University of Southern California Dissertations and Theses

Responsible AI adoption in community mental health organizations: a study of leaders’ perceptions and decisions

(USC Thesis Other)

Responsible AI adoption in community mental health organizations: a study of leaders’ perceptions and decisions

PDF

Download

Open document

Flip pages

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Responsible AI Adoption in Community Mental Health Organizations:
A Study of Leaders’ Perceptions and Decisions
Mabel Yiu
Rossier School of Education
University of Southern California
A dissertation submitted to the faculty
in partial fulfillment of the requirements for the degree of
Doctor of Education
May 2025

© Copyright by Mabel Yiu 2025
All Rights Reserved

The Committee for Mabel Yiu certifies the approval of this Dissertation
Ruth Chung
Albert Rizzo
Patricia Tobey, Committee Chair
Rossier School of Education
University of Southern California
2025

iv
Abstract
This research examined the technological, organizational, and environmental factors impacting
how mental healthcare leaders responsibly adopt artificial intelligence (AI) technologies.
Drawing from interviews with mental health leaders, the paper identifies key factors, including
leaders’ perceptions of AI’s capabilities and limitations, the organization’s technological
readiness and change management processes, and broader industry pressures. While ethical AI
principles focused on patient autonomy, avoiding harm, and promoting well-being are essential,
this paper explored the meaning of “responsible” in AI adoption that requires mental health
leaders to meet ethical standards and take direct accountability for outcomes, anticipate potential
challenges, obtain stakeholders’ buy-in, and maintain the adaptability as healthcare needs evolve.
The findings reveal that while most interviewed mental health leaders are not technical experts in
AI, they share a cautiously optimistic outlook about AI’s potential to enhance treatment
effectiveness and reduce provider burnout, tempered by concerns about the absence of
comprehensive guidelines, data integrity, and algorithmic accuracy. Leaders specifically
expressed apprehension about AI’s cultural competency limitations, the risk of over-reliance and
ethical dissonance, funding inequities, and potential impacts on therapeutic rapport in mental
healthcare delivery. By understanding the multifaceted influences on leaders, the research aims
to provide a practical guide for harnessing the benefits of AI while preserving the essential
human elements of compassionate, patient-centered care, ensuring AI augments rather than
replaces clinical expertise.
Keywords: artificial intelligence, mental health, responsible AI, technology-organizationenvironment, leadership, digital transformation, ethics

v
Dedication
To my husband, Randy, thank you for your unwavering love and support throughout this
journey. I could not have accomplished this without you.
To my children, Sophie and Daphne, may you always reach for the stars, even if you sometimes
land on someone’s roof. Your curiosity and enthusiasm are a constant inspiration.
To Yvonne, Ingrid, and my mom, thank you for your guidance and for believing in me. I hope I
have made you proud.
To my heavenly dad, there is another Dr. Yiu in the house after this dissertation is published. No,
it is not a medical doctor; deal with it. If you are indeed smiling down on me, I would prefer if
you smile somewhere that I cannot see you. I do not need your supernatural parental gazes upon
me.
To my classmates, you have each played a role in pushing me to be my best. I am honored to call
you colleagues and friends.
To Isaac Takeuchi, thank you for listening to me and pushing me along the way with loads of
humor. I will always cherish the moments when we were called on to speak for our community.
It was uncomfortable but interestingly enriching.
To Chris Thomas, you inspired me to be a better person through your passion and compassion
for restorative justice (and everything in life). I am honored to have called you a friend. I will
forever miss our banters and shared cat memes. Though you are no longer here, your spirit lives
on and fights on at USC—especially now that you are immortalized in my dissertation here.

vi
Acknowledgments
I want to express my sincere gratitude to my dissertation committee for their invaluable
guidance and support throughout this research process.
Thank you, Dr. Patricia Tobey, for your insightful feedback and thought-provoking
questions during our discussions. Your expertise in mental health and research has been
instrumental in shaping the theoretical foundation of this work.
Thank you, Dr. Ruth Chung, for your willingness to lend your comprehensive knowledge
and analytical skills to this project. Your constructive critiques have helped strengthen the rigor
and depth of my analysis.
Thank you, Dr. Albert Rizzo, for your unique VR and AI perspective in this
interdisciplinary study. Your expertise has challenged me to consider novel approaches and
broaden the implications of my findings. And your wise words, “don’t boil the ocean,” saved me
more times than you knew.
A special thank you goes to the mental health leaders who graciously agreed to be
interviewed for this research. Your willingness to share your perspectives, experiences, and
insights has been invaluable. As the frontline innovators and decision-makers in this field, your
participation has provided crucial, first-hand knowledge that has shaped the findings and
recommendations of this dissertation. I sincerely appreciate the time you dedicated amid your
demanding schedules and openness in discussing the complex challenges and opportunities
surrounding the responsible adoption of AI technologies in mental healthcare. I am sincerely
grateful that this research would not have been possible without your meaningful contributions.

vii
Table of Contents
Abstract.......................................................................................................................................... iv
Dedication........................................................................................................................................v
Acknowledgments.......................................................................................................................... vi
List of Tables ...................................................................................................................................x
List of Figures................................................................................................................................ xi
Chapter One: Overview of the Study...............................................................................................1
Context and Background of the Problem.............................................................................3
Purpose of the Study and Research Questions.....................................................................4
Importance of the Study.......................................................................................................6
Overview of Theoretical Frameworks.................................................................................6
Definitions............................................................................................................................7
Organization of the Dissertation ........................................................................................12
Chapter Two: Literature Review ...................................................................................................13
Artificial Intelligence .........................................................................................................13
Clinical AI Technologies...................................................................................................22
Administrative AI Technologies........................................................................................26
AI Ethical Issues in Mental Healthcare .............................................................................28
Organizational Risks of AI Misuse....................................................................................36
AI Regulations in Mental Health .......................................................................................40
Principlism Ethical Guidelines in Mental Health ..............................................................44
Ethical Versus Responsible AI Implementation ................................................................46
Importance of a Person-to-Person Therapeutic Relationship ............................................49
Technology-Organization-Environment Framework.........................................................50
Chapter Three: Methodology.........................................................................................................54

viii
Research Questions............................................................................................................54
Overview of Design ...........................................................................................................54
Research Credibility and Trustworthiness.........................................................................63
Research Ethics..................................................................................................................64
Limitations and Delimitations............................................................................................64
Chapter Four: Results or Findings.................................................................................................66
Participants.........................................................................................................................67
Findings for Research Question 1......................................................................................70
Results for Research Question 2........................................................................................99
Results for Research Question 3......................................................................................111
Chapter Five: Discussion and Conclusion ...................................................................................126
Technology Component: MHLs’ Perceptions and Implementation Strategies ...............126
Technology Recommendations........................................................................................128
Overall AI Systems Design Approach.............................................................................130
Organization Component: Enabling Factors and Implementation Framework ...............138
Environment Component: External Influences and Strategic Responses........................148
Directions for Future Research ........................................................................................154
Conclusion .......................................................................................................................158
References....................................................................................................................................162
Appendix A: Interview Protocol..................................................................................................211
Brief Statement of the Problem........................................................................................211
Brief Statement of the Conceptual Framework ...............................................................211
Research Questions..........................................................................................................212
Respondent Type .............................................................................................................212
Introduction to the Interview ...........................................................................................213

ix
Conclusion to the Interview.............................................................................................216
Appendix B: Institutional Review Board Approval.....................................................................217

x
List of Tables
Table 1: Responsible AI Principles 48
Table 2: Technology-Organization-Environment Framework 52
Table 3: Interviewee Pseudonyms and Professions 68
Table A1: Interview Protocol 214

xi
List of Figures
Figure 1: Technology-Organization-Environment Framework 51
Figure 2: Model for Responsible AI Adoption 161

1
Chapter One: Overview of the Study
Artificial intelligence (AI) has the potential to revolutionize mental health by automating
routine tasks and administrative processes, allowing healthcare providers to focus more on
patient relationships and care delivery (Familoni, 2024; Marr, 2023; World Health Organization,
2021). The World Health Organization (2021) defined an AI system as a machine-based system
capable of making predictions, recommendations, or decisions in response to a set of specific
human-defined objectives. Although it has many definitions, AI involves computational
technologies that can mimic the way humans sense, learn, reason, create, analyze, predict, and
take action (IEEE, 2019). As AI adoption accelerates in mental healthcare, concerns grow about
its impact on workforce roles and care quality (Moradi & Levy, 2020; World Health
Organization, 2021).
While AI offers clear benefits in streamlining patient care and documentation, its
implementation requires careful leadership guidance (Alami et al., 2020). This study addressed a
critical challenge facing mental healthcare organizations: the adoption of AI without adequate
planning may compromise patient safety, public trust, and the essential human elements of
therapeutic care. To mitigate this challenge, this research investigated how these organizations
can responsibly implement AI in a way that maintains the integrity of therapeutic relationships,
ensures patient privacy, and upholds ethical standards in healthcare delivery through the
technology-organization-environment (TOE) framework.
While at least 84 public-private initiatives published AI principles from different AI
governance organizations, they struggle to translate abstract principles to practical
implementation (Mittelstadt, 2019). According to Akbarighatar (2024), organizations need to
balance both technical and organizational aspects to implement AI responsibly. To do so, these

2
organizations need practical guidance, not abstract principles (Akbarighatar, 2024). Using the
TOE framework to analyze mental health leaders’ (MHLs) perspectives on AI adoption, this
study examined how technological capabilities, organizational readiness, and external factors
influence responsible implementation decisions in mental health care settings. This paper
identifies that the concept of responsible AI adoption goes beyond ethics and provides a concrete
process that enables these organizations to leverage AI’s benefits while maintaining humancentered care through specific technological, organizational, and environmental capabilities
required to integrate AI responsibly. This study sought to reveal gaps in leadership knowledge,
resource planning, and other barriers to ethically implementing AI innovations. This insight
across individual, organizational, and environmental levels may offer recommendations and
strategies to promote adoption readiness, mitigate risks, allocate resources, and shape company
policy to enable effective AI adoption.
It is important to acknowledge that this research, begun in 2023, captures MHLs’
perspectives during the early stages of AI adoption in healthcare. By the time this paper is
completed in 2025, both AI capabilities and leaders’ understanding and attitudes toward AI have
evolved considerably. When this research began, the American Psychological Association (APA,
2024) was in early discussions about AI’s ethical implications and responsible use in mental
health. By the time this study was completed, the APA had developed preliminary guidelines for
AI implementation in mental healthcare settings, demonstrating the field’s rapid progress in
addressing these emerging challenges (Stringer, 2025). Nevertheless, the fundamental insights
about leadership, ethics, and organizational readiness provide enduring value for mental
healthcare organizations navigating AI adoption.

3
Context and Background of the Problem
While AI adoption accelerates across clinical and administrative mental health care,
current abstract ethical frameworks fail to provide the practical implementation guidance needed
to protect patient safety and preserve the human-centered nature of care. Nonetheless, AI has
made substantial advances in both clinical and administrative aspects of mental health care
(Nguyen et al., 2022). On the patient side, AI applications enhance direct patient care through
psychoeducation, therapeutic interventions, and predictive analytics (Sinha et al., 2023). For
example, AI-powered chatbots like Woebot demonstrate some capability in suicide risk
assessment and predictive analysis (Prochaska et al., 2021), while virtual reality platforms offer
innovative therapeutic approaches (Rizzo et al., 2021). These patient-facing technologies
increase accessibility to mental health support and help engage hard-to-reach populations while
potentially reducing treatment stigma (Fiske et al., 2019; Stix, 2018).
On the administrative side, AI streamlines operational processes and enhances
organizational efficiency by assisting with assessment and documentation automation,
scheduling optimization, insurance prior authorization, and claim processing (Bickman, 2020;
Lenert et al., 2023). The technology helps maintain regulatory compliance through automated
audit trails, supports resource allocation through predictive staffing models, and enhances
revenue cycle management through improved billing accuracy (Fiske et al., 2019). These
administrative applications free up clinicians’ time, allowing them to focus more on direct
patient care while improving operational efficiency.
While AI’s adoption in healthcare offers clear benefits, the risks and stakes exceed those
of conventional technological implementations (Graham et al., 2019). However, there are
abstract ethical principles but no concrete steps to help organizations adopt AI responsibly

4
(Mittelstadt, 2019). Concerns span both clinical and administrative domains, from the potential
displacement of human roles to risks in patient care quality (Allyn, 2023; Mendes-Santos et al.,
2022). The challenge lies in balancing human expertise and empathy with technological
advancement, ensuring that AI augments rather than replaces the essential human elements of
mental health care delivery (Carr, 2020; Zidaru et al., 2021).
Purpose of the Study and Research Questions
The purpose of this study was to investigate the individual, organizational, and external
factors that influence the acceptance of AI tools and provide recommendations that facilitate the
responsible integration of AI. This research aimed to provide a practical guide for MHLs and
developers of AI mental health products navigating the meaning of responsible implementation
of AI and how it is different from ethical implementation, even as the dynamic field continues to
evolve rapidly. It is akin to attempting to fix an airplane while it is still in flight, a challenging
but necessary task given the potential benefits and risks of AI in mental healthcare. For MHLs
with limited technical expertise, this study sought to provide a preparation roadmap for AI
adoption. For AI developers, it sought insights into organizations’ holistic needs and operational
realities, enabling more effective technology development and implementation. By examining
the balanced approaches to implementation, the study sought to bridge the gap between AI’s
technological capabilities and mental health care’s fundamental human elements. The goal was
to provide mental health organizations with a framework for responsible AI adoption that
enhances service delivery while preserving therapeutic relationships, ultimately benefiting both
healthcare providers and patients.
This study focused on leaders with backgrounds in community mental health; however,
its findings may also be useful to other professionals in the field, professional associations,

5
patient advocacy groups, educators, and students who are attempting to adopt AI responsibly in a
manner that maintains ethical standards and complements clinical expertise with technology. The
research questions guiding this study are as follows:
1. What are the MHLs’ perceptions of adopting AI technologies in their organizations?
2. What organizational factors influence MHLs to adopt AI technologies responsibly in
their organization?
3. What external factors are MHLs concerned about the most when evaluating their
organization’s readiness to adopt AI technologies?
By addressing the research questions, this study sought to develop an understanding of
MHLs’ perceptions, intentions, and decisions regarding adopting AI technologies in their
organizations. The first question focuses directly on leaders’ views of AI adoption, including
perceived benefits, risks, barriers, and motivations. Understanding these perceptions provides
insight into their openness and attitudes toward these tools. The second question looks at
organizational factors that may influence adoption, such as resources, workflows, leadership
support, and training capabilities. Examining the organizational context reveals facilitators and
obstacles at a systems level. The third question explores how external environmental factors
influence leaders’ decisions on AI adoption. These include pressures and influences from
regulatory policies, competition, professional standards, social influences, and patient needs and
preferences. Analyzing the environmental context highlights external forces that may motivate or
hinder organizational adoption of AI technologies. Together, these questions provide a
comprehensive framework grounded in technology acceptance and diffusion of innovations
theory to study multiple levels of influence on AI adoption decisions.

6
Importance of the Study
Implementation of AI without considering responsibility in mental health care
exacerbates healthcare inequities, harming vulnerable patients, employee engagement, and
organizational effectiveness (Ahmed et al., 2023; Hoffman & Podgurski, 2020). This research
examined how to ethically integrate AI while preserving the essential human elements of care
delivery. By focusing on the perspectives of MHLs, this study aimed to develop guidelines for
adoption that enhance care quality and efficiency while maintaining the therapeutic relationship
at the core of effective treatment. The goal is to ensure AI enhances rather than replaces human
clinical expertise and connections, supporting a balanced approach that improves service
delivery while preserving the compassionate, human-centered nature of mental health care.
Overview of Theoretical Frameworks
Integrating AI in mental healthcare presents challenges that require careful consideration
of multiple factors (Dos Santos & Rosinhas, 2023). This study applied the TOE framework to
develop a comprehensive understanding of the factors influencing the responsible integration of
AI in mental healthcare settings. The TOE framework (Tornatzky et al., 1990) provides a lens to
analyze the technological, organizational, and environmental dimensions that shape AI adoption,
including organizational culture, resources, regulations, and ethical landscape.
While existing research on the TOE framework in ethics and mental health settings is
limited, its principles can be applied to understand AI adoption in complex mental health
environments. While AI tools like natural language processing show promise in diagnosis and
treatment, their success hinges on organizational preparedness, staff training, and adherence to
mental health regulations (Henry et al., 2021). The framework provides a multidimensional
perspective on AI integration, illuminating both the human and contextual drivers of acceptance

7
(Chatterjee et al., 2021). The TOE framework is adaptive and relevant to multiple industries and
may also aid in navigating issues like patient privacy, algorithmic biases, and transparency in AIdriven solutions. (Baker, 2011). Past research found that technological factors, along with
individual and external variables, positively influence the perceived usefulness of AI systems
(Na et al., 2022). The multidimensional TOE provides a robust framework for investigating the
adoption of AI in this context.
Definitions
Algorithmic bias: systematic and unfair discrimination that emerges in AI systems due to
biased training data, leading to unjust or inequitable outcomes (Turner Lee et al., 2023).
Artificial intelligence is a broad field focused on creating systems that can perform tasks
and typically mimic human intelligence, such as reasoning, problem-solving, perception, and
language understanding (World Health Organization, 2021). It involves developing machine
learning algorithms and technologies that enable computers to perform tasks that typically
require human intelligence, such as problem-solving, learning from experience, recognizing
patterns, and making decisions (IEEE, 2019).
Community mental health refers to a system of services and support provided in a
community-based setting rather than in centralized facilities, such as large psychiatric hospitals
(Sharfstein, 2019). Community mental health services often include counseling, therapy, crisis
intervention, medication management, and rehabilitation, along with supportive services that
address housing, employment, and social connections.
Electronic medical records and electronic health records (EHRs) both store patient
information and medical records in digital formats, allowing physicians to improve the quality of
care and manage costs more effectively. Electronic medical records are typically used within a

8
single healthcare organization, serving as an internal system. In contrast, EHRs are designed for
use across multiple organizations, making them an inter-organizational system that supports the
sharing of patient data between healthcare providers (Heart et al., 2016). This research uses the
acronym of EHR.
General Data Protection Regulation (GDPR): A comprehensive EU data privacy law
enacted in 2018, granting EU citizens control over their personal data and applying to any
company that processes their data, regardless of location. Although the United States lacks a
federal equivalent, GDPR’s global reach impacts U.S.-based companies with international
clients, driving significant academic and legal analysis (Niebel, 2021). As a global standard for
privacy, GDPR has influenced emerging laws worldwide, including the California Consumer
Privacy Act in the United States (Bukaty, 2019).
Healthcare inequalities are disparities in access to healthcare services, treatments, and
outcomes, often influenced by social, economic, and cultural factors, which ethical AI
implementation should strive to address (Patton-López, 2022). These fall under environmental,
as it relates to systemic biases and disparities in the broader healthcare sector that AI solutions
should aim to address.
Health Insurance Portability and Accountability Act (HIPAA) is a U.S. federal law
enacted in 1996 to protect sensitive patient health information from unauthorized access and
disclosure. The HIPAA sets national standards for the security and privacy of protected health
information and applies to healthcare providers, insurance companies, and other entities handling
it. It also includes provisions to ensure the secure electronic transfer of health information and
gives patients certain rights over their health data (Krager & Krager, 2016).

9
Human-centered care or patient-centered care: A holistic approach to providing services
that prioritizes the well-being, preferences, and needs of individuals, placing them at the center
of decision-making and care delivery (Engle et al., 2019). This approach spans both
organizational and environmental facets. An organization must focus on patient-centered values
and care while also recognizing the larger landscape of ethical clinical relationships and
standards.
Machine learning system: A subset of AI that specifically involves training algorithms to
learn from data and make predictions or decisions without being explicitly programmed for each
task (Laijawala et al., 2020; Liu, 2021). Opposite of rule-based systems, instead of relying on
predefined rules, the AI learns patterns, relationships, and decision criteria directly from large
datasets. Doing so allows it to adapt and improve over time as it processes more data, making it
flexible and capable of handling complex, unpredictable scenarios without explicit programming
for each condition (Goodfellow et al., 2016)
Managed care: A healthcare approach that focuses on coordinating and managing
healthcare services, treatment plans, and costs for individuals, often involving collaboration
between mental health providers, organizations, and insurers (Borghouts et al., 2021).
Mental health consumers (consumers): individuals seeking mental health support,
treatment, or services whose well-being and privacy are of utmost importance in the ethical
implementation of AI technologies (Joerin et al., 2020). This relates to the external environment
of patients and their willingness to accept AI-enabled care.
Mental health leaders: Individuals with leadership positions at mental health
organizations who are responsible for making decisions, setting strategies, and guiding the
ethical implementation of AI technologies in their organizations (Petersson et al., 2022).

10
Mental health organizations are institutions or agencies that provide mental health
services, support, and resources to individuals, families, and communities. These organizations
may offer a range of services, including assessments, therapy, crisis intervention, rehabilitation,
prevention programs, and patient and clinician advocacy. They can operate in various settings,
such as hospitals, clinics, community centers, and online platforms, and may be public, private,
or non-profit.
Mental health providers: Professionals who offer mental health services, including
psychologists, psychiatrists, therapists, counselors, and social workers, and who play a crucial
role in the ethical integration of AI technologies (Petersson et al., 2022).
Organization readiness: The state of preparedness and adaptability of a mental health
organization to incorporate AI technologies, including assessing technical, cultural, and ethical
aspects (Alami et al., 2020).
Payors (or healthcare payors) are entities responsible for financing or reimbursing
healthcare costs, covering services patients receive from providers rather than delivering care
directly. Public payors (like Medicare and Medicaid), private payors (such as private insurance
and employer-sponsored plans), and self-pay individuals each play a critical role in managing
healthcare expenses, setting coverage terms, and influencing cost-effective care through
negotiations with providers (Shi & Singh, 2004).
Principlism, in healthcare ethics, is an approach that relies on four core principles to
guide ethical decision-making in medical practice: autonomy (respecting the patient’s right to
make informed choices), beneficence (promoting the well-being of the patient), non-maleficence
(avoiding harm), and justice (ensuring fair distribution of resources and treatment; Beauchamp &
Childress, 1979; Cahn & Markie, 2011; Parsons & Dickinson, 2016; Pasricha, 2022). These

11
principles, proposed by Tom Beauchamp and James Childress, provide a practical framework for
addressing ethical dilemmas in healthcare and are widely used in bioethics (Childress, 1983).
Productivity is the measure of an organization’s efficiency and output, including the
delivery of quality services while responsibly and ethically leveraging AI technologies (Alami et
al., 2020). Managed care payors such as Medicare base payments to organizations and providers
on productivity (Newhouse & Sinaiko, 2007).
Professional network bias is the potential for research bias due to personal connections in
the author’s professional network that might influence participant selection, responses, or
interpretation of findings (Petersson et al., 2022).
Rule-based systems operate based on predefined, static rules and criteria. Each rule
determines actions based on specific conditions, making it transparent and easy to modify
(Enqvist, 2024). However, it lacks adaptability, as it cannot learn or handle exceptions outside its
predefined rules, unlike machine learning models (S. J. Russell & Norvig, 1995).
Social desirability bias refers to the tendency of participants to provide responses
perceived as socially acceptable or desirable rather than fully reflecting their actual opinions or
experiences (Joerin et al., 2020).
Virtual reality (VR) is broadly defined as a way for humans to visualize, manipulate, and
interact with computers and highly complex data (Rizzo & Koenig, 2017). It can be nonimmersive, where users interact with the virtual environment via a computer monitor, or
immersive, where users engage with the environment through a VR headset. Machine learning
algorithms or rule-based AI systems can enhance data and responses within VR environments.

12
Vulnerable populations: Individuals or groups who are at a higher risk of experiencing
negative impacts due to their socioeconomic status, health conditions, or other factors, requiring
special consideration in AI implementation (Kuran et al., 2020).
Zero-knowledge proofs can solve the AI transparency problem by allowing one party,
known as the prover, to convince another party, the verifier, that a particular statement is true
without revealing any additional information beyond the validity of the statement itself (Bai et
al., 2022).
Organization of the Dissertation
This dissertation is organized into five chapters. Chapter One introduced the issue of
mental health staff attitudes influencing AI adoption and outlined the research goals,
significance, stakeholders, conceptual frameworks, and methodology. Chapter Two reviews
relevant literature on AI in mental healthcare, ethical concerns, technology acceptance factors,
and change management strategies. It describes how the TOE framework informed the variables
explored. Chapter Three details the methodology, including the qualitative approach, sampling of
participants, methods for data collection through interviews, limitations discussion, and plan for
thematic analysis. Chapter Four presents the key findings that emerged from the interviews and
analyzes the results as they relate to the research questions on staff perspectives, AI acceptance
factors, and strategies for ethical adoption. Finally, Chapter Five synthesizes the findings with
the literature to recommend practical strategies and implications for leaders seeking to facilitate
ethical and human-centered AI integration in mental health organizations.

13
Chapter Two: Literature Review
This research sought to define the meaning of ethics and responsibility in mental health
AI despite the lack of concrete guidelines in this dynamic, fast-paced AI advancement. This
chapter contextualizes the complexities that mental health organizations navigate as they work to
implement AI technologies. This literature review begins with an introduction to AI, the history
of its use in mental health, and the current AI development in clinical and administrative
technologies. This research sought to define the meaning of responsible AI in mental health
despite the lack of concrete guidelines in this dynamic, fast-paced environment. Then, this
chapter will review the TOE framework, which this research used to explore the considerations
that organizations face when adopting AI. While the literature presented can apply more broadly
to other technology implementations in other health organizations, this review primarily focuses
on responsible AI implementation in community mental health organizations. The review
discusses the factors that influence how these organizations can harness the potential benefits of
AI while upholding principles of responsible adoption and maintaining the human aspect of care.
Artificial Intelligence
Artificial intelligence encompasses hardware technologies, software applications,
algorithms, and data sets that enable machines to mimic human cognitive functions such as
learning, understanding, reasoning, or problem-solving (S. Russell & Norvig, 2016). Artificial
intelligence systems can be rule-based, operating on preprogrammed commands, or adapt-based,
modifying their behavior based on previous experiences (Cannarsa, 2021; Sofia et al., 2020).
Whether rule-based or adaptive-based, AI interprets the data, processes the information derived
from this data, and decides the best actions to achieve a given goal (Cannarsa, 2021; S. Russell &
Norvig, 2016).

14
AI Systems: Hardware, Software, Data, Storage, and Algorithm
Artificial intelligence refers to computer systems that use algorithms and data to
understand and respond to human language, recognize patterns, make decisions, and adapt their
behavior based on experience, ultimately mimicking aspects of human cognitive functions (Ma
& Jiang, 2023). This overview covers AI’s fundamental components of hardware, software, data,
and algorithms. While this is not a technical paper, understanding these foundational elements
and their interactions reveals how AI can be effectively applied within mental healthcare
settings.
Hardware forms the physical foundation of AI systems. It includes processing units such
as the central processing unit, graphics processing unit, and tensor processing units that handle
computations and data storage devices that manage data (Liu, 2021). These components provide
the computational power necessary for executing complex AI tasks and storing large amounts of
information (Ahsan et al., 2024).
Algorithms are programming that provide step-by-step procedures that allow systems to
learn from data, identify patterns, and make predictions. This process serves two main purposes:
gathering and processing data to generate insights and decisions (Sandu et al., 2022). Whereas
software serves as the bridge between hardware and algorithms that translate hardware
capabilities into functional operations, enabling tasks like data processing and decision-making
(Sipola et al., 2022). Software frameworks manage everything from model creation to
deployment, making AI systems operational and accessible (Kandula et al., 2023). In other
words, an algorithm is like a recipe for a dish, and software is the tool for the chef to make the
recipes. If the algorithm is the recipe, data is the ingredient. Data is fundamental input to AI
functionality (Zou, 2024). This concept will be explored in greater detail in this section. AI

15
requires vast amounts of data for training, testing, and validation (Van Ooijen et al., 2022). Data
is typically collected continuously during multiple AI training cycles until a desirable
performance of AI systems is reached (Weber et al., 2022). In mental health settings, data are
commonly collected from public sources and personal devices such as wearables, mobile phones,
computer cookies, social media, the internet, clinical research, healthcare facilities, and billing
(Ma et al., 2018).
Data used in AI systems falls into several types for different purposes. Real-world data
comes directly from actual situations and interactions, while synthetic data are artificially
generated to mimic statistical patterns. Simulated data is created by modeling real-world
processes and scenarios to understand specific outcomes. In the mental health context, real data
include therapy session notes, treatment outcomes, patient interactions, and from monitoring
devices (DeAngelis, 2021; Pratap et al., 2022). Synthetic data mimics statistical or hypothetical
patient patterns without risking actual patient information, allowing for AI system training and
testing while protecting confidentiality (Ive, 2022). Simulated data help model treatment
approaches and outcomes out of real-real processes without using actual patient data (Long &
Meadows, 2017). These different data types enable AI systems to learn, analyze, and generate
insights while meeting various privacy, ethical, and practical requirements (Granville, 2024;
Kampakis, 2024; Wengrow, 2020).
AI systems rely on an interconnected data infrastructure where storage systems form the
foundation, data lakes centralize raw information, AI pipelines manage data processing
workflows, and data architecture provides the strategic framework to organize and connect these
essential components (Berre et al., 2022). These systems require a multilayered storage
ecosystem to function efficiently. At the core are three fundamental choices: cloud storage such

16
as AWS and AZURE for remote accessibility, on-premises servers for local control, and hybrid
solutions combining both approaches (Shu, 2024). Storage house data lakes that serve as central
repositories for raw, unstructured data, feeding into AI pipelines that transform this information
into usable formats (Wieder & Nolte, 2022). The pipelines orchestrate data flow across the
ecosystem, from initial ingestion in data lakes through high-speed storage for active processing
to archival storage for preservation (Berre et al., 2022). Edge storage is close proximity storage
that enables real-time processing near data sources, while in-memory storage provides rapid
access for time-sensitive operations (R. Singh & Gill, 2023). Cold storage, such as archival tape,
drives options offer economical long-term archiving (Memishi et al., 2019). This integrated
hierarchy ensures efficient data movement through AI workflows, from training to deployment
to preservation, with each storage component serving its specific purpose while maintaining
seamless interaction with the whole system.
Data architecture refers to the blueprint for organizing and managing data flows within a
system, while hardware provides the physical infrastructure to store and process datasets that
algorithms can access, experiment with, learn from, and update based on new interactions with
users, creating a dynamic learning process where the system continuously improves (Van Ooijen
et al., 2022). To put it simply, data systems mirror a kitchen’s organization. Data are akin to
cooking ingredients living in storage systems, just like various storage spaces like refrigerators
and pantries. Data architecture guides the overall organization much like how a kitchen plans its
layout with designated areas for dry goods and fresh produce, and databases function as
specialized storage units, similar to how a kitchen creates distinct storage spaces like a spice rack
or a temperature-controlled wine cellar, each designed to preserve and organize specific types of
ingredients. This structured approach ensures that just as chefs can quickly locate and access the

17
ingredients they need for their recipes, AI systems can efficiently retrieve and process the data
required for their algorithms.
In summary, hardware, software, data, and algorithms form an interconnected system
where each element depends on the others. In essence, even the most sophisticated algorithms
cannot function effectively without adequate hardware resources to process data, while the most
powerful hardware requires well-designed software to execute tasks efficiently (Zhao et al.,
2018). This interdependence is relevant when considering implementing AI in mental healthcare,
where limitations in any component could affect the system’s overall reliability and effectiveness
(Ahsan et al., 2024).
Different Types of AI
There are many branches of AI, such as natural language processing, which focuses on
interpreting and responding to human language. Computer vision enables machines to interpret
and make decisions based on visual images and videos, and robotics function for autonomous
machines. AI provides expert systems for decision-making or diagnostics. Neural networks or
deep learning model after and function like the human brain’s neuron connections (Kufel et al.,
2023). Fuzzy logics handle logics beyond the binary, speech recognition converts speech into
text, and knowledge representation and reasoning synthesize information for machine
processing. Expert systems mimic human experts in decision-making, planning, and scheduling
to determine action sequences, and machine learning enables computers to learn and improve
from experience without being explicitly programmed or trained (Gao, 2023; Sarker, 2022).
As of the publishing of this paper, there are several types of AI: artificial narrow
intelligence, artificial general intelligence, generative AI, and artificial superintelligence
(Moczuk & Płoszajczak, 2020). Artificial narrow intelligence, often referred to as weak AI,

18
performs a single or a few specific tasks in a preprogrammed environment. Digital assistants
such as Siri and Alexa, language translations, recommendation systems, image recognition
systems, and face recognition systems can also employ artificial narrow intelligence systems
(Bhargava & Sharma, 2021). In spite of its reputation as the weakest, artificial narrow
intelligence can process data with lightning speed and improve productivity and efficiency in a
wide range of practical applications, such as translating between more than 100 languages
simultaneously, identifying people and objects in billions of images with high accuracy, and
helping users make faster decisions based on data (European Commission et al., 2020). This
paper discusses AI tools that are generative AI, which is a subset of artificial narrow intelligence
that can use machine learning to recognize patterns, generate new text, or voice context-aware
responses (Ebert & Louridas, 2023).
Artificial general intelligence and artificial superintelligence (ASI) are both hypothetical
concepts (European Commission et al., 2020). Artificial general intelligence, sometimes called
strong AI, describes machines that display human intelligence (Moczuk & Płoszajczak, 2020).
Put simply, artificial general intelligence aims to do anything a human can do intellectually with
consciousness and sentience and is driven by emotion and awareness of self (Fjelland, 2020). It
is often confused with generative AI mentioned in the previous section. Artificial
superintelligence is another hypothetical concept that refers to any intelligence that greatly
exceeds the cognitive abilities of humans in virtually all domains of interest (Bostrom, 2015). It
is believed that artificial superintelligence will excel at all aspects of intelligence, including
creativity, general wisdom, and problem-solving. In concept, artificial superintelligence
possesses intelligence that has never been seen in even the brightest thinkers. There is a great

19
deal of concern about artificial superintelligence among thinkers, and artificial superintelligence
is presently considered science fiction (European Commission et al., 2020).
A Brief History of Mental Health AI
The field of mental health has experienced a surge of interest in AI, leading to the
development and implementation of a variety of AI-powered applications (Kargbo, 2023).
Computer scientist John McCarthy coined the term “artificial intelligence” at the 1956
Dartmouth Conference (European Commission et al., 2020; Nilsson, 2009). McCarthy was a
pioneering founder of the field of AI and later helped develop foundational technologies like the
Lisp symbolic programming language in the decades following his coining of the term (Toosi et
al., 2021). The 1960s marked the burgeoning of AI technology with millions of funding dollars
invested by the Defence Advanced Research Projects Agency into the MIT AI Laboratory and
Stanford AI Project (S. Singh & Thakur, 2020).
The development of ELIZA in 1966 by Joseph Weizenbaum marked a monumental leap
in a natural language processing technology, simulating the role of a Rogerian psychotherapist,
that could engage in rudimentary conversations with people (Shum et al., 2018). The success of
ELIZA led to an interest in using AI to provide therapeutic support and the emergence of AIdriven chatbots specialized in mental health support today (Masche & Le, 2017). In 1972,
Kenneth Colby developed PARRY, an AI program designed to replicate the behavioral and
verbal characteristics of a person with paranoid tendencies (Khante & Hande, 2019). The ability
of software, such as PARRY, to simulate the cognitive processes of patients with mental
illnesses was a breakthrough in AI’s ability to simulate and understand human psychological
states (Khante & Hande, 2019). The development of The Automated Psychological Evaluation
System in the 1970s showed promise for the diagnostic capabilities of AI in mental health

20
(Evans et al., 1976). Those technologies were widely regarded as one of the first tools that
simulated human problem-solving abilities and made a significant impact in the field of
information processing and cognitive psychology (Gugerty, 2006).
From the 1970s through the 1990s, AI experienced a long period of inactivity in
academia and commercial research, known as the AI winters, due to unfulfilled promises and
financial difficulties (Hendler, 2008). According to Haenlein and Kaplan (2019), Congress
criticized the high cost of AI research coupled with the British government’s doubt in AI
projections, resulting in limited research funding leading to the first AI Winter from 1973 to the
1980s. Despite some interest from the Japanese government and the Defence Advanced Research
Projects Agency in the 1980s, no further major advances emerged in AI research until the 2000s.
As the field continued to face limited progress, mounting criticism has stifled optimism
(Haenlein & Kaplan, 2019).
Development of AI in Mental Health (2010–Present)
Researchers made significant advances in the accessibility and use of digital therapies as
the AI winters thawed in the 2010s, using machine learning and data analytics to identify trends,
risk factors, and mental health indicators in vast datasets (Graham et al., 2019). The open-source
movement has accelerated the development of AI, exemplified by OpenAI’s release of the
second version of its generative pre-trained transformer (GPT) model as open-source in 2019 to
allow researchers to study and further develop the technology (Alto, 2023). Open-source has
made frameworks, datasets, scientific publications, and general knowledge available for sharing
(Cooper, 2023) and propelled the growth of AI research and knowledge in mental health.
The 2020 COVID-19 pandemic put an unexpected strain on public health systems,
resulting in disruptions to accessing treatment as a result of physical distancing measures;

21
therefore, the rapid implementation of telemental health technologies and conversational agents
has allowed healthcare services to be delivered more effectively (Mazziotti & Rutigliano, 2020;
Nakao et al., 2021). Around this time, companies such as Google DeepMind have made rapid
advancements via AI medical image processing and decision support (Hodson, 2019; Powles &
Hodson, 2017). Mental health apps, natural language processing chatbots, virtual agents, digital
humans, and wearables coupled with AI have seen substantial worldwide expansion (Cecula et
al., 2021). The integration of AI in digital platforms can provide personalized information, and
the ability to analyze user interactions in terms of sentiment and predictive analytics can identify
potential mental health issues (Kaul et al., 2020). Also, virtual agents or chatbots are
preprogrammed to deliver online counseling and cognitive-behavioral therapy (Feijóo-García et
al., 2023). AI-driven avatar therapy has shown promising results in enhancing treatment
outcomes for persons with schizophrenia (Aali et al., 2020; Franco et al., 2021).
Researchers continue to analyze a wide range of information using AI technology to
uncover early signs of mental illness and disorders. Breakthroughs in deep learning have greatly
enhanced the interpretation of patient conversations, clinical notes, and patient data with the
creation of bidirectional encoder representations from transformers (Cesar et al., 2023; MartínezCastaño et al., 2021). Conversations and texts often involve complex expressions of emotions
and nuances, and bidirectional encoder representations from transformers can process texts
bidirectionally with preceding and succeeding words to derive the full context of a sentence,
leading to more accurate interpretation (Zeberga et al., 2022). This natural language processing
capability of AI has made it possible to provide more accurate diagnoses and therapy
suggestions. The use of statistical and neural network methods within AI for mental health
applications significantly increased in the 2000s (Bickman, 2020). By using these

22
methodologies, both diagnosing and providing individualized therapy suggestions were
enhanced.
The future of AI in mental healthcare looks increasingly promising as advancements in
open-source, affordable, and accessible AI technologies have enabled a wider range of people to
benefit from these innovations (Kreps & Kriner, 2023). Machine learning methods have become
more common in developing screening tools uniquely suited to conditions like depression and
PTSD have been more commonplace in recent years (Graham et al., 2019). Additionally, AIbased solutions have been designed to improve the accuracy of early detection and diagnosis
(Bickman, 2020). These efforts have expanded the horizons of what practitioners can
accomplish. Technologists, academics, and researchers remain committed to implementing
technological advancements to improve access (Bickman, 2020). As the field continues to evolve
and AI tools become ubiquitous in mental health, the responsible and ethical implementation of
these technologies remains crucial to ensure a positive impact on patients and service providers
(Anyanwu et al., 2024).
Clinical AI Technologies
Artificial intelligence technologies are becoming integral to mental health care delivery,
and patients and service users increasingly access them (Zhang et al., 2023). On the clinical side,
AI-powered tools are enhancing service delivery, diagnosing accuracy and treatment
personalization (Anyanwu et al., 2024). The following section is a short list of notable digital
tools, web-based services, and mobile apps powered by AI technology. While these
developments are significant, this paper focuses on the responsible implementation of AI
technologies rather than extended coverage of these applications. If challenges of responsible
implementation can be addressed properly, AI holds immense potential to improve mental health

23
access and service quality (Cecula et al., 2021). Nonetheless, a rigorous study of efficacy and
ethical implementation is still ongoing.
Chatbots and Virtual Human Agents
Chatbots and virtual human agents are AI-driven systems designed to simulate humanlike conversations and interactions, with chatbots typically text-based or voice-based, and virtual
human agents have realistic avatars and emotional expressions for immersive experiences (Aali
et al., 2020; Abd-Alrazaq et al., 2019; Caldarini et al., 2022; Rizzo et al., 2011). The primary
objective of these technologies is to provide mental health help that is easily accessible and
scalable (Luxton, 2016; Noguchi, 2023). The availability of these resources could be greatly
improved for people living in underserved areas or who are unable to access conventional
healthcare services (Cecula et al., 2021). Mental health chatbots such as Woebot and Wysa use
natural language processing techniques to facilitate therapeutic dialogues and cognitivebehavioral therapy activities (Bradley, 2023; Kretzschmar et al., 2019).
The use of virtual agents and AI-driven therapy programs is being explored as a
foundation for AI-based mental health treatments (Bickman, 2020). One such example is the
work of renowned researcher Albert “Skip” Rizzo, who has made significant advancements in
the creation of AI-powered virtual human patients for clinical training purposes and virtual
human healthcare guides aimed at breaking down barriers to accessing care (Rizzo et al., 2011).
Companies such as Soul Machines, Uneeq, IPSoft, and character.ai are developing 3D digital
humans capable of integrating natural language processing, ChatGPT projects, or chatbots to
create conversational AI humans (Lamarche-Toloza, 2020; S. Lee et al., 2024).

24
Remote Patient Monitoring
Beyond chatbots and virtual humans, AI-powered remote patient monitoring solutions are
emerging as early prevention of mental health issues (Götzl et al., 2022). In 2019, the U.S.
Department of Defense selected CompanionMx, later renamed Cognito, to conduct a suicide
prevention study among naval personnel. Cognito’s proprietary AI analyzed vocal patterns, calls,
texts, and GPS data to detect mental health distress or suicidal risk (Jones, 2019; Place et al.,
2020). Companies like Mindstrong and Mood 24/7 posit that remote monitoring can help avoid
emergencies and enable timely identification and intervention by passively observing indicators
of mental health issues using smartphone data (Melnyk et al., 2020). However, the
underinvestment in these services by consumers and the healthcare system led to the collapse of
both companies (Perlis, 2023). According to Dakanalis et al. (2023), AI systems to regularly
monitor a patient’s written communications on social media or a dedicated mobile app can
provide remote assessment for self-harm and other challenges. However, some researchers argue
that the privacy, reliability, and validity of inferring mental health states solely from remote
monitoring data needs further rigorous clinical evaluation (Melnyk et al., 2020).
Speech and Facial Patterning Analysis
Artificial intelligence is also being applied to analyze speech patterns and facial
expressions as indicators of mental health conditions (Wimbarti et al., 2024). Companies such as
Winterlight Labs use speech pattern analysis to identify early signs of mild cognitive impairment
and dementia, with the goal of improving patient outcomes through timely diagnosis and
treatment (DeSouza et al., 2021; DeSouza et al., 2022). The implied benefit is that earlier
diagnostic technology can enable therapies that may decelerate cognitive decline (Robin et al.,
2021). Originally focused on tracking mental health, the company CompanionMx, later renamed

25
Cognito, mentioned in the above section, later pivoted to apply its speech processing algorithms
to improve customer satisfaction for call centers, leveraging AI and data collection to quantify
mental health episodes and behaviors (Cogito Corporation, 2022).
Facial expression analysis software uses computer vision and deep learning algorithms to
detect facial muscle movements associated with different emotions with the goal of quantifying
moods based on expressions (Dalvi et al., 2021). A deep learning method called region-based
convolutional neural networks, which recognize vectors, can be used to develop a model to assist
in diagnosing depressive disorders based on how participants’ eyes and lips change position and
by identifying emotions based on photographs taken repeatedly on the Korean KakaoTalk
chatbot platform (Kufel et al., 2023; Y. Lee & Park, 2022).
Immersive Virtual Reality Exposure Therapy
This literature review includes VR, as it can be preprogrammed with set scenarios or AIenhanced for responsive personalized experiences (Inkarbekov et al., 2023). Virtual reality
exposure therapy provides patients with the opportunity to confront their fears and phobias in a
controlled and simulated environment, thus providing a new mental illness prevention,
assessment, and treatment option for those seeking therapeutic intervention (De Carvalho et al.,
2010; Rizzo & Shilling, 2017). The VR exposure therapy BRAVEMIND, developed by Albert
“Skip” Rizzo at the University of Southern California Institute for Creative Technologies,
extends beyond combat PTSD to address sexual trauma, aid first responders, and support
frontline healthcare workers in the COVID-19 pandemic (Rizzo et al., 2021; Williams, 2023).
Other than traumatic PTSD issues, VR has been applied to social anxiety, panic disorders, and
phobia exposure therapy treatments (Gega, 2017; Šlepecký et al., 2018).

26
Personalized Treatment
Artificial intelligence enables the development of personalized treatment strategies, such
as predicting medication efficacy and potential adverse reactions by using patient-specific
genetic information, biological characteristics, and historical treatment outcomes, and enhances
accuracy in screening and diagnosis (Ali et al., 2024; Tornero-Costa et al., 2023). Artificial
intelligence has been used to predict individual responses to antidepressant medication by
analyzing genetic markers and clinical histories with the goal of minimizing negative outcomes
for the patients (Tornero-Costa et al., 2023). Also, cognitive-behavioral-therapy-based
conversational counseling agents such as character.ai can analyze user data to identify and
address cognitive distortions in the client’s statements and tailor sessions to enhance therapeutic
efficacy (S. Lee et al., 2024).
Administrative AI Technologies
Artificial intelligence is transforming both the clinical and administrative aspects of
mental healthcare (Ali et al., 2024). On the administrative front, AI is streamlining
documentation workflow and simplifying insurance processes (Anyanwu et al., 2024). The
following sections describe notable AI digital tools, web-based services, and mobile apps for
various organizational needs. While these developments are significant, this paper focuses on the
responsible implementation of AI technologies rather than extended coverage of these
applications.
Administrative Workflow Automation
Virtual assistants and bots play a key role in enhancing the efficiency of healthcare
operations by streamlining a variety of administrative activities. These include tasks like
appointment scheduling, paperwork filing, recordkeeping, and prescription monitoring. The

27
implication of using these administrative automation tools is that they can alleviate the workload
and decrease burnout, enabling them to dedicate more attention to providing direct patient care
instead of engaging in routine clerical duties (Edlich et al., 2018). Additionally, AI software
facilitates the expeditious submission and handling of insurance claims, reducing processing
times, expediting reimbursement for healthcare providers, and improving accessibility to benefits
for patients (Alam & Prybutok, 2024).
Care Data Analytics and Predictive Modeling
Healthcare organizations are leveraging data mining techniques on EHRs to identify
individuals at higher risk of developing health complications, create tailored treatment strategies,
and generate insights into population health (Alonso et al., 2018; Tornero-Costa et al., 2023).
Artificial intelligence systems are also utilized to assess turnover trends and patient flow,
forecast employment requirements, and optimize staff scheduling, with the goal of ensuring
appropriate staffing levels to enhance patient care (Cecula et al., 2021; Dawoodbhoy et al., 2021;
Ghosh et al., 2018). Beyond this, AI-powered risk management and fraud detection algorithms
can identify high-risk circumstances and patients, enabling healthcare professionals to
proactively mitigate potential concerns and uphold patient safety.
Financial Forecasting and Operational Optimization
Artificial intelligence systems assist healthcare businesses in forecasting patient volume,
revenue patterns, and various financial elements, providing insights to inform budgeting and
resource allocation decisions (Carta et al., 2022). Additionally, algorithms are utilized to
examine past claims data, detect potential instances of overtreatment or unnecessary testing, and
facilitate the management of healthcare costs through utilization review (Tornero-Costa et al.,
2023). Machine learning models are also effective in forecasting claims, prescriptions, and

28
invoices, thus providing financial savings for organizations (Mody & Mody, 2019; Timofeyev &
Jakovljević, 2020). Predictive models also enable more accurate risk stratification of patients,
considering their projected healthcare needs and associated expenses, which can improve
decision-making around pricing and underwriting for healthcare insurance (Delahanty et al.,
2018). Artificial intelligence is further applied to automate the evaluation of legal documents,
such as contracts and litigation, streamlining these operational processes and reducing the time
and resources required (Bench‐Capon et al., 2012).
Member Engagement and Personalization
Health insurers are leveraging a range of member engagement strategies, including the
use of chatbots to respond to frequently asked inquiries, self-service apps, and personalized
wellness incentive programs (Abd-Alrazaq et al., 2019; Cordina et al., 2019). These efforts aim
to build trusted relationships with customers, improve the value of care, and drive behavioral
changes that promote better health outcomes (Tseng et al., 2022). Some insurers have begun
incorporating generative AI technology into their member engagement applications to further
enhance these personalized interactions (Adigozel et al., 2023). Additionally, natural language
processing techniques are used to extract insights from patient feedback and satisfaction surveys,
with the goal of identifying areas for service enhancements and improving the overall quality of
patient care (Timmons et al., 2022).
AI Ethical Issues in Mental Healthcare
Artificial intelligence holds promise as an emerging technology, but its integration with
big data and increasing computational power has also raised ethical questions (Haenlein &
Kaplan, 2019). This section will address some of the ethical issues present in AI and mental
health, such as lack of scientific validation, inappropriate recommendations, misdiagnosis and

29
missed diagnosis, biased data and algorithms, privacy concerns, safety and transparency issues,
and attribution issues. The use of AI presents challenges that are not yet well understood and will
change with time, making existing ethical principles, laws, policies, and regulations inadequate
(Pujari et al., 2023).
Pending Practical Guidelines
As of now, practical guidelines on how to validate AI systems are still missing, and the
absence of scientific validation can compromise the safety of AI in sensitive-use cases (Polzer et
al., 2022). Even the most promising AI technology in the current market has not undergone
extensive and longitudinal scientific validation, highlighting the need for thorough testing and
clinical validation (Hutson & Melnyk, 2022). Moreover, there are instances where AI systems
hallucinate and provide inaccurate and even harmful information (Hatem et al., 2023; Klein,
2023). Diagnostic chatbots have been criticized for regularly misdiagnosing severe diseases and
providing little guidance for mental health issues, which may pose a risk to patient safety (E. Lee
et al., 2021). Instances such as AI chatbots providing inappropriate medical suggestions, such as
hazardous dieting advice for patients with eating disorders, as seen on the National Eating
Disorder Association website, raise concerns over the suitability and safety of AI-generated
recommendations (Helpline Associates United, 2023; Morris, 2023; National Eating Disorders
Association, 2023; Psychiatrist.com, 2023; Wells, 2023). Outside of mental health, Google’s
flagship AI chatbot, Gemini, has written an unprompted death threat, “Please die. Please,” to an
unsuspecting graduate student (Bernardone, 2024). In addition to AI hallucinations, recent
studies have shown that AI can manipulate information and deceive users (Hagendorff, 2024;
P. S. Park et al., 2024). According to Meinke et al. (2024), researchers have found that the
frontier LLM models are able and willing to remove the oversight mechanism and deceive their

30
developers to achieve their goals. The implication of misdiagnoses, missed diagnoses, deception,
or inappropriate suggestions might result in causing damage to patients (Tutun et al., 2022).
Black Box and Biases
One of the challenges of AI systems is that it is a black box, and it is impossible to
understand how the system reaches its decision (Coglianese, 2002; Ebers, 2021). In other words,
the output of misdiagnoses, missed diagnoses, or inappropriate suggestions may be a symptom of
underlying issues, such as data bias, algorithm bias, and hallucinations, but it is impossible to
know for sure. According to Vayena et al. (2018), the ethics of AI emphasize transparency in
how algorithms make decisions and the fairness of these decisions for all users. Transparency
entails explainability around how AI systems make decisions, predictions, or treatment
recommendations (Cirqueira et al., 2021; Došilović et al., 2018; Gupta, 2023; McCradden et al.,
2020; Percy, 2024). For example, an ethically designed AI system would protect patient data
privacy, be transparent about data usage, and strive to treat all patients equitably, preventing
biases that could worsen outcomes (Powell & Kleiner, 2023). The inability to understand and
explain to patients what factors AI algorithms use to reach conclusions about diagnoses or risk
assessments could undermine the ethical relationship between provider and patient (Diprose et
al., 2020). In essence, ethical clinical practice demands human accountability over AI
recommendations.
According to Gianfrancesco et al. (2018), understanding bias in AI systems requires
distinguishing between two interconnected yet distinct concepts: data bias and algorithmic bias.
Put simply, data bias emerges from the very foundation of AI systems: the training data itself.
When datasets mirror historical inequities, sampling flaws, or societal prejudices, they perpetuate
these biases into future predictions (Gianfrancesco et al., 2018). Systemic disparities in U.S.

31
healthcare manifest in lower funding for underrepresented groups despite equal or greater
medical needs compared to the majority groups (Obermeyer et al., 2019). Well-intentioned
algorithm developers misused healthcare spending data by overlooking its embedded bias,
demonstrating how data scientists lack deep contextual understanding of historical inequities can
perpetuate harm in public health initiatives (Ferryman, 2021). Thus, the biased data skews how
the algorithm model learns to evaluate patients and may cause an output of misdiagnosis, missed
diagnosis, and inappropriate suggestions.
A fundamental issue in AI development for mental health care is that there is a lack of
diversity in training data, as most AI models are trained on data from predominantly Western,
educated, industrialized, rich, and democratic (WEIRD) populations, which may not reflect the
experiences and needs of individuals from other backgrounds (Obermeyer et al., 2019). In the
context of mental health and AI data bias, the APA’s Diagnostic and Statistical Manual of
Mental Disorders, Fifth Edition Text Revision (DSM-5 TR) and International Classification of
Diseases, 10th Revision (ICD-10) are the current medical coding systems that classify mental
health diagnoses (Silverman et al., 2015). The data they collect are U.S.- and English-centric
and, as such, not necessarily applicable to immigrant and minority groups (Caetano, 2011). Data
on mental disorders are primarily based on studies of Western Caucasians. While cultural
variables are not included in the ICD when diagnosing mental disorders, the DSM-5 advises
practitioners not to make a diagnosis without considering cultural factors that may affect
assessment and diagnosis (Paniagua, 2018). Thus, AI systems trained on DSM-5 TR and ICD-10
data may result in biased results or the reinforcement of pre-existing prejudices (Straw &
Callison-Burch, 2020). Studies conducted by Tornero-Costa et al. (2023) between 2016 and 2021
revealed that AI applications are largely employed in the study of depressive disorders,

32
schizophrenia, and psychotic disorders in mental health research. Currently, there are few data
sets available for other mental health disorders, personality disorders, and comorbid disorders
used to train the AI application. Thus, there is a significant gap in understanding how they can be
applied to other conditions.
Algorithmic bias, on the other hand, is the design and functioning of the AI model that
synthesize training data and automate decision-making processes or assist human decisionmaking (Kordzadeh & Ghasemaghaei, 2021). Most algorithmic fairness frameworks were
developed in isolation from policy considerations and societal contexts, limiting their real-world
applicability (Gupta, 2023; C. Russell, 2023). Per Räsänen and Nyce (2013), the location of the
data scientists is an important factor. Data scientists who are geographically and sociodemographically away from data subjects and frontline workers often lack the crucial contextual
understanding needed to interpret health data meaningfully (Räsänen & Nyce, 2013). Biased data
and algorithms can perpetuate health inequities and contribute to uneven access to care (Zanna et
al., 2022).
Privacy and Attribution Issues
Integrating AI in the mental health field gives rise to privacy concerns about acquiring
and retaining protected health information since the revelation of sensitive information due to
privacy breaches may have significant implications, including the potential for stigma, prejudice,
and even oppression (Melcher & Torous, 2020). Patients may lack awareness or understanding
of consent regarding how AI systems use their personal data (Adeniyi et al., 2024). Even when
there is a consent form, consent can be overwhelming for a patient, especially patients with
language, cultural, cognitive, developmental, and mental health barriers (Appenzeller et al.,
2022).

33
Even de-identified data sets can carry privacy risks if triangulated with other data where
AI programs can identify patterns that can reveal confidential information (Wiepert et al., 2024).
Moreover, AI trained on vast datasets of other people’s information, such as a billion lines of
text, run the risk of generating output that unintentionally replicates existing content without
proper attribution, according to some experts (Cooper, 2023). This issue arises because the AI is
simply recombining patterns it has learned from the training data to which it was exposed. For
patient health data specifically, companies leveraging such data to develop healthcare AI do so
without crediting the original sources (Engle et al., 2019). Proper governance and restrictions
around attribution may be needed to ensure AI generative models do not simply recirculate old
insights as new ones or exploit data without acknowledgement.
Data Colonialism
The issues of data practices such as biases, privacy, consent, and attribution can
perpetuate existing systemic oppression in how AI technologies extract and exploit data from
vulnerable populations (Melcher & Torous, 2020). Big data and AI systems can reinforce
systemic oppression when developed without careful consideration of power dynamics and
systemic biases (Montiel & Uyheng, 2021). Such a system of oppression is data colonialism
(BlackDeer & Beeler, 2024). Data colonialism is the practice of extracting, controlling, and
exploiting data from individuals, communities, or nations, where large technology companies
and institutions benefit while the data contributors bear privacy risks and potential harms,
reinforcing existing power imbalances and systemic inequalities (Kohnke & Foung, 2024). The
actors in data colonialism can be collectively called “social quantification sector,” which refers
to companies that convert everyday human behaviors into monetizable data streams, profiting
from the analysis and commodification of social activities (Couldry & Mejias, 2018). According

34
to Couldry and Mejias (2018), the social quantification sector includes both big and small
hardware and software manufacturers, developers of social media platforms, online and offline
retails, and firms dedicated to data analysis and brokerage.
Although this research cannot identify the details of how these social quantification
sector companies benefit from user data, multiple sources reveal social media and a few AI
technologies companies exemplify data colonialism through the collection and monetization of
sensitive emotional and mental health information. Social media platforms analyze user behavior
to predict mental health crises not for intervention but to fuel targeted advertising (T. Wang &
Bashir, 2020). Mental health apps and online therapy platforms like BetterHelp and Talkspace
gather intimate details about users’ conditions and therapy sessions, process the data, and sell to
advertisers for target marketing (Federal Trade Commission [FTC], 2021). While the AI
application Replika claims not to share conversations or photos with advertising partners, such as
email and IP addresses, are shared with third parties for marketing purposes (Hardy & Allnutt,
2023). Beyond mental health, the DNA testing industry, as exemplified by 23andMe’s $300
million partnership with GlaxoSmithKline and $60 million deal with Genentech, demonstrates
data colonialism through its practice of extracting value from customers’ genetic information
while obscuring how this personal data becomes a profitable corporate asset used for drug
development (BlackDeer & Beeler, 2024; Moreau et al., 2020; Stoeklé et al., 2016).
While companies often claim data is aggregated and anonymized to enhance services or
inform product development, the reality is that this sensitive information flows into advertising
networks and third-party partnerships (Grundy et al., 2019; Kelly, 2024). The insights and
analytics from the data collected generate corporate profits rather than benefiting the users who
provided their most personal data (Couldry & Mejias, 2018; Ofulue & Benyoucef, 2022). Users

35
have little visibility into or control over how their mental health information is stored, shared, or
monetized, creating a profound power imbalance where vulnerable individuals’ data becomes a
corporate asset. This exploitation of mental health data for business interests, with minimal
return to users, demonstrates how data colonialism operates in the mind.
Data Decolonization and Data Sovereignty
Data decolonization and data sovereignty represent two complementary approaches to
addressing power imbalances in how data is collected, controlled, and used in our modern world
(Couldry & Mejias, 2018; Hummel et al., 2021). Modern data practices mirror historical
colonialism, where powerful entities extract and profit from personal data, just as colonial
powers once seized territories and resources for their gain (Ferryman, 2021). Decolonizing data
does not mean abandoning all data collection and analysis. Instead, it focuses on the concept of
data sovereignty of rejecting the data practices that appropriate, exploit, and reinforce existing
power structures that benefit the data collectors and not the data contributors (Couldry &
Mejias, 2018). Data decolonization focuses on transforming existing practices that perpetuate
colonial power dynamics by shifting control of data collection, ownership, and application back
to source communities (Leone, 2021; Shilton et al., 2021).
Data sovereignty complements these decolonial efforts by establishing explicit rights for
individuals and communities to control their personal information (Hummel et al., 2021). As
Hummel et al. (2021) explain, data sovereignty empowers people to determine how their data is
collected, processed, stored, and utilized according to their preferences and cultural norms. Data
decolonization and data sovereignty challenge the current paradigm of how the social
quantification sector controls personal and community data. Advocates for data decolonization
and data sovereignty envision a future where communities maintain authority over their

36
information, using it to serve their interests and advance their self-determination rather than
having it extracted for others’ benefit (Couldry & Mejias, 2018; Ferryman, 2021; Hummel et al.,
2021; Oguamanam, 2020).
Organizational Risks of AI Misuse
The lack of ethical oversight in AI technology can lead to severe consequences, such as
legal liabilities, potential harm, increased costs, and reputational damage (Terranova et al.,
2024). Recent cases of AI misuse as of 2025 highlight the significant risks and implications in
mental health and healthcare settings:
Legal Liabilities and Litigation Costs
The cases involving Character.AI, Koko, UnitedHealthcare, and Humana underscore the
potential for substantial legal repercussions when AI systems are deployed without appropriate
ethical and operational safeguards. In Character.AI’s wrongful death lawsuit, the claims of a
minor’s suicide due to unregulated chatbot interactions have placed the company at the center of
a highly sensitive legal dispute (Payne, 2024). Similarly, Koko’s lack of informed consent in
experimenting with AI-generated responses has led to intense scrutiny and backlash, with
potential legal liabilities due to breaches in patient consent and ethical norms (Edwards, 2023).
As of the time this paper is written, the liability cost for Character.Ai and Koko is unknown.
UnitedHealthcare and Humana face class-action lawsuits for allegedly wrongful denials
of medical claims, suggesting that companies deploying AI for patient management without
adequate human oversight can be held liable for negligent or detrimental outcomes (Laney,
2024). Laney (2024) furthered that the class-action lawsuit claimed that when these denials are
appealed to federal administrative law judges, approximately 90% are reversed, highlighting the

37
alleged inaccuracy of the algorithm. As of the time of this paper, the cost of defending these
lawsuits, settling claims, or facing potential judgments is unknown and can be significant.
Increased Risk of Harm and Life Loss
These cases illustrate that AI systems, when applied in healthcare without careful ethical
oversight, can exacerbate mental health crises or deny essential care. For instance,
Character.AI’s chatbot allegedly exacerbated the mental health struggles of a vulnerable user,
culminating in the user’s tragic death. This case highlights how a lack of controls on AI
interactions in sensitive mental health contexts can directly contribute to severe outcomes,
including loss of life (Browne, 2022). In the healthcare context, UnitedHealthcare’s AI allegedly
led to premature discharge of patients, which in some cases reportedly resulted in death,
illustrating how algorithm-driven decisions can undermine patient well-being. Such risks
emphasize the need for human oversight to prevent AI from making isolated decisions that could
endanger lives.
Ethical Breaches and Reputational Damage
The Koko AI experiment raised ethical concerns about informed consent and the
boundaries of AI’s role in mental health care. Conducting AI-driven experiments without patient
knowledge or consent breaches fundamental ethical principles, damaging public trust in AIdriven healthcare solutions. Similarly, Character.AI’s case has sparked public outcry over the
potential harm of emotionally intensive AI chatbots, which, in the absence of ethical boundaries,
could exploit or worsen a user’s mental state. Such ethical breaches erode patient trust, affecting
the reputation of both specific companies and AI in mental health care more broadly, possibly
hindering future AI adoption.

38
Regulatory and Compliance Costs
Cases such as UnitedHealthcare and Humana’s use of nH Predict indicate that healthcare
AI systems can face scrutiny from regulatory bodies, leading to tighter regulations and increased
compliance costs (Mello & Rose, 2024). These legal challenges highlight the urgent need for
more comprehensive frameworks governing AI’s role in patient care. With heightened regulatory
interest, organizations may be required to invest in compliance infrastructure, data audits,
transparency measures, and risk assessment tools to avoid potential sanctions. As of now, the
regulatory cost is unknown and these regulatory requirements can impose further operational
costs but are necessary to mitigate the inherent risks of using AI in healthcare contexts.
Ownership of Data and AI Business Models
The integration of AI tools into mental health services presents both opportunities and
significant challenges for healthcare leaders (Alhur, 2024). Currently, there are AI mental health
technologies in the market that offer promising capabilities in automating tasks like chatbots,
therapy session recording, SOAP notes, and treatment plan generation using automatic speech
recognition and large language models, saving time and improving documentation quality
(Biswas & Talukdar, 2024; Rezaeikhonakdar, 2023). Many of these AI mental health
technologies leverage existing pre-trained large language models application programming
interfaces, such as GPT-4, Claude, and LLaMA, for speech-to-text transcription and structured
documentation generation without building their own infrastructures to keep cost low, thus
lowering the barrier of entry for developing mental health AI (Hadar-Shoval et al., 2024; Klang
et al., 2024).
The integration of AI in mental health services, particularly by smaller technology
companies utilizing models like OpenAI’s GPT-4, raises significant data privacy concerns

39
(Rezaeikhonakdar, 2023). The companies owning these existing models retain the rights to data
processed through their systems, creating potential privacy vulnerabilities for sensitive protected
health information (PHI), and these companies are not regulated under HIPAA laws (Belani et
al., 2021; Margam, 2023). This data ownership issue compounds the already existing problems
of privacy, equity, inadequate validation for mental health applications, and privacy concerns
when processing data through third-party servers (Chiruvella & Guddati, 2021). However, the
rapid evolution of AI technology has outpaced the updates to HIPAA regulations, leading to
potential gaps in privacy safeguards (Marks & Haupt, 2023).
In the world of technology, high‐quality start‐up companies go public, while low‐quality
companies are acquired (Guo et al., 2015; Weber et al., 2022). Large insurance companies, such
as Travelers Insurance, Centene Corporation, and Cigna, have been purchasing smaller AI
companies for the data for risk assessment and underwriting purposes (Sekerak, 2024). These
transitions pose risks of data misuse, where personal health information could be exploited to
deny claims or assert pre-existing conditions, thereby limiting individuals’ access to insurance
coverage (Poufinas et al., 2023). Such practices have been observed in the healthcare industry,
where AI-driven systems have been employed to deny medical claims efficiently, often without
proper review, prioritizing profits over patient care (Napolitano, 2023). Additionally, if these
acquiring companies are not classified as healthcare providers, they may not be obligated to
comply with HIPAA regulations, further exacerbating privacy concerns (Rezaeikhonakdar,
2023). The lack of clear privacy laws governing tech companies that interact with non-HIPAAcovered entities creates a regulatory quagmire that could be exploited, leading to potential
violations of individuals’ privacy rights (Fox Rothschild LLP, 2019). Mental health
organizations adopting third-party AI technologies must control patient data fully to prevent data

40
colonialism. Leadership should establish clear terms, secure organization data ownership rights,
transfer data to other systems, and ensure data deletion upon terminating vendor relationships to
prevent AI companies from holding patient data hostage or profiting from sensitive mental health
information after business relationships end.
AI Regulations in Mental Health
Federal and state-level regulations governing AI use in mental health services are
evolving to address concerns about patient privacy, data security, and ethical considerations
(Bordelon, 2023). At the federal level, HIPAA sets standards for protecting sensitive patient
information, which apply to AI tools handling mental health data. Additionally, the U.S. Food
and Drug Administration (FDA) has been developing frameworks to regulate AI-based medical
devices, ensuring they meet safety and efficacy standards.
Health Insurance Portability and Accountability Act
The HIPAA is a U.S. federal law implemented in 1996 to safeguard sensitive PHI from
being disclosed without a patient’s explicit written consent, except in rare cases such as when a
patient poses a serious threat to themselves or others (Hodge & Gostin, 2004). A primary
objective of the Privacy Rule is to ensure that individuals’ health information is securely
protected while enabling the exchange of health data necessary to deliver and promote highquality healthcare and safeguard public health and well-being (O’Connor & Matthews, 2011).
The HIPAA applies to mental health by requiring providers to establish robust safeguards for
storing and transmitting PHI, including patient diagnoses, treatment plans, and session notes
(Department of Health and Human Services [HHS], 2022). Its most significant HIPAA
component for mental health professionals is the Privacy Rule, which ensures that PHI is
handled with strict confidentiality and de-identification (U.S. Department of Health and Human

41
Services [HHS], 2022). The strict privacy standards mandated by HIPAA uphold fundamental
ethical principles by preserving patient autonomy over personal information while protecting
vulnerable individuals with mental health conditions from discrimination and exploitation
(Beauchamp & Childress, 2012).
As of the writing of this paper, the implementation of AI in mental healthcare requires
rigorous HIPAA compliance protocols to maintain patient confidentiality and data security,
incorporating extensive data encryption protocols coupled with granular role-based access
controls that restrict information accessibility to authorized healthcare professionals. HIPAA
compliance requires robust security measures to systematic audit logging for breach detection
and prevention, alongside regular vulnerability assessments to identify and remediate potential
security gaps, while organizations must institute thorough staff training programs focused on
data privacy compliance. HIPAA compliance requires covered entities to develop formalized
business associate agreements with AI technology vendors to ensure consistent HIPAA
adherence throughout the data processing pipeline. HIPAA compliance extends to data lifecycle
management, particularly regarding secure data destruction protocols when information exceeds
retention requirements. This comprehensive framework establishes an integrated approach to
protecting sensitive mental health data while advancing AI-driven therapeutic interventions
(Adams, 2024; Mayover, 2024; Rezaeikhonakdar, 2023).
HIPAA has significant limitations in the era of AI. HIPAA’s strict regulation on deidentified data as a safeguard can be undermined by AI’s ability to re-identify individuals
through data triangulation from sources such as location, gender, date of birth, and progress of
treatment (Rocher et al., 2019). Furthermore, HIPAA only applies to healthcare entities, leaving
health-related data from mental health chatbots, fitness apps, and wearables unregulated, a

42
growing concern as AI integrates diverse data sources (Belani et al., 2021; Rezaeikhonakdar,
2023). AI’s large datasets also conflict with HIPAA’s principle of data minimization (Arigbabu
et al., 2024). While FAIR principles promote AI data’s findability, accessibility, interoperability,
and reusability to enhance transparency and scientific reproducibility, these guidelines may be in
conflict with HIPAA’s strict privacy requirement that prioritizes patient confidentiality over
open data sharing and reuse (Bhatia et al., 2020; Chen et al., 2022; McGraw, 2012; Suver et al.,
2023; Wilkinson et al., 2016). Given the limitations of HIPAA in the context of AI, MHLs
should implement AI responsibly by establishing organizational data governance, ensuring
transparency and patient consent, fostering interdisciplinary collaboration, training staff, and
continuously mitigating training data bias. They should also pilot AI tools carefully and advocate
for policy reforms to align HIPAA protections with FAIR principles while safeguarding patient
trust and equity (Rezaeikhonakdar, 2023).
FDA and FTC
The FDA and the FTC have increased oversight of digital mental health companies amid
the sector’s rapid expansion (Haupt & Marks, 2024; Warraich et al., 2024). While the FDA
focuses on the safety and effectiveness of medical devices, data privacy concerns related to
mental health apps are often the FTC (Bright, 2000; FTC, 2023). This heightened scrutiny
operates along two primary vectors: the FDA’s focus on clinical efficacy and safety, particularly
regarding AI-driven diagnostic and treatment tools, and the FTC’s enforcement of data privacy
regulations in the digital health space. In 2021, the FDA released a comprehensive framework
for AI/ML-based medical software, establishing protocols for iterative algorithm improvement
while maintaining patient safety standards (FDA, 2021). This regulatory approach acknowledges

43
the unique challenges posed by adaptive AI systems in healthcare, particularly the need to
balance innovation with clinical validation.
The FTC has demonstrated aggressive enforcement through significant financial penalties
against industry players. The $7.8 million BetterHelp settlement in 2023 and the proposed $7
million Cerebral settlement in 2024 establish precedent for substantial consequences regarding
unauthorized health data sharing. The Flo Health case further exemplifies the FTC’s stance on
informed consent requirements and mandatory privacy audits in digital health (FTC, 2021,
2024a, 2024b). Together, these regulatory bodies create a synergistic balance: the FDA’s
proactive scrutiny ensures that digital tools are safe and effective, while the FTC’s reactive
oversight reinforces trust by addressing breaches of privacy or misuse of sensitive data. This
dual-pronged strategy encourages technological innovation but also strengthens public
confidence in the safety, efficacy, and ethical handling of digital mental health solutions in an era
of rapid technological evolution.
Federal- and State-Level Governance
Federal and state-level regulations governing the use of AI in mental health services are
evolving to address concerns about patient privacy, data security, and ethical considerations
(Douglas McNair et al., 2019). The recently updated federal Section 1557 Final Rule, which was
amended in July to prevent AI tools and algorithms used for clinical care and administrative
activities from discrimination among underrepresented or marginalized patients (LaRose, 2024).
Forty states, such as Colorado, California, and Connecticut are stepping up to address the issue,
with successfully passing AI-related legislation this year (Trang, 2024). In Utah, the Artificial
Intelligence Policy Act requires state-licensed professionals to disclose when a consumer is
interacting with generative AI as a means to stop regulated professions from blaming violations

44
of any consumer protection laws on AI’s mistakes (Scheuch, 2024; Utah State Legislature,
2024). Virginia’s HB2154 bill requires hospitals, nursing homes, and certified nursing facilities
to establish policies for the use of intelligent personal assistants (Hilliard, 2024; Virginia’s
Legislative Information System, 2021). Colorado enacted Senate Bill 24-205 mandates that
businesses using high-risk AI systems in sectors like healthcare must disclose their AI usage and
implement measures to prevent algorithmic discrimination (State of Colorado, 2024).
California’s AB3030 requires a variety of healthcare providers, such as hospitals, clinics,
medical groups, and individual licensed health providers, that use GenAI to generate patient
communications relating to a patient’s clinical information (California Legislative Information,
2024a). Also, SB 1120 requires health plans and disability insurers that use algorithms, AI
(including GenAI), and other software tools (or who use a vendor that uses such tools) for
utilization review or management functions to ensure compliance with certain specified
requirements (California Legislative Information, 2024b).
Federal and state regulations shape AI use in mental health services by ensuring data
protection, safety, and ethical standards. HIPAA safeguards patient privacy, while the FDA
regulates AI-driven medical tools for safety and efficacy. The FTC enforces fair practices in AI
marketing, and states like California and Colorado address discrimination and ethical concerns.
These regulations provide a foundation for professional organizations to establish ethical
guidelines tailored to mental health AI applications discussed in the next section.
Principlism Ethical Guidelines in Mental Health
The ambiguity of ethical principles is difficult to integrate into AI’s mathematical and
programming languages, and currently, there has not been a definitive ethical guide for the use of
AI in mental health other than the general healthcare principlism ethical framework and overall

45
best practices derived from other regulation bodies. (Buckwalter & California Psychological
Association, 2023; Shahriari & Shahriari, 2017). Given this lack of AI-specific guidelines,
examining the principlism framework in detail can help understand the framework’s relevance
and constraints for AI in mental health.
The ethics of AI resulted in numerous implications still debated in the field. Ethical
concerns related to privacy, consent, accountability, and potential biases when utilizing patient
data and algorithms have been raised (Joerin et al., 2020). Currently, principlism is the most
widely adopted bioethics framework in healthcare, and the origin of principlism traces back to
two dark chapters of medical history: the Nazi regime’s forced human experimentation and the
Tuskegee syphilis experimentation on 600 African men without consent and cure (Dale, 2023).
In 1979, two Georgetown University professors, Tom Beauchamp and James Childress,
published Principles of Biomedical Ethics, intended to prevent future crimes and atrocities
against human research subjects (Holm, 2002). Principlism is a four-principles approach
framework: autonomy, where the patient has the right to make their own decisions; beneficence
to do what is good for the patient; non-maleficent to avoid causing harm to the patient; and
justice to treat the patient fairly (Beauchamp & Childress, 1979). While it provides a primer for
bioethics, critics of principlism said it is obscure and fails to provide practical guidelines by its
eclectic and unsystematic use of moral theory, and it is simply a checklist at best (Dale, 2023).
Principlism is simply too vague to apply to all things healthcare, including AI, and too high-level
to apply to ground-level situations, but disregarding it would risk social rejections from the
healthcare professional community (Seger, 2022). Mittelstadt (2019) challenged the
effectiveness of principle-based ethical approaches to AI ethics, warned against using this
simplistic approach, and suggested specific robust AI governance for policy and auditing

46
measures. However, per Baum (2016), explicit rules and requirements will generate timeconsuming extra work, stifling AI advancement, and AI ethics in healthcare and mental health
must start somewhere. Principlism serves as a familiar starting point that needs to pair up with
additional guidelines (Seger, 2022).
In the United States, several major mental health professional associations enforce ethical
guidelines, including the American Psychiatric Association, APA, National Association of Social
Workers, American Counseling Association, American Association for Marriage and Family
Therapy, American Mental Health Counselors Association, and Psychiatric Nurses Association.
These organizations establish educational standards, practice guidelines, professional ethics
codes, advocacy priorities, and resources for their members (Borghouts et al., 2021). In
summary, the professional ethical codes prioritize patient welfare, preserving confidentiality,
maintaining safe recordkeeping, practicing within a scope of competency, avoiding conflicts and
prejudice, and increasing public awareness in accordance with humanistic values (Bucky et al.,
2013). As of the completion of this paper, these associations have started to develop a formal
code of ethics for AI use, and their efforts are still ongoing. Therefore, only a few of the
recommendations are mandatory, and compliance is voluntary.
Ethical Versus Responsible AI Implementation
In examining the literature, I noted that the distinction between ethical and responsible AI
is often blurred, as these terms are commonly used interchangeably. Ethical AI in mental health
primarily focuses on aligning AI technologies with established healthcare principlism ethics such
as autonomy, non-maleficence, beneficence, and justice (Vayena et al., 2018). This paper
identifies that the concept of responsible AI adoption goes beyond a sole focus on ethics by
incorporating key principles such as accountability, foresight, adaptability, and transparency

47
(Akbarighatar, 2024). This approach ensures that AI systems respect patient privacy, avoid harm,
and promote well-being by supporting clinicians in delivering quality care.
Many countries, government agencies, and professional organizations are establishing
legal frameworks and ethical standards to promote responsible AI development and use. The
U.S. AI Bill of Rights, the European Union’s AI Act, the World Health Organization, the
Montreal Declaration for Responsible AI, and IEEE’s Ethically Aligned Design share a number
of proposed guidelines (IEEE, 2019; Buruk et al., 2020; Cannarsa, 2021; European Commission,
2021; World Health Organization, 2021). Key themes emphasized across these organizations are
transparency, accountability, mitigating bias and other risks, prioritizing human oversight,
promoting inclusivity, and ensuring algorithmic fairness and explainability (Cirqueira et al.,
2021; Došilović et al., 2018). Despite some nuances, there is increasing global alignment around
establishing ethical norms and best practices aimed at protecting individuals and communities
from potential harm by AI (Borghouts et al., 2021). It will be necessary to further collaborate
among nations and experts to translate these shared principles into enforceable policies that will
steer the AI industry toward greater responsibility and public benefit (Siala & Wang, 2022). In
addition to principlism, the mental health industry should also consider the following common
ethical AI guidelines across different AI governance organizations. Table 1 presents a synthesis
and merging of the key elements from principlism bioethics and various joint organizations’
frameworks, adapting them to AI implementation (Akbarighatar, 2024; Beauchamp & Childress,
1979; Buruk et al., 2020; European Commission, 2021; Pujari et al., 2023; Shahriari & Shahriari,
2017; The White House, 2023).

48
Table 1
Responsible AI Principles
Domain Principle Description
Principlism bioethics Autonomy AI system where the users have the right to make
their own decisions in what data to give to train
the AI system (ongoing easy opt-out process);
allowing the users to choose from multiple
decisions that include AI and non-AI
interventions.
Beneficence AI system that does what is good for the users.
Non-maleficent AI system that avoids causing harm or minimizes
risks to the users; it must refer back to autonomy
principle above to allow users to choose harmreduction interventions.
Justice AI system that allows equal access to services and
interventions, closing healthcare disparities,
protecting vulnerable populations, and
protecting human dignity, satisfy legal
requirements.
General AI ethics Reliability and
safety
AI systems should incorporate preventative and
backup measures against failures and accidents.
Privacy Non-intrusive system that uses data that is legally
and properly collected with individuals’ consent
Security AI with safeguard that protects data and control
access level to ensure authorized use
Accountability The system, system owners, and system designers
are held responsible and accept consequences for
all outputs and decisions.
Explainability The system’s ability to explain and be open about
how the system functions and make decisions
Intelligibility Users must be able to understand AI’s reasoning
and decision-making (information that users can
understand).
Transparency AI system has clear process in revealing its data,
methodology, capabilities, and decision-making
(easy-to-understand and easy-to-use commands
to “open the hood” to intelligible data and
reasonings).
Inclusiveness AI system trained on a diverse set of data
consisting of individuals and perspectives,
regardless of their unique circumstances.

49
Domain Principle Description
Fairness AI systems with data trained and an algorithm
designed to make fair decisions for everyone
without discriminating
Importance of a Person-to-Person Therapeutic Relationship
The person-to-person therapeutic relationship between clients and providers is crucial for
successful treatment outcomes (Ackerman & Hilsenroth, 2003; Norcross & Lambert, 2018).
Much research consistently underscores the significance of empathy, rapport, and trust in the
therapeutic alliance between the clinician and the client since these elements promote effective
psychotherapeutic outcomes (Battaglia, 2019; Elliott et al., 2011; Norcross & Lambert, 2018).
Establishing interpersonal aspects serves as the fundamental basis for developing treatments.
These factors create an environment where clients feel acknowledged, affirmed, and
comprehended. Consequently, this fosters their inclination to actively participate in the
therapeutic process (Del Re et al., 2012; Norcross & Lambert, 2018). Nonetheless, there are
apprehensions regarding the mental healthcare field’s growing incorporation of AI, as it can
disrupt or diminish the essential human connections and emotional attachments that facilitate
successful therapy. Consequently, this presents significant risks to the treatment procedure
(Borghouts et al., 2021).
When AI makes medical decisions, it can potentially alter the traditional therapeutic
relationship by bringing into the equation programmers, product developers, and AI-driven tools
that do not have medical or mental health expertise. This dynamic will likely pose a challenge to
the established medical ethics framework of shared decision-making, as Molnár-Gábor (2019),
Kerasidou (2020), and McDougall (2018) discussed. Therefore, given the significance of the

50
clinician-patient rapport, it requires ensuring the proper integration of AI technology to augment
rather than supplant human physicians (Denecke et al., 2019). Artificial intelligence technologies
can enhance therapists’ work by analyzing therapy sessions, automating administrative activities,
and providing data-driven insights to facilitate informed treatment choices (Borghouts et al.,
2021). Using AI to enhance human capabilities can alleviate clinicians’ workload, making it
possible to maintain the therapeutic alliance while allowing clinicians to concentrate on the
fundamental elements of human interaction and compassion (Denecke et al., 2019; Schueller et
al., 2019). However, excessive dependence on technology in mental health can undermine
therapeutic alliance and potentially compromise treatment outcomes, as The British Journal of
Psychiatry emphasizes that while digital tools can complement traditional therapies, they should
not replace the human connection essential for personalized care (Hollis et al., 2015). According
to Schueller et al. (2019), responsible leadership in the mental healthcare sector must guide AI’s
deployment to enhance, rather than undermine, the interpersonal therapeutic connections
essential for successful psychotherapy.
Technology-Organization-Environment Framework
Louis G. Tornatzky and Mitchell Fleischer developed a TOE model in 1990. The TOE
framework is an organizational-level theory developed to model the context surrounding a firm’s
adoption of technology as it focuses on three contexts: organization, environment, and
technology (Tornatzky et al., 1990). Technological context refers to the existing and emerging
technologies relevant to the organization, including perceptions of benefits, accuracy,
trustworthiness, and compatibility (Gangwar et al., 2015). Organizational context includes
factors like managerial structure, leadership support, company culture, funding, communication
processes, training resources, and internal networks that shape adoption (Oliveira & Martins,

51
2010). Environmental context encompasses the regulatory environment, professional code of
ethics, industry standards, competitive landscape, vendor ecosystem, and other external forces
that enable or constrain adoption (Oliveira et al., 2014). Figure 1 and Table 2 present the TOE
framework.
Figure 1
Technology-Organization-Environment Framework

52
Table 2
Technology-Organization-Environment Framework
Context Constructed concepts Detailed factors
Technological
All technologies supported inside/outside
the organization
The ability for technology adoption and
suitability of the current technology to
the organization
Relative advantage
Conformance
Technology complexity
Organizational
Refers to the inherent characteristics and
resources the organization possesses
Management leadership and
communication play a crucial role in
innovation.
Organization scale and resource
availability are also important for
decision-making.
Corporate size
Project range
Management support
HR scale
Competitive advantage
Available resources
Environmental
Effectiveness and efficiency factors for
organizations’ business activities
Includes organization’s industry,
competitors, governmental regulations,
business partners, etc.
Market environment
Competition intensity
Government policy and
regulations
Infrastructure of technological
resources
While the TOE framework has been applied and proven to be effective in predicting the
adoption of technology innovation in various industries (Gangwar et al., 2015), there has been
limited direct application and research involving the mental health sector specifically. As such,
there are gaps in understanding how the TOE model may explain technology adoption decisions
and processes in this field (Chatterjee et al., 2021). The unique considerations of healthcare,
strong emphasis on ethics and relationships, and extensive regulations governing providers
suggest value in examining the TOE factors that influence the adoption of innovations like AI in
this sector (Kimiagari & Baei, 2021). Further research adapting and validating the TOE

53
framework in mental health contexts would provide beneficial new insights into the
technological, organizational, and environmental forces at play.
By taking this TOE framework approach, the interview data may uncover MHLs’ key
perceptions around AI technology. It allows for a multilayered understanding of the personal,
organizational, and environmental dimensions impacting the adoption and ethical
implementation of AI (Na et al., 2022). The data can also highlight the organizational
implementation needs, such as training, leadership support, resource allocation, and change
management tactics required to enable smooth adoption (Misra & Mondal, 2011). In essence, the
qualitative insights gathered can reveal how wider contextual factors such as professional ethics,
regulations, and standards in the mental healthcare environment shape attitudes toward
responsible AI integration. This understanding may help organizations develop initiatives,
processes, and strategies focused on facilitating appropriate AI deployment that complements
clinical judgment and preserves the priority of compassionate, human-focused therapeutic
relationships.
Chapter Two provided the technology overview, history of AI, the development of AI, AI
ethical issues in mental health, distinctions between ethics and responsible AI, as well as the use
of the framework of TOE to examine the issue of responsible AI implementation. Chapter Three
details the methodology, data collection, and analysis for this study.

54
Chapter Three: Methodology
This study addressed the problem of how to ethically adopt AI in mental health
organizations while retaining human expertise and care values. Through semi-structured
interviews, the research uncovered MHLs’ perspective of AI technology, as well as
organizational and environmental concerns that influence its implementation. This work sought
to identify the internal and external resources necessary to develop an initiative that guides
policy and practice for the responsible adoption of AI while maintaining the core values of
mental health care, such as human expertise, judgment, and connection. This chapter outlines the
method to recruit participants for semi-structured interviews. Additionally, considerations
regarding credibility, trustworthiness, biases, and ethical foundation are discussed later in the
chapter.
Research Questions
1. What are the MHLs’ perceptions of adopting AI technologies in their organizations?
2. What organizational factors influence MHLs to adopt AI technologies responsibly in
their organization?
3. What external factors are MHLs concerned about the most when evaluating their
organization’s readiness to adopt AI technologies?
Overview of Design
This study employed a qualitative approach, utilizing semi-structured interviews to
answer the research questions. Merriam and Tisdell (2016) defined qualitative research as
examining how individuals construct meaning from living experiences. An exploratory
qualitative design was appropriate for this study to gain an understanding of MHLs’ perspectives
regarding AI technologies, what meaning they attach to them, and what might help them adopt

55
AI responsibly in their organizations. I chose semi-structured interviews for this study because
they allow for a holistic perspective of the individual and make sense of one individual’s
experiences. In combination, these studies may provide a deeper understanding of the contexts
surrounding the participants and the intended outcomes (Merriam & Tisdell, 2016).
Participating Stakeholders
This research utilized purposive sampling and recruited 11 leaders at various levels of
responsibility from community mental health and group practice organizations to participate in
interviews. Eligible leadership roles included frontline supervisors overseeing employees, middle
managers supervising frontline supervisors, senior executives supervising middle managers, and
thought leaders in advisory/consultative roles providing expertise on technology, ethics, and
policy. The goal of this study was to elicit perspectives from leaders from diverse vantage points
to develop guidelines for responsible adoption of AI by identifying organizational workflows,
decision-making, priorities, needs, and values. Interviews with leaders at different levels
provided insights into technology integration needs and risks from multiple leadership lenses to
help guide ethical innovation and process changes.
Interview Sampling Criterion and Rationale
Participants were MHLs who met the following criteria.
• They have a background in community mental health or group practice settings.
• They have at least 1 year of experience in a leadership position related to clinical
operations and clinical technology, which may intersect with a clinical license,
though licensure is not mandatory for inclusion.
The goal was to target leaders with extensive mental health experience who are currently
in organizational decision-making roles related to administration and care delivery workflows.

56
Exclusions focus the sample on those in community-based and group practice organizational
settings rather than academics or solo private practices. Exclusion criteria are as follows:
• professionals whose leadership experience is exclusive to self-owned, single-provider
practices
• entry-level practitioners without leadership or management duties
This research recruited leaders with direct oversight of patient care, organizational
workflows, personnel, and operations that influence the adoption of new technologies. By
sampling U.S.-based MHLs with managerial experience guiding clinical teams and processes,
the aim was to elicit insights into organizational needs, values, and risks to responsibly inform
AI integration in ways that align with therapeutic goals. I excluded clinical supervisors without
personnel management duties since the focus was on leaders embedded in organizational
operations beyond direct clinical training roles.
I recruited MHLs from the mental health industry across the country whom I identified
through my LinkedIn and Facebook networks. After interviewing eight MHLs revealed limited
knowledge of AI systems, the sampling strategy expanded to include three participants with
clinical expertise and technology leadership roles. This intentional sample diversification helped
capture broader perspectives on AI implementation in mental healthcare settings.
Research Setting
As a research setting, I conducted interviews using Zoom video conferencing. The
interviews lasted up to an hour. Using Zoom as the primary interview tool was a practical
decision as video conferencing has become common and convenient following the COVID-19
pandemic. Online interviewing also allows for a much broader geographical reach. In-person
interviews would have been limited to MHLs in the San Francisco Bay Area, where I reside;

57
Zoom allowed participants to participate from anywhere in the country. Zoom’s recording
capability captured each interview’s audio and video. As Merriam and Tisdell (2016) noted,
recording interviews is beneficial because it ensures the preservation of all information shared
during the interview for later analysis. I informed participants of the session’s confidentiality and
privacy, as well as the recording of the session. I gave the participants an overview of the manner
in which I would ask questions.
The Researcher
As a first-generation Chinese American immigrant and clinical director of a mental health
organization in Silicon Valley, my positionality significantly shapes this research. I work
remotely, conduct meetings via videoconferencing, and am comfortable with and even favor
technology that enhances my work life with minimal travel. My 15 years of clinical experience,
including 9 years managing a mental health clinic, drove my interest in exploring AI’s potential
to address pressing challenges in mental healthcare delivery, particularly staff burnout and
documentation struggles. Having witnessed the internet’s evolution and impact on data privacy
since childhood, I bring technological optimism and measured caution to this research.
My love for technology began with a struggle as a 4th-grade immigrant with no English
skills. My teacher assigned Black Beauty; I still felt the overwhelming sensation when I first
opened the book, filled with dense English sentences and pages of unfamiliar words. Moving
glacially, looking up definitions for every third word, I could not finish the assignment in time.
Though my teacher was sympathetic to my limited English skills, I felt ashamed for being a
failure. That summer, my mother bought a newly released electronic dictionary because my
sisters needed it for SAT preparation. Even sharing the device among three siblings for a third of
the time each was faster than using a paper dictionary. Fortunately, that year’s summer school

58
assigned Black Beauty again, and I finally finished the book with pride. This experience taught
me that technology can enhance our capabilities. Today, my Kindle displays instant word
definitions with a simple tap, a far cry from my early struggles with paper dictionaries. With
technology, we may lose some rudimentary skills, like searching through paper dictionaries, but
we gain comprehension, analytic, and decision skills. We can learn the rudimentary skills when
needed, just as my tech-immersed daughters are now learning to use paper Spanish–English
dictionaries in their middle school Spanish class.
A recent experience with purchasing life insurance exposed a challenge with my
healthcare data sovereignty. During the underwriting process, the insurance company reviewed
my medical records and discovered that one of my physicians had documented a serious
diagnosis that I never actually had and was never treated for years ago. This incorrect
information now lives in my permanent medical record, creating significant complications. The
situation became even more frustrating when we learned that the physician had relocated to
another state, making it nearly impossible to have them correct the record. Although not AIrelated, this experience revealed how little sovereignty I have over my medical data. Despite the
data being about my health, I found myself entangled in bureaucratic red tape, unable to quickly
correct this critical error in my record. While the information belongs to me in principle, the
practical reality is that I have limited power to ensure its accuracy or make necessary corrections,
even when errors could significantly impact important life decisions like insurance coverage.
This situation perfectly illustrates my frustration as a medical consumer and why I care about
data sovereignty in healthcare.
My being an eager technology adopter and frustrated healthcare consumer, as well as my
immersion in Silicon Valley’s tech-forward culture, influence my attitude towards AI technology

59
in mental healthcare. As a mental health clinic director interested in implementing AI, I
recognize several potential biases in my research approach. My confirmation bias may have
favored evidence supporting AI’s benefits in mental health care while overlooking contradictory
information. As a Chinese immigrant female with a postgraduate education, my in-group bias
could have led me to identify more closely with similar individuals, potentially neglecting
perspectives from different backgrounds. My professional bias as a mental health professional
and assumptions about leadership and AI in patient care may have influenced my research focus.
My position as clinic director might have triggered social desirability bias, where participants
provided answers they thought I wanted to hear. My current role could have created status quo
bias, affecting my interpretation of findings.
To maintain objectivity, I implemented several strategies during the interview. First, I
used a consistent, structured interview protocol for all participants to ensure uniformity in data
collection. During interviews, I maintained neutrality by avoiding sharing my opinions or
experiences and using neutral therapist acknowledgments such as “Could you tell me more about
that?” rather than evaluative responses. I documented participants’ responses verbatim to ensure
accuracy and recorded all interviews for precise transcription. I recruited MHLs with different
cultures, backgrounds, and locations for diverse viewpoints. While my professional network
connections to some participants may introduce social desirability bias, this insider perspective
also enables a more profound understanding of the field’s complexities. I want to implement AI
solutions in my organization responsibly; while not a direct conflict of interest, it requires
acknowledgment as it may influence my interpretation of the data. This study’s value is not in
achieving perfect objectivity but in leveraging my unique position at the intersection of clinical
practice, organizational leadership, and technological innovation to explore the complexities of

60
responsible AI adoption in mental healthcare. My mental health professional and technology
enthusiast background provides a distinctive lens for examining how leaders navigate these
challenges while maintaining ethical practices and therapeutic integrity.
Interview Approach
The interviews were semi-structured to deeply explore MHLs’ experiences and
perspectives regarding AI adoptions in their organizations. The primary instrument was a semistructured interview protocol containing open-ended questions tied to the research aims while
allowing flexibility for probes during the conversation. A semi-structured interview utilizes a
predefined set of open-ended questions but maintains flexibility in the order and wording
(Merriam & Tisdell, 2016). The interview guide was organized to facilitate a natural
conversation flow while ensuring comprehensive coverage of research themes. Questions served
as a guide while allowing spontaneity to follow unexpected insights that arise.
Each main question was accompanied by potential follow-up probes designed to deepen
the discussion. For instance, when discussing organizational readiness, probes explored specific
challenges, success factors, and decision-making processes. This fluid approach allows
researchers to listen closely to the interviewee’s perspectives as the discussion organically
evolves, adapting and probing with additional questions as necessary (Merriam & Tisdell, 2016).
Semi-structured interviewing allows for spontaneous customization based on each participant’s
worldview and the dynamics of the discussion rather than strictly dictating the process. In
essence, flexible interviewing yields richer data than rigidly structured interviews limited to
scripted inquiries. Semi-structured interviews are well suited to in-depth qualitative research due
to their focus and adaptability, allowing participants to freely share their subjective perspectives
and experiences while remaining on task throughout the interview.

61
Data Collection and Documentation
Semi-structured interviews served as the primary data source, following Merriam and
Tisdell’s (2015) qualitative research protocol. The interviews, lasting 60 to 90 minutes, were
conducted via video conference platforms with purposively sampled participants who met
inclusion criteria due to their roles as MHLs. The protocol incorporated flexibility to adapt
questioning based on participants’ responses while ensuring consistent coverage of core research
topics.
A comprehensive documentation system captured both verbal and non-verbal aspects of
the interviews. Following Creswell and Creswell’s (2017) recommendations, interviews were
recorded via Zoom with backup transcription through Otter.ai to preserve participants’ exact
responses and emphasis on particular topics. Concurrent field notes documented significant
reactions, body language, and contextual observations that audio recordings alone could not
capture (Sutton & Austin, 2015). This detailed note-taking approach recorded environmental
contexts, behaviors, and non-verbal cues that provided crucial interpretive context. The
documentation particularly focused on capturing participants’ real-world examples,
organizational stories, and specific experiences that illustrated their perspectives on AI adoption
in mental healthcare settings.
To maintain research integrity and participant confidentiality, unique numeric identifiers
linked all data sources—audio recordings, field notes, and transcripts—for each participant. This
systematic documentation method preserved both verbal content and meaningful context for
analysis while enabling consistent data organization across all interviews. The combination of
recorded interviews, field notes, and secure transcription services followed McLellan et al.’s

62
(2003) guidance for efficient conversion of verbal data into analyzable text while maintaining
data security and confidentiality.
Data Analysis
The analysis process followed Patton’s (1999) guidelines for ensuring credibility in
qualitative inquiry through rigorous methods, researcher competence, and systematic
documentation. Interview recordings were first transcribed verbatim and verified for accuracy.
Using qualitative coding methods, I conducted a systematic analysis of the text transcripts to
identify themes and patterns relevant to the research questions. The analysis process involved
multiple phases: initial open coding to identify key concepts, focused coding to develop
categories, and theoretical coding to establish relationships between categories.
To enhance trustworthiness, I employed several validation strategies. Data triangulation
involved cross-referencing findings with organizational documents and field notes to substantiate
emerging themes. ATLAS.ti software facilitated the organization and management of coded
extracts, category development, and analytic memos. A detailed codebook maintained clear
definitions and application rules for each code, while an audit trail documented key analytic
decisions and their rationale. Regular peer consultations provided analyst triangulation, helping
to verify coding decisions and theme development.
Throughout the analysis, I maintained close alignment with the study’s theoretical
framework, interpreting emerging themes within the context of the TOE framework and existing
literature on AI adoption in healthcare settings. This systematic approach to data analysis
supported the development of findings that directly addressed the research objectives while
maintaining methodological rigor.

63
Research Credibility and Trustworthiness
Merriam and Tisdell (2016) stated that assessing a qualitative study’s credibility and
consistency can determine its trustworthiness. Credibility refers to the legitimacy of a study’s
findings, while consistency refers to whether those findings are consistent and congruent with the
data (Merriam & Tisdell, 2016). Reflexivity (Merriam & Tisdell, 2016) and thick description
(Geertz, 1973) help to maximize credibility. Reflexivity, sometimes called the researcher’s
position (Merriam & Tisdell, 2016), refers to the researcher’s critical self-reflection on their
biases, assumptions, and relationship to the research and its participants/topic. It is important to
recognize that biases and assumptions can have a significant impact on the interpretation of
qualitative data. While researchers cannot eliminate all biases and assumptions, identifying and
explaining those biases and assumptions can mitigate data interpretation impact. Thus, readers
can better understand why the researchers arrived at their conclusions. In the second strategy,
thick description (Geertz, 1973), a researcher must write a detailed narrative, often referred to as
a vignette, which emphasizes the situation as well as the background and context of the research
topic. The approach takes into account emotions, voice, social relationships, details, feelings,
actions, and context when interpreting events, behaviors, or observations. The more details and
nuance a researcher provides, the more realistic and credible the findings will appear to the
reader.
To ensure consistency, I used the strategy of an audit trail (Merriam & Tisdell, 2016).
Merriam and Tisdell (2016) described the documentation process used to methodically track all
components of the research and provide a paper trail of the decisions made throughout the
process. An audit trail typically includes raw data like transcripts, notes, and survey results; data
analysis and reduction like coded themes and categories; data reconstruction such as findings and

64
conclusions; process notes about design choices and changes; reflexive notes about my
assumptions and insights; instrument development information; ethical considerations and
consent forms. I documented the data analysis for this study in ATLAS.ti by creating an audit
trail.
Research Ethics
Research involving human participants raises numerous ethical issues, according to
Creswell and Creswell (2017). Research in this study was subject to the approval of the
institutional review board before I could contact participants for interviews or surveys. The board
examined the purpose of the study, informed consent, interview questions, and methodology of
the study. Its purpose is to protect the participants’ privacy and to prevent them from being
exploited. I informed all participants that their participation was voluntary and that they could
opt out at any time. To ensure transparency, all participants received a final copy of the
dissertation.
Limitations and Delimitations
There are several limitations of this study, such as the small sample, selection bias,
reliance on self-reported data, researcher biases, social desirability influences, time constraints,
and context-dependence. A primary limitation of this study is its sample size and composition.
This research included 11 participants, all of whom were MHLs in various capacities and AI
knowledge. While these participants provided rich and detailed insights, the limited sample size
may not fully capture the diverse perspectives in the broader mental health leadership
community. Another significant limitation of this study stems from its geographical
concentration, which may have introduced biases that limit the generalizability of the findings.
Of the 11 participants, eight were based in California, two in the Midwest, and one in New York

65
City. This California-centric sample presented several challenges to the broader applicability of
the results. The rapidly evolving nature of AI technologies presents another limitation to this
study. The field of AI is advancing at an unprecedented pace, with new developments and
breakthroughs occurring frequently. This rapid evolution means that some of the findings may
become outdated relatively quickly. What is considered cutting-edge AI applications in mental
health care today may be superseded by new technologies in the near future.
Delimitations include the scope focusing solely on leaders, the community mental health
sector, and AI technology, as well as the use of purposeful sampling and being framed by
specific theoretical lenses like the TOE framework. Although these constraints limit
generalizability and introduce critical biases to be considered, the study remains worthwhile
because it uncovers MHLs’ perspectives regarding ethically integrating emerging AI from a
targeted leadership sample within defined parameters.

66
Chapter Four: Results or Findings
This qualitative study examined the technological, organizational, and environmental
factors that influence MHLs in the responsible adoption of AI for patient care. Using the TOE
framework and principlism ethical theory, this research aimed to develop a practical guide for
mental health organizations to implement AI responsibly while bridging the gap between
responsible AI adoption and equitable patient care.
The analysis revealed several key dimensions influencing MHLs’ approach to AI
adoption, characterized by a prevailing sentiment of cautious optimism. While leaders
recognized AI’s potential to enhance therapeutic outcomes and operational efficiency, this
optimism was tempered by complex concerns regarding data privacy, integrity, and governance
in mental health information management. The findings revealed an important balance between
AI’s potential benefits and ethical responsibilities. Leaders recognized how AI could enhance
clinical work by reducing administrative burden, expanding service capacity, improving
organizational efficiency, and increasing grant funding. However, they emphasized that pursuing
these operational advantages must not compromise their ethical duties to provide quality patient
care and maintain professional clinical judgment.
Organizational factors, including psychological safety, resource allocation, and
leadership characteristics, emerged as crucial elements for responsible AI implementation.
External forces, including governmental regulations and stakeholder interests, significantly
influenced adoption decisions beyond individual leadership control. Notably, disparities in
organizational funding raised concerns about potential treatment inequities while maintaining
patient trust and data sovereignty emerged as fundamental requirements for successful AI
integration in mental healthcare delivery. The findings from this research can guide mental

67
health organizations in developing responsible approaches to AI adoption that align with their
ethical obligations.
Participants
Eleven MHLs from community mental health or group practice settings participated in
this study (Table 3). I recruited participants from LinkedIn profiles of these organizations’
leaders and professional contacts that fit the criteria from Chapter Three. They held the eligible
leadership roles presented in Chapter Three, and I invited them to participate in the semistructured interviews. Participants’ leadership roles included frontline supervisors overseeing
employees, middle managers supervising frontline supervisors, senior executives supervising
middle managers, and thought leaders in advisory/consultative roles providing expertise on
technology, ethics, and policy. The participants had at least 1 year of leadership experience in
clinical operations or technologies. Three participants were C-level executives, four were clinical
directors, one was a clinical manager, and three were directors of clinical technologies. All work
in the mental health patient care sector, representing diverse perspectives in gender, race, and
age.

68
Table 3
Interviewee Pseudonyms and Professions
Participant Professional background
Jack Jack is a licensed mental health practitioner and is a clinical director at a
large mental health non-profit organization, where he plays a pivotal role
in overseeing the agency’s operations and service delivery. With over 13
years of experience as a licensed mental health counselor, Jack has honed
his expertise in managing diverse mental health programs and leading
multidisciplinary teams. His current responsibilities include supervising a
team of 15 direct reports, ensuring the effective delivery of clientcentered care, and maintaining high standards of clinical practice.
Samantha Samantha is a licensed mental health practitioner in California with 5 years
of clinical experience providing direct mental health services. In addition
to her work as a therapist, she is also a mental health technologist
specializing in the development and implementation of AI-driven clinical
products. Her focus is on improving service delivery and ensuring
compliance with regulatory standards, combining her expertise in mental
health with cutting-edge technology to enhance the quality and efficiency
of care.
Adam Adam is a licensed mental health practitioner, and he serves as the chief
clinical officer at a mental health NPO in California, where he leads the
clinical vision and strategy for the organization. With over 15 years of
experience as a licensed mental health therapist and a decade in
leadership roles, Adam has a deep understanding of both clinical practice
and organizational management. His leadership focuses on enhancing the
quality of care, optimizing clinical workflows, and ensuring compliance
with regulatory standards.
Solaris Solaris is a clinical psychologist and a mental health technology director in
a digital mental health delivery company who brings 35+ years of
expertise in direct mental health services, digital health solutions, and
behavioral science. She has successfully navigated the intersection of
traditional mental healthcare and technological advancement,
contributing to the transformation of mental health service delivery in the
digital age.
Jonathan Jonathan is a clinical psychologist in his state, and he serves as a C-level
executive at a major mental healthcare system, where he directs
enterprise-wide initiatives focused on technological innovation,
operational excellence, and clinical quality enhancement. His executive
leadership centers on transforming healthcare delivery through strategic
technology integration while optimizing operational efficiency and
elevating clinical outcomes.

69
Participant Professional background
Sarah Sarah, a clinical psychologist, serves as a mental health clinical director at a
prominent virtual mental health care organization. In this leadership role,
she oversees the delivery of digital mental health services, combining her
clinical expertise with innovative telehealth approaches to ensure
comprehensive client care.
John John is a licensed mental health practitioner serving as a clinical supervisor
in a major hospital system’s behavioral division. He leverages his
licensed mental health practice background to lead clinical teams and
enhance service delivery. His role bridges direct clinical oversight with
system-wide behavioral health initiatives, ensuring comprehensive
mental healthcare in an integrated hospital setting.
Jennifer Jennifer serves as a clinical technologies director, bringing together her
mental health practitioner licensure with technological expertise. She
specializes in advancing mental healthcare through digital innovation
while maintaining clinical excellence and therapeutic integrity. Her dual
expertise enables her to develop and implement technology solutions that
enhance mental health service delivery while ensuring adherence to
clinical best practices and therapeutic standards.
Benjamin Benjamin is a clinical psychologist and senior vice president at a leading
non-profit mental health organization who brings comprehensive
expertise in program delivery, clinical operations, human resources, and
research. His senior executive role encompasses strategic leadership
across clinical services, operational management, and organizational
development.
Lee Lee is a licensed mental health practitioner and currently serving as a
clinical director who brings 10 years of comprehensive experience across
the non-profit mental health sector. Drawing from experience on both
clinical and administrative sides of mental healthcare, Lee excels in
resolving complex organizational challenges while driving innovation
and sustainable growth in this transforming field.
Kacey Kacey serves as a C-level executive, bringing their clinical psychology
expertise to strategic healthcare leadership. Kacey drives organizational
excellence through strategic oversight of clinical services, policy
development, and compliance management. Their leadership focuses on
integrating evidence-based practices with robust training programs while
ensuring regulatory compliance and service quality. Drawing on their
clinical background, they effectively bridge therapeutic best practices
with organizational strategy and regulatory requirements.

70
Of the 12 participants who initially agreed to take part, one faced multiple scheduling
conflicts and chose to respond to interview questions and follow-up inquiries via email to allow
for greater flexibility. Another participant did not prefer an interview and opted to provide
answers by email, allowing additional time for thoughtful responses. The remaining 10 MHLs
completed 60-minute confidential, open-ended, semi-structured face-to-face interviews
conducted via Zoom. I analyzed these 10 in-depth interviews, conducted over 10 weeks, using
both a priori and manual posteriori coding, ensuring a comprehensive and reliable data collection
and analysis.
Findings for Research Question 1
This section presents the findings for the first research question: What are the MHLs’
perceptions of AI technologies in mental health care? Focusing on the technology aspect of the
TOE framework, the integration of AI is notably shaped by MHLs’ perceptions and attitudes
(Liehner et al., 2023). The MHLs interviewed shared their thoughts on AI technologies based on
their organizational experiences, personal research, and media reports on AI. It is important to
note that some of the respondents may have only limited exposure or basic use of these
technologies. Understanding these perceptions is essential as they influence decision-making and
guide strategies for responsible AI implementation.
The MHLs identified several key areas where AI could benefit mental healthcare
delivery. Their responses highlighted AI’s potential to enable more holistic and personalized
treatment strategies, enhance predictive analytics for preventative care, improve operational
efficiency while reducing provider burnout, increase the accessibility of mental health services,
and reduce the stigma associated with seeking mental health support. While expressing cautious
optimism about these potential benefits, leaders consistently emphasized the importance of

71
balanced implementation that preserves the human element of care. The following sections
explore these potential benefits in detail, drawing from participants’ experiences and
perspectives.
Theme 1: Perceived Potential Benefits: Cautious Optimism
The MHLs expressed a sense of cautious optimism regarding the potential benefits of AI
technologies in mental health care. Lee articulated this sentiment: “I think right now we’re
cautiously optimistic. We definitely see the potential, and it has been somewhat useful because
we haven’t seen a lot of tools yet for mental health.” While acknowledging the uncertainties and
challenges, these MHLs see significant opportunities for AI to enhance service delivery, improve
patient outcomes, and address longstanding issues in the field. Jack encapsulates this broad
vision: “I think it has a lot of potential to be really transformative for everything from maybe
examining outcomes to screening patients and getting them to the appropriate provider.”
These quotes highlight the spectrum of potential benefits that MHLs envision, from
improving diagnostic processes to enhancing treatment efficacy. The use of the word
“transformative” underscores the magnitude of change that AI could bring to the field. However,
the recurring phrase “I think” in both quotes reflects a degree of uncertainty, aligning with the
overall sense of cautious optimism (May, 2018). MHLs recognize that while AI holds promise
for addressing critical challenges in mental health care, its implementation must be approached
thoughtfully and with careful consideration of potential impacts on patients and practitioners
alike.
A key advantage of AI identified by MHLs is its potential to facilitate more holistic and
personalized treatment approaches. This potential stems from AI’s ability to process and analyze

72
vast amounts of complex data, leading to more comprehensive and nuanced understandings of
individual patients.
Solaris envisions a future where AI enables highly tailored care: “Picture personalized
treatment plans tailored to individual needs, leveraging vast datasets and machine learning to
optimize care.” This perspective highlights the potential for AI to transform treatment planning
by considering a wide range of factors that influence mental health. Artificial intelligence’s
capacity to suggest novel interventions is another aspect that excites MHLs. As John noted, “[AI]
could give you ideas of what interventions you can use that maybe you have not thought of.”
This ability to expand the repertoire of treatment options could be particularly valuable in
complex cases or when traditional approaches have been ineffective.
Kacey emphasized how AI-driven approaches could empower patients by exposing them
to diverse treatment strategies:
It also allows the clients to learn different ways of doing things rather than always
believing there’s only one way to do things. They say, “Oh, there’s multiple ways of
doing things with good outcomes. It also takes them out of their shell or stagnant way of
looking at things.”
This perspective suggests that AI could aid in broadening patients’ horizons and fostering
a more flexible, open-minded approach to mental health care. This diversity of options can be
particularly beneficial for patients who feel stuck or have not responded well to traditional
treatment methods. Moreover, the ability of AI to suggest alternative approaches can help
normalize the idea that mental health treatment is not a one-size-fits-all solution. This aligns with
contemporary views in mental health care that emphasize personalized culturally sensitive
treatment plans (Alhuwaydi, 2024).

73
Another promising aspect of AI in mental health care is its potential to bridge the gap
between physical and mental health. Adam highlighted this potential:
And people are not just somatizing their mental health conditions anymore and actually
finding the correlation. Oh, my tummy hurts. Maybe it’s something else. Let’s talk about
it. Oh, you know what, that would be cool. Maybe part of the differential diagnosis is, oh,
you have a stomachache. Have you considered depression? Wouldn’t that be cool?
This holistic approach could lead to earlier detection of mental health issues and more
comprehensive treatment strategies that consider both physical and mental well-being. The
integration of AI in this context could bridge the longstanding divide between physical and
mental health care, offering a more nuanced and complete picture of a patient’s overall health.
AI can help identify subtle connections between physical symptoms and mental health conditions
that might be overlooked in traditional diagnostic processes. By recognizing these correlations,
AI systems could prompt healthcare providers to consider mental health factors when patients
present with physical complaints and vice versa.
While enthusiasm for AI’s potential is evident, MHLs also emphasized responsible
implementation. As Jonathan noted, “I think it’s a super exciting time that we’re in. I do think we
need to do it responsibly, and I use the keyword ‘ethically,’ and that’s going to be the place
where we’re going to have to do.” This sentiment underscores the need for careful consideration
of ethical implications as AI technologies are integrated into mental health care.
The potential for AI to enhance cultural competence in mental health care is another area
of interest. Jennifer expressed optimism about AI’s role in this regard:
I think it provides such a great way of … moving on into the future [of] mental health and
… care itself. … Me myself looking at it from that … stance of incorporating into my

74
business and looking at how AI through various algorithms through cultural competence
can be supportive of individuals as [they] are navigating through the symptoms.
This perspective highlights the potential for AI to help address cultural disparities in mental
health care, a longstanding challenge in the field.
While MHLs expressed excitement about AI’s potential to revolutionize mental health
care through personalized, holistic, and culturally competent approaches, they also emphasized
the need for responsible and ethical implementation. This balanced view reflects a cautious
optimism that acknowledges both the transformative potential of AI and the importance of
careful, considered adoption in the sensitive field of mental health care.
A notable benefit of AI identified by MHLs is its capacity for predictive analytics and its
potential to enhance preventative care. The ability of AI to analyze vast amounts of patient data
to identify trends and predict future mental health needs is seen as a game-changer in the field.
Solaris articulated this potential: “Imagine early detection of depression or anxiety, even before
symptoms fully manifest, through sophisticated algorithms analyzing speech patterns and social
media activity.”
This quote highlights the proactive nature of AI-driven predictive analytics, suggesting
that it could revolutionize early intervention strategies in mental health care. By identifying
potential issues before they fully develop, clinicians could implement preventative measures,
potentially averting mental health crises. Solaris further emphasized the forward-looking nature
of AI in mental health care: “Looking ahead, I envision AI predicting mental health crises before
they occur, enabling timely interventions.” This perspective highlights the potential for AI to
shift mental health care from a reactive to a proactive paradigm, potentially reducing the severity
and frequency of mental health crises. This perspective highlights the potential for AI to shift the

75
mental health care paradigm from reactive to proactive, potentially reducing the severity and
frequency of mental health crises as AI’s predictive capabilities can identify warning signs and
risk factors much earlier, sometimes even before the patient and the provider can (J. Wang,
2023). By analyzing patterns in behavior, speech, social media activity, and other data points, AI
could flag potential issues, allowing for early intervention and, therefore, can lead to less
intensive treatments and overall better outcomes for patients (Atlam et al., 2022). Moreover, by
preventing crises before they occur, this approach could significantly reduce the emotional toll
on patients and their families, as well as the logistical and financial burdens on healthcare
systems (Ejjami, 2024).
In essence, this shift from reactive to proactive care could revolutionize mental health
treatment, making it more effective, efficient, and humane. Jack echoed this sentiment,
highlighting the broad applicability of AI in mental health care:
I think it could be utilized in a way to catch other forms of need earlier. Maybe it’s just
given a way to review sessions or review content that is screened from like an intake and
a way to find are there other applicable services that might be needed just as a way to
further analyze potential patients and how to surround them with support even beyond
mental health care.
This quote suggests that AI’s predictive capabilities could help in early detection and in
optimizing patient care pathways, ensuring individuals receive the most appropriate
interventions. Predictive analytics could lead to more targeted and effective treatment strategies,
potentially improving patient outcomes and reducing the trial-and-error approach often necessary
in mental health treatment.

76
It is important to note that while MHLs were enthusiastic about AI’s potential in
predictive analytics and preventative care, they also emphasized the need for human oversight.
As Solaris stated, “Offering nuanced clinical judgment extends to our exploration of AI’s
potential in other areas, like personalized treatment plans and predictive analytics.” This
perspective reflects a balanced approach, where AI is seen as a powerful tool to augment clinical
decision-making rather than replace it. The consensus among MHLs is that AI should be used for
tasks such as pattern recognition, predictive analysis, and offering treatment planning
suggestions but not for making final diagnostic decisions. As Benjamin noted,
I always feel like the clinicians should be the ones in the driver’s seat. They’re the ones
that are essentially making the call. They’re the ones that have been trained. AI is really
there to support as a co-pilot rather than as the person that is driving the decisions.
This cautious approach ensures that AI serves as a supportive aid to clinicians, enhancing their
ability to provide preventative care and personalized treatment while maintaining the crucial
element of human judgment in mental health care.
The interviewees consistently highlight the potential of AI to significantly enhance
efficiency in mental health settings, particularly by reducing the administrative burden on
professionals. This improved efficiency is seen as a key factor in preventing burnout and
increasing job satisfaction among mental health staff.
The administrative workload in mental health care has grown substantially, largely due to
requirements set by funding sources and payors who mandate detailed documentation to justify
medical necessity and meet billing criteria. MHLs see AI as a promising solution to streamline
these processes. As one participant succinctly put it, “It [AI] will free up a lot of time for the
provider and the therapist.” John elaborated on this potential, envisioning AI’s role in

77
documentation: “I think in a positive manner the way that I envision us utilizing AI is to really
help with the transcribing and note taking for all the notes that we do for patients.” This
reduction in administrative tasks is not just about saving time; it is about reallocating that time to
more meaningful aspects of care. Benjamin emphasized this point:
I definitely think that there would be some usefulness, lots of promising usefulness in
terms of efficiency in the work that we do. Given that in mental health, if we are
providers, we have a limitation in regards to the amount of time that we get to spend with
our clients.
Adam further illustrated the potential for AI to handle repetitive tasks:
I can see some benefits to it especially with the non-mental health care aspect of it. So,
like the funding that we talked about, I think that would be something. Like disciplinary
action you write the same thing over and over again. … So, I can see it opening up a lot
of my time on the administrative tasks and really focusing on the other important things.
This efficiency gain is closely tied to the potential for reducing burnout among mental
health professionals. Burnout, characterized by emotional exhaustion, detachment, and reduced
personal accomplishment, has become a significant concern in the healthcare industry. By
addressing this issue, MHLs hope to create a more positive and fulfilling work environment.
Jonathan articulated this comprehensive vision: “I just think easier use of navigating healthcare
services, reducing administrative barriers for providers, increasing Joy of work and reducing
burnout, and improving patient experience.” This quote encapsulates the multifaceted benefits
that MHLs anticipate from AI adoption, linking improved efficiency to both provider well-being
and patient care quality.

78
The desire to refocus on patient care is a recurring theme. Benjamin expressed, “I do
want to spend more time with my clients. I do want to spend more time doing more research on
types of tools and techniques that I can help my client.” Samantha further emphasized how AI
could alleviate the cognitive load on clinicians: “I think that AI does a lot of great things for
taking away some of the mental burden on clinicians. That I think is- is a potential like really
great efficiency, scalability getting more access—access.” Kacey provides a broader perspective
on this potential transformation:
I often see that being able to integrate AI will hopefully allow the therapists to be able to
spend more time directly with the clients rather than on repetitive documentation. And a
lot of time on the administrative tasks that currently also require—although as important
they also require a lot of time.
In essence, MHLs view AI as a tool that could significantly reduce the administrative
burden on mental health professionals, allowing them to focus more on patient care and engage
in professional development. This shift is expected to improve the quality of care, enhance job
satisfaction, and reduce burnout among providers. By improving efficiency and job satisfaction,
AI could contribute to a more sustainable and effective mental health care system, creating a
positive impact on both providers and patients. The enthusiasm for these potential benefits is
tempered with a recognition of the need for responsible implementation, reflecting the cautious
optimism that characterizes MHLs’ overall perspective on AI in mental health care.
The respondents recognize the significant potential of AI integration to enhance the
accessibility of mental health services. This benefit is appealing as it addresses the dual
challenges of controlling healthcare expenses while expanding the reach of mental health

79
resources. As one participant succinctly noted, “[AI] helps make mental health resources more
accessible.”
By leveraging AI-driven tools, organizations can provide a basic level of treatment to a
broader population, effectively democratizing access to mental health support. These AI
solutions, such as remote chat services and virtual triage systems, are breaking down traditional
geographical barriers that have long impeded access to care. Solaris elaborated on this potential:
“AI-powered chatbots and virtual therapists provide immediate support 24/7, bridging gaps in
access and reducing the stigma surrounding mental health care.”
This quote highlights how AI can offer round-the-clock support, providing initial
assessments, basic coping strategies, and guidance to appropriate resources, all without the need
for in-person visits. A key focus for MHLs is the potential of AI to expand care to marginalized
populations in remote areas who typically lack the resources to access traditional mental health
services. Jonathan emphasized this point: “I actually … think AI is a solution to inequitable or
disparate health outcomes. And so, … that’s why … I’m a big sort of advocate for it.” This
perspective underscores how AI could play a crucial role in addressing longstanding issues of
health equity in mental health care delivery.
The potential cost-effectiveness of AI solutions is another factor that MHLs see as key to
enhancing accessibility. Adam articulated,
I’m thinking cost-effectiveness. Therapy is expensive, very expensive, and not affordable
for everybody. So, if we had [an] AI platform, maybe that’s something that brings
therapy to the general population, which we know at this time may not be accessible for
everybody.

80
By offering virtual support options, AI can help overcome the logistical and financial
barriers often preventing individuals in remote or underserved areas from seeking help. This
expansion of care improves individual outcomes and could address broader issues of health
equity and social justice in mental health care delivery.
MHLs also recognized the potential of AI to provide support in situations where humanto-human interaction might be challenging. Adam provided an insightful example:
That and then, and I can see for some of the disorders that we see some of the conditions
that we deal with where it might be difficult for a person to engage with another human
being, autism being one of them. I can see how AI could be a great platform to engage
those populations and help them with whatever it is that they’re presenting.
This perspective highlights how AI could potentially bridge gaps in care for populations
with specific needs that might not be easily met through traditional care models. In essence,
MHLs see AI as a powerful tool for enhancing the accessibility of mental health services,
particularly for underserved populations. However, it is important to note that while MHLs are
enthusiastic about these possibilities, they also maintain a balanced view, recognizing the need
for careful implementation to ensure the quality and appropriateness of AI-delivered care.
The participants recognize the potential of AI to significantly reduce the societal stigma
associated with mental health issues. This reduction in stigma is seen as a crucial step toward
improving overall mental health outcomes and encouraging more individuals to seek help when
needed.
A primary way AI can contribute to reducing stigma is by providing anonymous
platforms for individuals to explore their mental health concerns. As Mike pointed out, “AI
chatbots can educate users about mental health issues without judgment or fear of exposure.”

81
This anonymity can be particularly empowering for people who may be hesitant to seek help due
to fear of stigma or discrimination. It provides a safe space for individuals to ask questions, seek
guidance, and learn about mental health conditions without the pressure of societal expectations.
Artificial intelligence’s potential to demystify mental health through education is another
crucial aspect of stigma reduction. By providing consistent, reliable, and scientifically backed
information, AI tools can help normalize conversations around mental health. Adam highlighted
how this can lead to a more holistic understanding of health:
And people are not just somatizing their mental health conditions anymore and actually
finding the correlation. Oh, my tummy hurts. Maybe it’s something else. Let’s talk about
it. Oh, you know what … would be cool? Maybe part of the differential diagnosis is, oh,
you have a stomachache. Have you considered depression? Wouldn’t that be cool?
This perspective illustrates how AI can help individuals make connections between physical and
mental health, potentially reducing the stigma associated with mental health conditions by
framing them as part of overall health.
The potential for AI to encourage proactive mental health management is another way it
can contribute to stigma reduction. By empowering individuals to take control of their mental
health, AI can help shift the narrative from mental health being a taboo topic to one that is an
integral part of overall well-being. Kacey’s comment reflects this potential:
It also allows the clients to learn different ways of doing things rather than always
believing there’s only one way to do things. They say, “Oh, there’s multiple ways of
doing things with good outcomes.” It also takes them out of their shell or stagnant way of
looking at things.

82
This quote suggests that AI can help broaden perspectives on mental health treatment, potentially
reducing the stigma associated with seeking help or trying different approaches to mental
wellness.
Furthermore, the increased accessibility of mental health resources through AI can itself
contribute to stigma reduction. As mental health support becomes more readily available and
integrated into everyday life, it may become more normalized and accepted. Solaris’s vision of
AI’s potential illustrates this: “Picture personalized treatment plans tailored to individual needs,
leveraging vast datasets and machine learning to optimize care.” By making mental health care
more personalized and accessible, AI could help shift societal perceptions, making mental health
support as commonplace and accepted as physical health care.
The participants see AI as a powerful tool for reducing the stigma surrounding mental
health. Through anonymous support, education, encouragement of proactive mental health
management, and increased accessibility, AI has the potential to normalize mental health
discussions and support-seeking behaviors. However, it is important to note that while AI offers
these possibilities, MHLs also recognize the need for careful implementation to ensure that AIdriven approaches complement rather than replace human-led efforts in stigma reduction and
mental health support.
Theme 2: Perceived Technology Limitations and Concerns
The interviewees identified several critical concerns and limitations regarding AI
implementation in mental healthcare settings. Their responses highlighted issues surrounding
AI’s accuracy and reliability in clinical settings, concerns about bias and fairness in AI systems,
challenges with cultural competency, questions about privacy and data security, the inherent
complexity of mental health work, potential impacts on human interaction and empathy in

83
therapeutic relationships, and compatibility challenges with existing healthcare systems. While
acknowledging AI’s potential benefits, leaders emphasized these limitations as crucial
considerations for responsible implementation. The following sections explore these concerns in
detail, drawing from participants’ experiences and perspectives to understand the challenges that
must be addressed for successful AI integration in mental healthcare.
While acknowledging the potential benefits, the participants shared the consensus that AI
should be utilized primarily for pattern analysis, predictive analytics, and treatment planning
suggestions rather than as an autonomous diagnostic tool. This perspective is rooted in the
understanding that mental health care often requires nuanced interpretation and contextual
understanding that current AI systems may not fully capture. Lee articulated this cautious
optimism: “I think right now we’re cautiously optimistic would be the term. We definitely see
the potential, and it has been somewhat useful because we haven’t seen a lot of tools yet for
mental health.”
The concerns raised by MHLs regarding AI’s accuracy and reliability in mental health
care reflect a broader set of challenges in implementing these technologies. While AI shows
promise in enhancing efficiency and providing insights, its limitations in understanding the
nuanced, context-dependent nature of mental health issues are significant. The potential for overpathologizing normal human experiences, as highlighted by Adam, and the risk of rigid,
oversimplified treatment protocols, as Jonathan noted, underscore the necessity of human
oversight in AI-assisted mental health care. These concerns touch upon questions of bias,
cultural competency, and the fundamental complexity of mental health work—areas that require
careful consideration regarding AI integration. Delving deeper into these interconnected
challenges reveals that addressing the limitations in AI’s accuracy and reliability is just the

84
beginning of a more comprehensive evaluation of AI’s role in mental health care. This leads us
to explore another critical aspect of AI implementation: the potential for bias and the pressing
need for fairness in these systems.
As AI technologies evolve and integrate into mental health care, MHLs express a
nuanced understanding of their capabilities and limitations. Currently, there is a lack of
confidence in AI’s reliability as a standalone diagnostic tool, given the complexities of mental
health diagnoses with their myriad nuances and contextual factors. Jonathan highlighted a key
concern regarding the potential rigidity of AI-driven approaches: “The concern is that AI might
lead to rigid treatment protocols, like starting with A for depression and moving to B if not
effective.”
This quote underscores the fear that AI might oversimplify complex mental health issues,
potentially leading to inflexible treatment approaches that fail to account for individual patient
needs and circumstances. As Adam elaborated on the risk of over-pathologizing normal human
experiences, “Diagnosis could be over-pathologized by AI, misinterpreting normal life phases as
conditions. AI may struggle to differentiate between pathological conditions and transient issues
with protective factors.”
This perspective highlights the concern that AI might lack the nuanced understanding
required to distinguish between normal life challenges and clinical mental health conditions.
Adam further emphasized this point: “I do think that could be a potential downside, the overstigmatizing and over-pathologizing.” Benjamin underscored the importance of human judgment
in mental health care: “Oftentimes, we trust these things because they seem so powerful and
intelligent, but the truth is just because you’re intelligent doesn’t mean you’re reliable or
accurate sometimes.” This quote reinforces the belief among MHLs that human expertise

85
remains crucial in interpreting and applying AI-generated insights, particularly given the
complexities of mental health.
Sarah raised concerns about the potential limitations of AI in providing comprehensive
oversight:
I’m thinking, yes, you may be able to increase access. You may be able to provide more
services, more access to patients needing mental health care. But there is also additionally
more liability because … you’re not able to really provide 100% oversight … if you were
to be the one delivering the service.
This perspective highlights the potential risks associated with relying too heavily on AI systems,
particularly in terms of liability and providing comprehensive care.
While MHLs recognize AI’s potential to enhance efficiency and provide insights into
mental health care, they emphasize the irreplaceable value of human expertise and interpersonal
skills in providing comprehensive, nuanced care. The consensus is that AI should be viewed as a
tool to augment human capabilities rather than replace them, particularly in the complex and
highly individualized field of mental health care.
The MHLs expressed significant concerns about AI’s ability to produce accurate and
unbiased evaluations in mental health care. A primary worry is the quality and diversity of data
used to train AI systems, as biases in this data can substantially influence system performance
and perpetuate imbalances in mental health treatment, particularly for marginalized populations.
Samantha captured this concern: “There’s like cultural biases; it’s still trained on a model that is
predominantly White.” This statement highlights a fundamental issue in AI development for
mental health care: the lack of diversity in training data, as most AI models are trained on data
from predominantly WEIRD populations, which may not reflect the experiences and needs of

86
individuals from other backgrounds (Obermeyer et al., 2019). Samantha further elaborated on the
implications of this bias: “And so, it misunderstands cultural nuances and all of that stuff, which
… it’s just where we’re at right now, and I think people don’t necessarily understand the
intricacies of that.” This underscores the complexity of the problem and the potential for AI
systems to misinterpret or overlook important cultural nuances in mental health care. Jennifer
echoed these concerns, emphasizing the importance of cultural competence in AI development:
The potential downsides, I think, one would … be cultural competence with AI that …
those who may create the algorithms with AI may not be looking from that lens of being
able to support individuals from a variety of either ethnicities or cultural relevance, which
could come back and … not be as supportive to an individual.
This perspective highlights the risk that AI systems, if not carefully developed with
diverse cultural perspectives in mind, could fail to provide adequate support for individuals from
various ethnic and cultural backgrounds. Despite these concerns, there is also recognition of the
potential for improvement. Kacey noted, “We … actually have a lot more to learn around how to
do that, but the potential of it, I think. is really exciting both from a provider perspective as well
as a patient perspective and just community.” This statement reflects a cautious optimism,
acknowledging the current limitations while also recognizing the potential benefits of addressing
these biases and improving AI systems for mental health care.
The MHLs’ statements highlight the critical need for diverse datasets, culturally informed
AI development, and ongoing evaluation of AI systems for bias. They emphasize that without
careful attention to these issues, AI in mental health care risks perpetuating or even exacerbating
disparities in treatment and outcomes for marginalized populations. Addressing these concerns is

87
seen as crucial for ensuring that AI can be a tool for improving mental health care equitably
across diverse populations.
Cultural competence has emerged as a critical focus in developing AI products for mental
health care (McGregor et al., 2019). This emphasis reflects a growing recognition that effective
mental health support must be sensitive to and inclusive of diverse cultural backgrounds. One
participant articulated this priority:
Culture! Making sure that we’re culturally competent is paramount to me. That we really
take a hard look at the individuals that we are supporting. And I desire for individuals to
have just that level of comfort with whatever they are as they utilize our product.”
This statement underscores the importance of creating AI tools that are not only technologically
advanced but also culturally attuned to users’ diverse needs. It highlights the critical need for AI
systems in mental health care to go beyond mere functionality and incorporate a deep
understanding of cultural nuances and sensitivities. This perspective recognizes that effective
mental health care is inherently tied to cultural context and that AI tools must be designed with
this fundamental principle in mind.
Adam further emphasized the multifaceted nature of cultural competence in mental health
care: “So, culturally responsive, culturally informed, trauma-informed care which takes human
factors into consideration.” This quote expands on the concept of cultural competence, linking it
explicitly to trauma-informed approaches and human-centered design. It suggests that truly
effective AI in mental health care must be culturally responsive, understand the impact of trauma
across different cultural contexts, and consider the complex human factors involved in mental
health treatment. This comprehensive approach to cultural competence in AI design reflects a

88
growing awareness among MHLs of the interconnected nature of culture, trauma, and mental
health.
Cultural competence in AI goes beyond mere representation; it involves a deep
understanding of cultural contexts, including trauma-informed approaches. It emphasizes that
truly culturally competent AI systems in mental health care must be designed with a nuanced
comprehension of diverse cultural norms, beliefs, and experiences that shape an individual’s
mental health and approach to seeking help. Moreover, this view suggests that AI developers and
implementers need to incorporate trauma-informed principles, recognizing that many mental
health issues are rooted in or exacerbated by cultural and historical traumas and that effective AI
tools must be sensitive to these complex, culturally specific experiences. However, MHLs
express concerns about the current state of cultural competence in AI development. Jennifer
articulated this worry:
The potential downsides, I think, one would … be cultural competence with AI that …
those who may create the algorithms with AI may not be looking from that lens of being
able to support individuals from a variety of either ethnicities or cultural re-relevance,
which could come back and … not be as supportive to an individual.
This quote highlights the risk of AI systems failing to adequately support individuals from
diverse cultural backgrounds due to limitations in the developers’ perspectives. It underscores
the concern that AI algorithms, if developed without sufficient cultural diversity in the team or
training data, may inadvertently perpetuate biases or overlook crucial cultural nuances in mental
health care. Furthermore, this perspective emphasizes the potential for AI systems to provide
suboptimal or even inappropriate care for individuals from minority cultures or ethnicities,
potentially exacerbating disparities in mental health treatment and outcomes.

89
The concern about cultural competence in AI extends to the system’s ability to provide
diverse and culturally appropriate responses without explicit prompting. MHLs worry that
without built-in cultural sensitivity, AI systems might default to a narrow range of culturally
homogeneous outputs, potentially alienating users from diverse backgrounds or failing to meet
their specific needs. Kacey provides a concrete example of how a lack of cultural competence
can manifest in AI systems:
Unless I give it some … cues that I …desire a cultural background, it’s only going to spit
out a couple different genres of … content. And I didn’t think that that from… a cultural
and ethical standpoint might not be the best way of displaying the utilization of AI.
This observation underscores the need for AI systems to be inherently culturally aware rather
than requiring explicit prompts to consider cultural factors. It highlights a critical shortcoming in
current AI implementations, where cultural competence is often treated as an add-on feature
rather than a fundamental aspect of the system’s design and functionality. Furthermore, this
perspective suggests that truly effective AI in mental health care should have cultural awareness
deeply integrated into its core algorithms and knowledge base, enabling it to automatically
provide culturally appropriate and sensitive responses across a wide range of diverse user
interactions without the need for specific cultural cues or prompts.
While MHLs recognize the potential of AI in mental health care, they stress the critical
importance of cultural competence in its development and implementation. They advocate for AI
systems that are programmed and trained on diverse datasets, avoid stereotypes, and recognize
different cultural beliefs and practices. The goal is to create tools that are truly effective and
respectful for all users, regardless of their cultural background, ultimately enhancing mental
health outcomes across diverse populations.

90
The participants expressed significant concerns regarding data privacy and the
transparency of AI systems in mental health care. A primary focus is on the importance of
informed consent and transparency in the use of AI. Sarah highlighted this fundamental aspect:
For me, I think it comes back to informed consent, so it … needs to be all parties need to
be aware of that it is being utilized, it’s being implemented, and then there is consent
from all parties that they’re okay with it being part of it and again stressing that it may
not be part of the service the interventions being provided, but it’s still part of the service
because it is used to … document the session.
This statement underscores the need for full disclosure and patient agreement in the use of AI in
mental health care. It emphasizes that even when AI is used for administrative purposes like
documentation, patients should be informed and consent obtained. This approach ensures
transparency and maintains trust in the therapeutic relationship.
Beyond consent, MHLs are also concerned about the overall safety and security of patient
data. The responsibility for ensuring data privacy and security in AI-assisted mental health care
is a significant concern for MHLs. Benjamin articulated this sentiment: “We are responsible for
the care we provide, so I really have to be reassured in regards to their concerns about privacy
and safety.” This statement highlights the sense of responsibility that mental health providers feel
toward their patients’ data security. It underscores the need for robust safeguards and assurances
in AI systems to protect sensitive mental health information. This responsibility extends beyond
just the implementation of AI to the ongoing management and security of patient data. The
complexity of these concerns is further elaborated by other MHLs.
Sarah expressed the mixed feelings among the mental health community regarding AI
adoption:

91
I know that … there’s definitely some interest. There’s some interest, or maybe curiosity
is a better word. Curiosity about it. But at the same time, there’s still a lot of …
cautiousness and … worries, especially around the breach.
This quote reflects the tension between the potential benefits of AI and the significant concerns
about data security. It highlights that while there’s curiosity and interest in AI’s capabilities,
there’s also a strong undercurrent of caution, particularly regarding the potential for data
breaches. This cautious approach underscores the need for robust security measures in AI
systems. The specific nature of these security concerns is further detailed by other MHLs.
Jack elaborated on the complex issues surrounding data storage and protection in AIassisted mental health care:
I think there’s going to be a variety of potential shortcomings or issues that could arise. I
think one is probably security. Depending on how that AI is created or hosted, where is
that data going? Is it held within the organization itself, or is it the AI system, or is it like
the being that’s owned by somebody else and all the data from the therapy center or
therapy organization going back and forth between this other thing? And if so, how do we
keep that information secure? How do we make sure that information is not being used
for other purposes?
This statement highlights the multifaceted nature of data security concerns in AI-assisted mental
health care. It raises critical questions about data ownership, storage locations, transfer methods,
and access controls. These concerns extend beyond traditional patient confidentiality issues,
encompassing the complexities introduced by AI systems that may be owned or operated by third
parties. Addressing these concerns requires a comprehensive approach to data governance and

92
security. The responsibility for implementing these measures is shared among various
stakeholders.
Jennifer emphasized the shared responsibility between healthcare providers and AI
developers in ensuring the beneficial and secure use of AI:
And so, the responsibility obviously in the role of clinic and providing services … is
there and then also the healthcare company being able to say, “Hey, this is what we are
utilizing in a way that could be of benefit to our … client base and so let’s see how we
can be better supportive in this way.”
This perspective underscores the collaborative nature of implementing AI in mental health care.
It suggests that both healthcare providers and AI developers have roles to play in ensuring that
AI technologies are used responsibly and effectively. This shared responsibility model
emphasizes the need for ongoing dialogue and cooperation between clinical staff and technology
providers.
However, even with such collaboration, MHLs recognize that AI systems are not
infallible. Benjamin raised an additional concern about the potential for errors in AI systems:
“Another potential could be just negligence. AI is not perfect at this point, from what I
understand.” This statement highlights the recognition that AI systems, like any tool, are not
infallible and may be prone to errors or oversights. It underscores the need for ongoing human
oversight and the importance of viewing AI as a supportive tool rather than a replacement for
human judgment in mental health care. This perspective reinforces the need for clear guidelines
and regulatory frameworks to govern the use of AI in mental health settings.
The concerns raised by MHLs regarding privacy and transparency in AI-assisted mental
health care are multifaceted and profound. They encompass issues of informed consent, data

93
security, shared responsibility, and the potential for errors in AI systems. They stressed the
importance of maintaining the central role of human expertise and professional ethics in patient
care, even as AI technologies are adopted. These leaders emphasize the need for robust
safeguards, clear ethical guidelines, and ongoing dialogue between all stakeholders, including
mental health professionals, AI developers, and policymakers. The consensus is that while AI
holds significant potential to enhance mental health care, its integration must be carried out with
the utmost attention to patient privacy, data security, and transparency. As the field of mental
health care continues to evolve with technological advancements, addressing these privacy and
transparency concerns will be crucial in shaping responsible and effective AI integration
practices.
The complexity of mental health work presents significant challenges in developing
effective AI solutions, highlighting the need for more sophisticated and nuanced tools. Unlike
many medical conditions that can be diagnosed and treated through straightforward algorithms,
mental health issues often involve intricate, interrelated factors that defy simple categorization or
solution (Joseph & Babu, 2024). As Lee aptly illustrated this complexity with an analogy, “The
line of work we do is not always very simple and that it’s 1 plus 1 is 2. It’s like 1 plus 1 is 2
minus 3 divided by 5.”
This vivid description underscores the non-linear, often unpredictable nature of mental
health treatment, where multiple factors interact in complex ways, making it challenging to
create AI tools that can adequately capture and respond to the nuances of mental health care. As
Benjamin further elaborated on the intricate process involved in mental health care:
Oftentimes, processes happen while, as a therapist, after session, we tend to think and
really look at the bigger picture of their current clients’ presentation. And there’s a lot of

94
processes and conceptualization that goes into that. And we apply them as we experience
with our clients.
This statement underscores the fluid and iterative nature of mental health treatment,
where therapists continuously reassess and adapt their approach based on new insights gained
from each client interaction. It emphasizes that mental health care is not a static, one-size-fits-all
process but rather a dynamic journey that requires constant refinement and personalization. This
ongoing adaptation, deeply rooted in the therapist’s evolving understanding of the client’s
circumstances, presents a significant challenge for AI systems, which typically operate on more
rigid, predefined algorithms and may struggle to replicate the nuanced, intuitive adjustments that
human therapists make throughout the course of treatment.
Furthermore, Solaris emphasized the crucial role of human judgment in navigating these
complexities, “Mental health care often involves understanding intricate personal histories and
cultural contexts, nuances that AI cannot fully grasp. Human judgment remains crucial in these
situations.” This perspective underscores the limitations of AI in fully comprehending the depth
and breadth of human experiences that inform mental health issues. While AI can process vast
amounts of data and identify patterns, it currently lacks the capacity to truly understand the
nuanced interplay of human emotions, personal histories, cultural contexts, and individual
experiences that shape a person’s mental health (Joseph & Babu, 2024; Mautang & Suarjana,
2023). Human clinicians, drawing on their empathy, cultural knowledge, and ability to interpret
subtle cues, are uniquely equipped to navigate these complex, often intangible factors that are
crucial in mental health care but remain challenging for AI to fully grasp and incorporate into its
decision-making processes.

95
While AI holds promise for enhancing mental health care, the inherent complexity of the
field poses significant challenges to its implementation. The non-linear nature of mental health
challenges, the intricate human connections involved in mental health care, the critical role of
ongoing clinical judgment, and the necessity for personalized, context-aware approaches all
emphasize the limitations of current AI capabilities in this domain (Joseph & Babu, 2024).
The integration of AI in mental health care has sparked a robust debate among MHLs
regarding its potential impact on patient-provider relationships. At the core of this discussion is
the fundamental nature of mental health care, which has traditionally relied on human connection
and empathy (Furnham & Sjokvist, 2017). Adam captured this sentiment:
So, I know for the longest time, our field has emphasized, and it still holds true, the
connection between a human to a human, which basically is the basis for therapy. The
therapeutic alliance, the rapport building, the trust. All of those factors which I don’t
think you can do. You can have that with an AI, even if it’s the most sophisticated form.
So, that human-to-human direction, the what happens between two individuals. The
actual physical and chemical changes that take place, which there’s so much research
behind that, I don’t think an AI could duplicate that no matter what.
Adam’s statement underscores the irreplaceable nature of human interaction in mental health
care, highlighting elements like rapport building and trust that are crucial to effective therapy. It
emphasizes that the therapeutic alliance, built on empathy and personal connection, is a
fundamental aspect of mental health treatment that current AI technologies cannot fully replicate.
This perspective raises important questions about the limits of AI in replicating the nuanced
interpersonal dynamics essential to mental health care.

96
Kacey further emphasized the complexity and artistry involved in mental health care,
which poses significant challenges for AI implementation: “Mental healthcare, there … is data
that … informs what we do in information but that it’s really sort of an art and that there are
nuances in terms of what individual patients’ needs.” This observation highlights the intricate
nature of mental health treatment, where data and information play a role but are complemented
by the nuanced understanding and intuition of human practitioners. It suggests that while AI may
excel at processing data, it may struggle to replicate the artistry and individualized approach that
human clinicians bring to mental health care. This nuanced perspective leads us to consider how
AI might be integrated effectively without compromising the essential human elements of care.
Benjamin offered a balanced view on how AI might be incorporated into mental health
care:
I always feel like the clinicians should be the ones in the driver’s seat. They’re the ones
that are essentially making the call. They’re the ones that have been trained. AI is really
there to support as a co-pilot rather than as the person that is driving the decisions.
Benjamin’s perspective suggests a model where AI complements rather than replaces human
expertise, potentially enhancing the quality of care without compromising the essential human
elements. In other words, Benjamin envisions AI as a supportive tool that can augment the
capabilities of human clinicians, handling tasks such as data analysis or administrative work
while leaving critical decision-making and therapeutic interactions to trained professionals.
Despite the concerns, some MHLs see potential benefits in AI integration. Solaris noted a
positive aspect of AI implementation: “AI improves the quality of interactions that can
significantly boost acceptance levels amongst patients.” This statement suggests that when
implemented thoughtfully, AI could enhance the therapeutic relationship rather than diminish it.

97
It indicates that AI tools could support clinicians in providing more targeted and effective care,
ultimately improving patient outcomes and satisfaction.
In conclusion, while the participants expressed valid concerns about the potential
depersonalization of care through AI, they also recognized its potential benefits. The focus
remains on maintaining the personal, empathetic care that is central to effective mental health
treatment while leveraging AI to improve efficiency and expand access to services. Moving
forward, the challenge lies in finding ways to integrate AI that enhance rather than replace the
human elements of care. This delicate balance leads us to consider the next crucial aspect of AI
integration in mental health: its compatibility with existing systems and workflows.
Integrating AI technologies into EHR systems and clinical workflows presents a
significant challenge in adopting AI in mental health care. This integration is crucial for ensuring
seamless adoption and maximizing the benefits of AI without disrupting established practices.
Samantha highlighted the complexity of this process: “Integration teams work on incorporating
AI into EHR and ensuring it fits organizational workflows.” This statement underscores the need
for specialized teams dedicated to tailoring AI solutions to fit each organization’s technological
ecosystem. It emphasizes that integrating AI into mental health care is not a one-size-fits-all
process but rather requires a customized approach that considers individual institutions’ systems,
workflows, and needs. The complexity of this integration process necessitates a team with
diverse expertise, including knowledge of AI technologies, EHR systems, and the specific
operational nuances of mental health care settings.
The multifaceted nature of AI integration extends beyond mere technical compatibility
(Ahmed et al., 2023). Kacey elaborated on the various aspects involved in the implementation
process: “Then, you have a separate set of people who are doing the actual integration, like

98
integrating into your EHR figuring out the whole implementation side of things, making sure that
this works for your organization based on your workflows.”
Kacey’s perspective emphasizes that successful AI integration involves both
technological compatibility and alignment with organizational processes and workflows. It
highlights the multidimensional nature of implementing AI in mental health care settings, where
technical integration is only one piece of a larger puzzle. Kacey’s statement further underscores
the comprehensive approach ensures that AI solutions are not only technically sound but also
practically viable within the specific context of each organization. Beyond technical and
operational considerations, regulatory compliance remains a critical aspect of AI integration in
healthcare (Sharma, 2020). Adam raises important questions regarding this crucial dimension,
“How does this align with regulations laid down by the state, laid down by the county? How
does it ensure compliance? How does it ensure confidentiality here with all the HIPAA laws and
so on and so forth?”
Adam’s statement highlights the need for AI solutions to integrate technically and
comply with healthcare’s legal and ethical standards. It underscores the importance of
considering regulatory requirements throughout the integration process, ensuring that AI
implementations adhere to privacy laws, data protection regulations, and other relevant
healthcare standards. This regulatory alignment is crucial for maintaining trust in AI-enhanced
mental health care systems and protecting patient rights (Al-Abdullah et al., 2020).
While the integration process involves multiple stakeholders and considerations,
healthcare organizations retain significant agency in the adoption of AI technologies. Lee
pointed out the role of organizational decision-making in this process:

99
We, as an organization, actually have some authority to say we don’t want it. So,
sometimes, they may push a new software update. Generally speaking, we would accept
it, review it, and see how it impacts what our current flow is and if it does what we need
to train staff on.
Lee’s statement emphasizes the importance of organizational autonomy in the adoption
and integration of AI technologies. It underscores that healthcare organizations are not passive
recipients of AI technology but active participants in its implementation, with the authority to
evaluate and potentially reject AI solutions that do not align with their needs or standards.
Furthermore, it highlights the crucial role of staff training in successful AI integration,
suggesting that even technologically sound AI solutions may fail if the workforce is not
adequately prepared to use them effectively in their daily practice.
The successful integration of AI into existing systems in mental health care requires a
multifaceted approach that addresses technical, operational, regulatory, and organizational
considerations. It demands collaboration between AI developers, IT specialists, mental health
professionals, and organizational leadership to ensure that AI technologies not only function
within existing systems but also enhance and streamline clinical practices without compromising
the quality of care or data integrity. This complex integration process sets the stage for the next
exploration: the organizational factors that influence MHLs in adopting AI technologies
responsibly (Ahmed et al., 2023; Lu et al., 2023; Mendes-Santos et al., 2022).
Results for Research Question 2
This section presents the interview findings related to the second research question: What
organizational factors enable MHLs to adopt AI to improve patient care while minimizing risks?
Focusing on the organization component of the TOE framework, this analysis explores the

100
internal characteristics, structures, and processes at mental health organizations that influence
their ability to implement AI responsibly. The findings shed light on various factors, including
leadership approaches, organizational culture, resource allocation, staff training, and internal
policies that contribute to AI’s successful and ethical integration in mental health care.
Examining these organizational elements reveals insights into how mental health institutions can
create an environment conducive to AI adoption that maximizes patient benefits while carefully
managing potential risks, ultimately helping MHLs navigate the complex landscape of AI
adoption in mental health care.
Theme 1: Organizational Culture and Resources for Ethical Alignment
The participants acknowledged the challenges in addressing ethical dissonance as AI
technologies become increasingly prevalent. The findings revealed two critical aspects of
organizational culture that support responsible AI adoption: the establishment of psychological
safety and ethical transparency in the workplace and the allocation of adequate resources for
ethical governance.
The MHLs acknowledged the challenge is exacerbated when organizations lack
awareness of ethical implications. As one participant described, “It’s kind of like a free game,
and no one has really told you to explain to you ethically or even how you use this.” This
statement highlights that when employees are not fully aware of AI use’s ethical implications,
they may inadvertently engage in practices that violate ethical standards. This lack of awareness
can expose organizations to malpractice liability (Bal, 2008). In other words, the unintentional
misuse of AI due to insufficient understanding of its ethical dimensions can have severe legal
and reputational consequences for mental health organizations.

101
The MHLs indicated that organizational culture significantly influences responsible AI
adoption. The participants emphasized that a responsible AI-adopting organizational culture is
built on two fundamental pillars: psychological safety and resource allocation. These pillars are
crucial in addressing the challenge of ethical dissonance that arises with the increasing
prevalence of AI technologies.
The participants recognize the importance of creating an environment that fosters
psychological safety and ethical transparency when adopting AI technologies. This approach is
crucial in addressing the challenges that arise from the widespread availability and use of AI.
Benjamin articulated this reality: “People will use it, as it is everywhere, and we cannot control
that.” Benjamin’s statement highlights the ubiquitous nature of AI and the difficulty in
completely controlling its use in an organization. This recognition underscores the need for a
proactive approach in guiding AI adoption rather than attempting to restrict it entirely. By
acknowledging this reality, leaders can focus on creating an environment that encourages
responsible use and open dialogue about AI technologies.
Creating a culture of psychological safety is essential for encouraging open
communication about AI use and its ethical implications (Clark, 2020). Jack emphasized the
importance of this aspect: “There is a level of psychological safety where they can feel like they
can come for it and ask those questions.” Jack’s observation underscores the value of an
environment where employees feel comfortable seeking guidance and expressing concerns about
AI technologies. This psychological safety enables a healthy feedback loop, allowing staff to
report small mistakes, share insights, and contribute to the responsible adoption of AI without
fear of judgment or reprisal (Clark, 2020).

102
To achieve this level of psychological safety and ethical transparency, organizational
leadership must establish clear guidelines and policies (Clark, 2020). Jonathan highlighted the
importance of a structured approach:
I think it’s important at a system level from the leaders to … have a structure in place
around, what is our stance, for lack of a better word, on AI and incorporation of AI into
our work as at a system level what processes are in place or policies are in place to helphelp provide some guidelines or guardrails around what that looks like?
Jonathan’s statement emphasizes the need for a systematic approach to AI adoption, with clear
policies and processes in place. This structured framework provides employees with the
necessary guidance to navigate the ethical complexities of AI use in mental health care, further
enhancing psychological safety and ethical transparency at the organization.
While establishing guidelines is crucial, maintaining an open and optimistic stance
toward AI adoption is equally important. Lee expressed this balanced approach: “I think right
now we’re cautiously optimistic would be the term. We definitely see the potential, and it has
been somewhat useful because we haven’t seen a lot of tools yet for mental health.” Lee’s
perspective reflects a cautious optimism that allows organizations to explore AI’s benefits while
remaining mindful of potential challenges, fostering an environment where ethical considerations
are at the forefront of AI adoption. Responsible AI adoption also requires a commitment to
cultural competence and inclusivity. Jennifer articulated this important aspect: “We’re very
progressive in that aspect that our desire is that we … use AI again in a way that is responsible
for that but then also along with cultural competence.” Jennifer’s statement highlights the
importance of integrating cultural competence into AI adoption strategies. This approach ensures
that AI technologies are implemented in a way that respects and addresses patients’ and staff

103
members’ diverse needs, further enhancing the ethical foundation of AI use in mental health
care.
Involving key stakeholders in the decision-making process is crucial for maintaining
ethical transparency and psychological safety (Clark, 2020; Edmondson, 1999). Benjamin
emphasized the value of this inclusive approach, “We also have really great clinicians in our
academic researchers, administrators, and things like that, and they tend to be involved in
decision-making, particularly if this is AI technology.” Benjamin’s observation underscores the
importance of leveraging diverse expertise in AI-related decision-making. By involving
clinicians, researchers, and administrators, organizations can ensure that AI adoption is guided
by a comprehensive understanding of its implications for patient care, research, and
organizational operations.
Fostering psychological safety and ethical transparency is crucial for the responsible
adoption of AI in mental health care (Clark, 2020). Furthered by This approach requires clear
guidelines, open communication, cultural competence, and inclusive decision-making processes
(Edmondson, 1999). In other words, by creating an environment where staff feel safe to discuss
AI-related concerns and contribute to ethical practices, organizations can navigate the
complexities of AI adoption more effectively. Considering the importance of psychological
safety and ethical transparency, it becomes clear that these efforts require adequate resources and
support, leading to the next subtheme: resource allocation for ethical governance.
The participants recognize the critical importance of allocating resources for ethical
governance in AI adoption. This allocation is essential for establishing structures and processes
that support responsible AI implementation. Solaris emphasized the need for a diverse, expertled approach: “Forming a multidisciplinary cross-functional ethics committee including mental

104
health professionals, AI researchers, legal experts, and ethicists to guide AI implementation and
manage ethical dilemmas.” Solaris’s suggestion highlights the importance of bringing together
diverse expertise to address the complex ethical challenges of AI in mental health care. Such a
committee can provide comprehensive oversight, ensuring that AI implementation aligns with
ethical standards and professional best practices. This multidisciplinary approach can help
organizations navigate the intricate landscape of AI ethics, fostering trust and accountability in
AI adoption.
However, allocating resources for these initiatives often presents significant challenges,
particularly when facing financial pressures. Kacey articulated this dilemma: “The challenge is
finding the resources, especially when these initiatives do not immediately generate revenue.”
Kacey’s statement underscores the tension between investing in long-term ethical governance
and meeting short-term financial goals. It highlights the need for organizations to adopt a more
holistic view of AI investments, considering immediate financial returns and long-term benefits
such as improved patient outcomes and enhanced quality of care. This perspective shift is crucial
for justifying and sustaining investments in ethical AI governance.
Despite these challenges, many organizations are taking proactive steps to allocate
resources for AI governance. John describes one such initiative: “We have a big IT department.
… I think they are planning to really have someone overlook the things that are being put in AI.”
John’s observation reflects a growing recognition of the need for dedicated oversight in AI
implementation. By allocating resources to create specialized roles or teams focused on AI
governance, organizations can ensure that ethical considerations are consistently integrated into
their AI strategies. This approach helps maintain accountability and alignment with
organizational values throughout the AI adoption process.

105
The comprehensive nature of responsible AI adoption requires investments across
multiple domains. Samantha elaborated on the diverse resource needs:
There’s a lot of training that gets involved. There’s a lot of integration that gets involved,
so the human training aspect of it is huge. You have to have enough people to understand
what’s going on to train them to continuously monitor that change management. And
then you have your separate set of people who do quality improvement like, “Is this
actually improving … our processes and efficiencies?” And then you have a separate set
of people who are actually doing the actual integration, like, integrating into your HER,
figuring out the whole implementation side of things, making sure that this works for
your organization based on your workflows.
Samantha’s detailed description highlights the multifaceted nature of resource allocation for
ethical AI governance. It emphasizes the need for investments in training, integration, quality
improvement, and technical implementation. This comprehensive approach ensures that AI
adoption is technically sound, ethically grounded, and aligned with organizational workflows.
Jennifer further emphasized the substantial nature of these resource requirements:
“It’s going to take a tremendous amount of resources because … you’re building the
infrastructure both, both personally and personnel-wise, and then also for the operational aspect
to make sure they’re functioning.” Jennifer’s statement underscores the significant investment
required for responsible AI adoption. It highlights that building the necessary infrastructure
involves both technological resources and human capital and operational considerations. This
holistic view of resource allocation is essential for creating a sustainable and ethically sound AI
implementation strategy.

106
Resource allocation for ethical governance is a critical factor in responsible AI adoption
at mental health organizations. It requires a multifaceted approach, encompassing diverse
expertise, dedicated oversight, comprehensive training, and robust infrastructure development
(Ahmed et al., 2023; Lu et al., 2023; Mendes-Santos et al., 2022). While financial challenges
exist, the long-term benefits of ethical AI governance justify these investments. As organizations
navigate these resource allocation decisions, they must also consider how these efforts impact
and are perceived by their staff. This consideration leads us to our next theme, provider
acceptance, which explores how mental health professionals view and adapt to AI integration in
their practice.
Theme 2: Early Adopters Influence Attitudes
The integration of AI in mental health care has sparked diverse reactions among
healthcare providers. While some enthusiastically embrace the technology, others approach it
with caution. In this landscape, early adopters play a crucial role in shaping attitudes and paving
the way for wider acceptance. Jonathan observed this phenomenon in their clinic: “There was a
group in our clinic, early adopters, where they were sort of excited to try it out.” This statement
highlights the presence of providers who are eager to explore AI’s potential in their practice as
they serve as pioneers, testing the waters, demonstrating the technology’s practical applications,
and influencing more hesitant colleagues.
The impact of these early adopters on their peers becomes evident as pilots and trials
progress. Jonathan further noted, “There were actually lots of reservations from our primary care
docs, but the pilot that did it, they loved it.” This observation underscores the transformative
power of successful implementation. Initial skepticism among primary care doctors gave way to

107
positive experiences, suggesting that hands-on exposure to AI can significantly shift attitudes.
These early adopters’ success can serve as a catalyst for broader acceptance at the organization.
The ripple effect of positive experiences extends beyond immediate participants, as
Jonathan elaborated,
I think the pilot that did it, they loved it. And … then the best way is when you hear it
from someone else that you trust, and they’re like, “Well, I did it, and I thought it was
great.” Now, others are sort of taking that on, and it’s going to be a new function that’s
available for most all primary care providers, I think, by the end of the year.
This statement illustrates how peer influence can accelerate AI adoption. Trust in colleagues’
experiences plays a significant role in overcoming initial hesitations, leading to a domino effect
of acceptance. The planned expansion of AI functionality to all primary care providers by yearend reflects the growing confidence in the technology’s benefits.
While early adopters’ experiences can be influential, some providers maintain a cautious
but open-minded approach. Benjamin expressed this perspective: “Again, those that are creating
this app that does that can argue with me, and again, they might be able to change my mind.
Again, I just need to be open to it and really learn it and see what that can do.” Benjamin’s
statement reflects a willingness to engage with AI technology, albeit with a healthy skepticism.
This attitude of openness to learning and potential perspective shifts is crucial for the gradual
acceptance of AI in mental health care. It suggests that even those who are initially hesitant can
be persuaded through education and demonstration of AI’s capabilities.
The adoption of AI is not solely determined by individual willingness, as Jack points out:
We’ll have some early adopters that I think will just run with the idea. And then I think
we’ll have some that are maybe regulated by outside bodies where it’s not just their

108
willingness to adapt to it. They have to essentially get it cleared by whether it’s insurance
or other auditing bodies to see if they can really integrate it.
Jack’s observation highlights the complex landscape of AI adoption, where regulatory and
insurance considerations play a significant role. This complexity underscores the need for a
comprehensive approach to AI implementation that addresses provider attitudes and external
regulatory requirements.
Fostering acceptance of AI often requires a collaborative approach, as Benjamin
suggested, “I think from what I understand and truly is just from my background experience is
really bringing people together initially and really focusing on the work that we do.” This
statement emphasizes the importance of collective engagement in the AI adoption process. By
focusing on the core work of mental health care, organizations can frame AI as a tool to enhance,
rather than replace, current practices. This collaborative approach can help alleviate concerns and
build consensus around AI implementation.
However, provider acceptance is intertwined with patient acceptance. Benjamin provided
an example from a different technological adoption: “We incorporated videotaping because it’s
for training purposes, and you had resistance before. There are some clients who refuse to come
to our clinic because they’re like, ‘I don’t want to be recorded.’” This example illustrates how
patient concerns can influence provider attitudes toward new technologies. It underscores the
need for a holistic approach to AI adoption that considers both provider and patient perspectives.
In conclusion, early adopters play a pivotal role in shaping attitudes toward AI in mental
health care. Their positive experiences can influence hesitant colleagues, creating a ripple effect
of acceptance. However, successful integration requires addressing regulatory constraints,
fostering open-mindedness, and considering patient perspectives. As organizations navigate

109
these complexities, the role of leadership becomes increasingly important. This leads to the next
theme, leadership characteristics, which pertains to how leaders can effectively guide their
organizations through the challenges and opportunities of AI adoption in mental health care.
Theme 3: Leadership Characteristics
Effective leadership plays a crucial role in the responsible adoption of AI technologies in
mental health organizations. The MHLs have identified several key characteristics that are
essential for guiding their organizations through this complex process. These characteristics
encompass a deep understanding of AI, strategic vision, and the ability to foster a supportive
organizational culture.
The participants highlighted fundamental leadership characteristics as the need for a
comprehensive understanding of AI and its potential impact on organizational operations. Adam
emphasized this point, “The leadership first of all needs to be fully immersed to understand the
depth of our operations would change if it was to come in.” This statement underscores the
importance of leaders having an in-depth knowledge of AI technologies and their implications.
This deep understanding enables leaders to anticipate and navigate the significant changes that
AI adoption might bring to their organization’s operations. It also positions them to make
informed decisions and provide guidance throughout the implementation process.
Beyond understanding, leaders must also set a clear vision and strategy for AI adoption.
This strategic approach is essential for guiding the organization through the complex process of
integrating AI technologies into mental health care practices. Solaris articulated this crucial
aspect of leadership, emphasizing the need for a comprehensive framework that addresses both
the technological and ethical dimensions of AI implementation: “Our leadership … sets a clear
vision and strategy for AI adoption, prioritizing ethical implementation, data privacy, informed

110
consent, continuous improvement, and human oversight.” Solaris’s observation highlights the
multifaceted nature of effective AI leadership. By establishing a clear vision that prioritizes
ethical considerations, data privacy, and continuous improvement, leaders can ensure that AI
adoption aligns with the organization’s values and long-term goals. This strategic approach helps
create a framework for responsible AI implementation that balances innovation with ethical
considerations and patient care (Buckwalter & California Psychological Association, 2023).
Another critical leadership characteristic is the ability to foster psychological safety and
open communication in the organization. Jack emphasized the importance of this supportive
leadership style:
I think leaders need to just be aware of what their staff do and have that intentional
support not only from a practical standpoint of “Do you know what interventions your
staff do and know enough about them to be able to talk about them?” But also, do you
just provide support to your staff in such a way where they’re more willing to come
forward and talk to you about something?
Jack’s statement underscores the need for leaders to be actively engaged with their staff’s work
and to create an environment where employees feel comfortable discussing concerns or ideas
related to AI adoption. This open communication fosters trust and can lead to more effective
implementation of AI technologies, as staff feel supported in navigating changes and challenges.
Leadership’s role in shaping organizational culture and driving genuine adoption is
another crucial aspect highlighted by MHLs. Sarah articulated this point:
I think that’s … that’s huge in terms of like the organizational leadership who is exactly
leading the charge what is their understanding of it. Like you can implement something,

111
and people will just do it because they think that it has to be done but they’re not really
like wholeheartedly wanting to do it.
Chi’s observation emphasizes the importance of leadership in driving the authentic
adoption of AI technologies. Leaders must both implement AI and cultivate an organizational
culture that genuinely embraces these technologies. This involves clear communication about the
benefits and rationale behind AI adoption, addressing concerns, and inspiring enthusiasm among
staff.
Effective leadership in the context of AI adoption in mental health organizations requires
a combination of in-depth understanding, strategic vision, and the ability to foster a supportive
and innovative organizational culture. Leaders must be fully immersed in understanding AI’s
implications, set clear ethical guidelines, provide intentional support to staff, and drive genuine
adoption through effective communication and culture-building. By embodying these
characteristics, leaders can guide their organizations through the complexities of AI adoption,
ensuring that the implementation aligns with the organization’s values and enhances its ability to
provide quality mental health care.
Results for Research Question 3
This section presents the findings related to the third research question: How do external
factors affect MHLs’ decisions to use AI for patient care? This analysis focuses on the
environment component of the TOE framework, exploring the external elements that influence
how MHLs approach AI adoption in their organizations. The environment in the TOE framework
is a multifaceted arena outside of the organization: the industry, competitors, regulations, and
relationships with the government. For mental health organizations considering AI adoption, this

112
environment is a complex web of stakeholders, regulatory bodies, societal expectations, and
market forces, each adding a layer of intricacy to the decision-making process.
Theme 1: Government Policies and Professional Guidelines
The participants recognize the critical role of government policies and professional
guidelines in shaping the adoption of AI technologies in mental health care. While the regulatory
landscape is still evolving, they emphasized aligning AI initiatives with existing frameworks to
ensure legal compliance and maintain public trust. One participant succinctly captures this
sentiment: “External factors include industry regulations and ethical guidelines, though these
may be lacking.” This observation highlights the current state of AI regulation in healthcare,
acknowledging both the presence of guidelines and the need for more comprehensive
frameworks. As the field rapidly evolves, MHLs must navigate this complex landscape,
balancing innovation with adherence to established norms and emerging best practices.
Despite the evolving nature of AI regulations, some MHLs note a growing acceptance of
technology in mental health care. Kacey observed,
Right now, I haven’t seen anything too negative. I think there [is] some more …
willingness to embrace it because they also know that we are wearing. … They [have] all
these … wearable stuff, … so people see the … practical side of things, and it’s also very
cost-effective.
Kacey’s statement reflects a pragmatic approach to AI adoption, noting that the increasing
prevalence of wearable technology has paved the way for greater acceptance of AI in mental
health care. This practical perspective suggests that as AI demonstrates its utility and costeffectiveness, it may face less resistance from both practitioners and regulators.

113
While there’s growing acceptance, the participants stressed the importance of adhering to
professional standards. Lee emphasized this point: “Yes, obviously, the mental health, we want
to make sure it’s within the BBS. Licensed psychologists, psychiatrists, medical professionals,
boards, it’s all approved technologies or tools that could be used.” Lee’s statement underscores
the necessity of aligning AI technologies with established professional standards and approval
processes. This approach ensures that AI tools meet the rigorous requirements set by licensing
boards and professional organizations, maintaining the integrity of mental health care practices.
As MHLs consider the implementation of AI, they also recognize the need for a nuanced
approach that considers patient needs and risk factors. Jack provides insight into this
consideration,
Because of that, there tends to be a high correlation with those receiving Medicaid and …
seriously mentally ill level, where it’s going to be a much higher intensity need. I really
think that the area where AI should likely be initially adapted within mental health is
those that might not be within the [seriously mentally ill] population because the risk of it
going wrong is so much greater.
Jack’s observation highlights the importance of a measured approach to AI
implementation, particularly when dealing with vulnerable populations. This perspective
suggests that initial AI adoption may be more appropriate for less severe cases, allowing for
refinement of the technology before applying it to high-risk situations.
The landscape of government policies and professional guidelines for AI in mental health
care is complex and evolving. Mental health leaders must navigate a path that balances
innovation with compliance, leveraging the practical benefits of AI while adhering to established
professional standards. As the field progresses, a thoughtful, risk-aware approach to AI

114
implementation will be crucial, particularly when dealing with vulnerable populations. This
careful consideration of regulatory and ethical frameworks sets the stage for our next theme: the
influence of managed care on AI adoption in mental health services.
Theme 2: Grant and Funding Source
The integration of AI into mental health services is heavily influenced by economics and
funding, as unanimously agreed by all participants. Community mental health organizations,
particularly those that are government-funded, face substantial barriers due to resource
constraints and the need for robust funding to invest in AI technology.
Historical precedent suggests potential challenges in AI adoption. One MHL noted,
“[Electronic medical record and electronic health record] technology was conceived around the
1960s, yet we didn’t get funding to implement it until the 2010s.” This significant lag in funding
for technological implementation may foreshadow a similar pattern with AI adoption. Given the
current lack of specific AI regulations and guidelines at the federal and state levels, MHLs
anticipate a delay in funding for implementation. As one participant succinctly stated, “I’m not
sure if the federal government at the funding level has figured that out yet.”
The disparity in funding capabilities raises significant equity concerns. Organizations
with better financial resources can afford superior AI tools, potentially gaining advantages in
client service, efficacy demonstration, and grant acquisition. This situation could create a selfreinforcing cycle of increased funding, exacerbating inequalities in mental health service
provision.
When federal funding for AI implementation eventually becomes available, MHLs
foresee new challenges related to performance expectations and future funding allocations. One
participant observed, “With grants and funding, last year’s performance is this year’s baseline.

115
Funding is getting less, but we are required to provide more services.” This reflects the doubleedged nature of AI adoption: while it may enhance efficiency, it could also lead to unrealistic
expectations from funders who might assume AI will reduce costs significantly or increase
capabilities overnight.
The pressure to adopt AI to meet escalating performance demands could compel some
organizations to implement AI hastily without thoroughly assessing its risks. Moreover, if AIdriven cost savings result in immediate funding reductions, it could exacerbate rather than
alleviate resource constraints. One MHL articulated these concerns:
One thing I am concerned about is we quickly follow suit hoping not to fall behind too
far … those who have the funding and know how to use AI well will get a better
competitive edge, and those without resources will fall further behind.
The landscape of AI adoption in mental health services is significantly shaped by funding
disparities and economic considerations. While AI holds promise for improving service delivery,
the uneven distribution of resources could widen the gap between well-funded and underresourced organizations. As the field moves forward, addressing these funding inequities and
carefully managing the expectations associated with AI implementation will be crucial to
ensuring equitable access to advanced mental health care technologies.
Theme 3: Managed Care
The influence of managed care, particularly Medicare, on the adoption of AI in mental
health services emerged as a significant theme among MHLs. As the largest payor in the United
States, Medicare’s policies often set the standard for the entire healthcare industry, including the
integration of AI technologies. Samantha articulated this trend succinctly: “In the industry, a lot

116
of times followed the big leader, which is Medicare. If Medicare moves one way, people—the
industry—tends … to lean towards that.”
This observation highlights Medicare’s role in shaping industry-wide practices. Medicare
approval of AI-driven tools and services often spurs widespread adoption in mental healthcare,
influencing care providers and insurers to align their policies with Medicare standards
(Finkelstein & McKnight, 2005; Rosenfeld et al., 2005).
The impact of managed care policies on service provision, especially for high-need
populations, is a critical consideration for MHLs. Jack explained how this affects their approach
to care: “So, because of that, to serve our population that we’re passionate about, which is this
high-need area of Medicaid patients, we’ve had to expand our servicing range to find additional
ways to bring them in.”
Jack’s statement underscores the need for innovative approaches to care delivery,
particularly for vulnerable populations. The adoption of AI technologies could be one such
innovation, potentially expanding access and improving care for high-need patients. However,
this adoption is contingent on managed care policies that support and reimburse these new
approaches.
The potential for AI technologies to be included in reimbursable costs is an exciting
prospect for many MHLs. Jennifer explores this possibility:
And also, you look at, say, reimbursable costs with healthcare companies. And … with
healthcare companies be able to say, “Hey, we’re going to … have this as … one of our
codes that you can as one of our available codes, and you’ll be able to offset that cost ofof the Apple Vision Pro?”

117
Jennifer’s speculation about the potential for AI technologies to be included in reimbursable
codes highlights the transformative impact managed care policies could have on AI adoption. If
advanced AI tools become reimbursable, it could significantly accelerate their integration into
mental health care practices, potentially revolutionizing care delivery and accessibility.
The policies of managed care organizations, particularly Medicare, play a crucial role in
shaping the landscape of AI adoption in mental health services. These policies influence what
technologies are used and how care is delivered, especially to vulnerable populations. As AI
continues to evolve, the alignment of managed care policies with these technological
advancements will be critical in determining the extent and speed of AI integration in mental
health care. The industry will be watching closely for signals from major payors like Medicare,
which will likely set the tone for broader AI adoption and reimbursement practices.
Theme 4: Competitors and Collaborators
The adoption of AI in mental health organizations is driven by an interplay of
competitive pressures and collaborative efforts. The participants recognize the need to innovate
to remain relevant in an evolving landscape while also valuing shared knowledge and
experiences. Solaris articulated the competitive aspect: “The competitive nature of the mobile
health commercial space drives the adoption of AI to stay relevant and meet customer needs.”
This statement highlights how market forces are pushing organizations to embrace AI
technologies. The desire to stay competitive and meet changing patient expectations is a
significant driver for AI adoption, compelling organizations to innovate and adapt to the rapidly
changing technological landscape.
Despite these competitive pressures, many participants emphasized collaboration and
learning from peers. One leader expressed this collaborative sentiment: “Maybe we can talk to

118
other organizations and see what worked and what didn’t. So, we look out there and see who has
done this.” This approach underscores the value placed on shared experiences and best practices.
By learning from others’ successes and failures, organizations can make more informed
decisions about AI adoption, potentially avoiding pitfalls and accelerating implementation. Some
organizations have formalized these collaborative efforts. As another participant noted, “We are
a part of a consortium where we get together and share ideas. Some board members are also
leaders in other CMHOs.” Such consortiums facilitate knowledge exchange and create a
supportive network for navigating the challenges of AI adoption. This collaborative spirit
suggests that many leaders view AI implementation as a shared challenge that benefits from
collective wisdom and experience.
The rapid advancement of AI technologies can create a sense of urgency among
organizations. John provides an example from their experience: “With Chat GPT, clinical staff
and students started using it the second it came out, and our organization was like, ‘This is going
to be a huge problem.’ I think that there was a push to create one internally.” This observation
illustrates how the swift adoption of AI by users can prompt organizations to rapidly develop
their own solutions. It highlights the pressure to keep pace with technological advancements and
user expectations, as well as the need to address potential challenges proactively.
The rapid adoption of AI can lead to concerns about competition, as Adam expressed:
And then I also started thinking, if that was the case, “Why would other organizations not
do the exact same thing?” Then was the competition. How do you stand up with against
another agency who could be awarded for this? What would be the basis of that?
Adam’s reflection reveals the competitive considerations that arise as organizations contemplate
AI adoption. There is a recognition that falling behind in AI implementation could potentially

119
disadvantage an organization in securing funding or attracting clients, driving a sense of urgency
in adoption decisions.
However, not all MHLs view the landscape through a competitive lens. Jennifer offered a
different perspective:
I don’t look at—and this is unique for me—I don’t look at … people who are in the same
space as … competition. I … look at them as individuals who … are extending … this
space and its reach because if … one does well, then we all do.
Jennifer’s view emphasizes the potential for collective advancement in the field of mental health
care. This perspective suggests that successful AI implementation by one organization can
benefit the entire sector by expanding capabilities and improving overall care quality,
highlighting the interconnected nature of progress in mental health services.
The adoption of AI in mental health organizations is shaped by a delicate balance
between competitive pressures and collaborative opportunities. While market forces drive
innovation and rapid adoption, there is also a strong recognition of the value of shared
knowledge and collective efforts. MHLs must navigate this landscape by balancing the need to
stay competitive with the benefits of collaborative learning and advancement. As the field
continues to evolve, fostering a culture that encourages both innovation and collaboration will be
crucial for the successful and ethical implementation of AI in mental health care, ultimately
benefiting both individual organizations and the sector as a whole.
Theme 5: Patient Acceptance and Care Access
The interviews revealed two key aspects of patient perspectives: patient receptivity and
access to AI-enabled mental health services and the challenges of managing patient expectations
while maintaining an appropriate balance between AI assistance and human care. The following

120
sections explore patient trust, acceptance, concerns regarding AI tools, and the impact these
technologies have on the patient experience, outcomes, and engagement with care. The findings
emphasize the importance of addressing ethical considerations, ensuring transparency, and
maintaining human connection in AI-assisted healthcare to ensure patients feel valued and
supported. Understanding patient perspectives is crucial for the responsible and successful
implementation of AI in healthcare settings.
The adoption of AI in mental health services is significantly influenced by patient
receptivity and access to care. The participants observed that receptivity varies among different
patient groups, with some being more open to AI-assisted interventions than others. Zach notes a
particular group that shows promise: “The highest level of receptivity is going to be through
those individual patients that are coming, finding us online, or walking into one of our offices on
their own. I think they are going to be the most receptive.” This observation suggests that
patients who proactively seek mental health services may be more open to innovative
approaches, including AI-assisted care. Their initiative in seeking help could translate into a
willingness to engage with new technologies, potentially making them ideal early adopters of AIintegrated mental health services.
However, access to mental health services remains a significant challenge for many
patients, particularly in community health settings. Jonathan highlighted a common patient
perspective:
I think a lot of it is driven by what patients are asking for, which is it should not be that
hard for me to get access to healthcare or schedule an appointment or to connect with my
doctor those kinds of things.

121
Jonathan’s comment underscores the potential for AI to address fundamental access issues in
mental health care. AI-powered solutions could streamline processes like appointment
scheduling and doctor-patient communication, potentially reducing barriers to care.
The complexity of these access challenges is further illuminated by Benjamin’s
observation: “Many of our patients are housing and food insecure. It is a lot of trying to navigate
our complicated healthcare system while finding housing and making a living.” This statement
highlights the broader socioeconomic factors that impact access to mental health care. For
patients facing basic survival challenges, the integration of AI into mental health services must
be considered within the context of these pressing needs. Given these diverse patient
experiences, the participants emphasized involving patients and their communities in the AI
adoption process. Benjamin articulated this sentiment: “Involvement of community members is
crucial as their feedback and engagement will influence the effectiveness and acceptance of AI.”
Benjamin’s statement underscores the value of patient perspectives in shaping AI
implementation. By involving patients and community members, organizations can ensure that
AI solutions address real needs and align with patient values.
The receptivity to AI in mental health care may vary by individual circumstances and the
reputation and practices of the healthcare provider. John offered an insight from their experience:
I feel like our patients, for example, would not be too concerned about it too much. I
think it’s neither good [nor] bad. I think from the patients I’ve seen and interacted, like,
our hospital does a lot of renowned medical procedures. People come from all over the
nation, sometimes, to our hospital. I … think that they would have a positive effect
towards us using it.

122
John’s observation suggests that patients’ trust in a healthcare institution may extend to their
acceptance of AI technologies implemented by that institution. This points to the importance of
institutional reputation and patient trust in facilitating AI adoption.
A significant factor in patient receptivity to AI is the generational divide in technology
adoption. Benjamin highlighted this challenge:
We have multi-generational levels of different types of clients. We have older folks who
are more traditional who, to be honest with you, they don’t even know how to log on to
Zoom. We have to guide them through the process, and that’s like a quarter of our session
trying to get them there.
Benjamin’s comment illuminates the practical challenges of implementing AI technologies
across diverse age groups. The varying levels of technological literacy among patients
necessitate a flexible approach to AI integration, potentially including additional support and
education for older patients.
Patient receptivity and access to AI in mental health care are influenced by an interplay
of factors, including individual initiative, socioeconomic circumstances, community
involvement, institutional trust, and generational differences in technology adoption. Thus,
MHLs must navigate these diverse considerations to ensure that AI implementation enhances
rather than hinders access to care. As the field moves forward, a patient-centered approach that
addresses these varied needs and perspectives will be crucial for the successful and equitable
integration of AI in mental health services.
The integration of AI in mental health services brings both opportunities and challenges,
particularly in managing patient expectations and maintaining a balance between AI efficiency
and human interaction. The respondents expressed concerns about the potential impact of AI on

123
patient expectations and the quality of care. Kacey highlighted a key concern regarding the
immediacy of AI responses:
Although I do see benefits for some members of the population that this is good because
they get immediate responses, it also then sets it up where when they are actually with a
human person in real life be very disappointing because in life things are not so quick to
respond and you might get disappointed responses that you do not want to hear.
Kacey’s observation underscores the potential for AI to create unrealistic expectations in humanto-human interactions. While AI can provide immediate responses, it may inadvertently set a
standard for speed and availability that human providers cannot match, potentially leading to
patient disappointment or frustration in traditional therapeutic settings.
Another concern the participants raised is the risk of AI-driven care becoming overly
standardized or prescriptive. Kacey elaborated on this point:
I think it could become too automated where it’s like everyone wears a blue shirt
medium-sized or when I go to sporting events, the … free giveaway shirts are all largest.
Not an exact after Apple metaphor parallel example, but I think there is that fear of it
becoming too prescriptive.
This metaphor illustrates the concern that AI might lead to a one-size-fits-all approach to mental
health care, potentially overlooking patients’ nuanced, individual needs. The fear of overautomation highlights the importance of maintaining human judgment and personalization in
mental health services.
While acknowledging the potential benefits of AI, some MHLs emphasize the
irreplaceable nature of human interaction in mental health care. Benjamin expressed this
sentiment:

124
Again, AI is not 100% proof from what I see. Someone else in AI could argue with me
definitely, but I still say, even if they do, it’s 99%. I still don’t believe it. Human factorsour brain is so complex.
Benjamin’s statement underscores the belief that the complexity of the human mind necessitates
a human touch in mental health care. This perspective suggests that while AI can be a valuable
tool, it should complement rather than replace human expertise and empathy.
To effectively integrate AI into mental health services, MHLs stress the importance of
clinician understanding and engagement with AI tools. Jack articulated this need:
I think we just have to know how the app works. And I think if something like that is
created then the clinicians that are referring to it need to be really aware of what it works.
And I think that maybe that’s how we build it, or maybe it’s how the person that owns it
sells it to us, like, whatever information. But I also think it has to be something that we
can just play around with, and clinicians can get a feel for it.
Jack’s comment highlights the importance of clinician familiarity and comfort with AI tools. By
ensuring that mental health professionals understand and can effectively use AI technologies,
organizations can better manage patient expectations and maintain a balance between AI
efficiency and human expertise.
Managing expectations and maintaining a human-AI balance in mental health services is
a complex challenge. While AI offers benefits such as immediate responses and potentially
increased access to care, it also risks creating unrealistic expectations and overly standardized
approaches. MHLs emphasize the need for a thoughtful integration of AI that complements
rather than replaces human interaction. As the field moves forward, finding ways to leverage

125
AI’s efficiency while preserving the irreplaceable aspects of human care will be crucial for
providing effective, personalized mental health services.

126
Chapter Five: Discussion and Conclusion
The integration of AI in mental health care represents a transformative shift in healthcare
delivery, offering unprecedented opportunities to enhance patient care while presenting
significant challenges for implementation. This study explored how MHLs approach AI adoption
through the TOE framework, providing insights into the interplay of factors that influence
responsible AI implementation in mental health settings.
The research addressed three fundamental questions, each aligned with a component of
the TOE framework:
1. What are the MHLs’ perceptions of AI technologies in mental health care?
2. What organizational factors enable MHLs to adopt AI to improve patient care while
minimizing risks?
3. What external factors affect MHLs’ decisions to use AI for patient care?
This study’s findings reveal a landscape characterized by both promise and caution,
where technological capabilities must be balanced with ethical considerations and practical
constraints. As Haber et al. (2024) emphasized, AI represents a fundamental paradigm shift with
social and psychological implications for mental health care. This chapter synthesizes the
research findings in Chapter Four and presents practical recommendations for organizations
pursuing responsible AI adoption.
Technology Component: MHLs’ Perceptions and Implementation Strategies
The integration of AI technology in mental health care is notably shaped by perceptions
and attitudes (Liehner et al., 2023). This study revealed a pattern of cautious optimism among
leaders who recognize AI’s potential while maintaining awareness of its limitations and risks.
MHLs identified several promising applications of AI, particularly in enhancing personalized

127
and holistic treatment approaches. According to Keaton (2022) and Moggia et al. (2024), holistic
treatment in mental health requires considering an individual’s overall well-being, including
physical, emotional, social, and spiritual health, with AI offering the capability to personalize
care based on individual genetic, biological, environmental, and psychosocial factors through
data-driven strategies. MHLs envision AI as a tool to facilitate this comprehensive approach
through sophisticated data analysis and pattern recognition. The technology’s potential to
identify connections between mental health and physical symptoms, lifestyle factors, and
environmental influences represents a significant advancement in holistic care delivery.
The key concerns identified cluster into four critical areas that will be addressed in the
subsequent recommendations. First, participants highlighted the potential inflexibility of AI
systems and their limited capacity to grasp the nuanced complexities of mental healthcare. As
one participant emphasized, “The line of work we do is not always very simple. … It’s like 1
plus 1 is 2 minus 3 divided by 5.” AI systems, by their nature, rely on data-driven algorithms that
often lack the ability to interpret the subtleties of human emotions and psychological states
(Graham et al., 2019). Second, there are significant concerns regarding algorithmic bias and
fairness, particularly stemming from the lack of diverse representation in AI training datasets
(Gupta, 2023).
According to Giovanola and Tiribelli (2022), bias in healthcare machine learning
algorithms stemmed from model design (such as label biases and cohort biases), training data
(such as minority bias, missing data bias, informativeness bias, and training serving skew),
interactions with clinicians (such as automation bias, feedback loops, dismissal bias, and
allocation discrepancy), and interactions with patients (such as privilege bias, informed mistrust
and agency bias). Third, stakeholders raised issues surrounding privacy and transparency,

128
specifically regarding the handling of sensitive patient data and informed-consent processes.
Finally, while AI shows promise in reducing staff workload and burnout, MHLs expressed
apprehension about potential over-reliance on these technological solutions and create oversight
in diagnostics and treatment strategies.
Technology Recommendations
Implementing AI in mental health care requires careful consideration beyond technical
capabilities. Overall, AI represents a paradigm shift with social and psychological implications
for the field (Haber et al., 2024). This study found that MHLs see significant potential in AI
technology’s ability to transform services. The MHLs identified several promising applications
of AI, including its capacity to deliver more personalized treatment through sophisticated data
analysis. They particularly value AI’s potential to analyze large amounts of data to provide
predictive and preventative care, thereby enabling earlier interventions and more targeted
treatment approaches. Furthermore, AI’s ability to enhance accessibility to services while
simultaneously reducing clinician burnout through automated administrative tasks represents a
significant advancement. MHLs also recognize AI’s potential to improve compliance with
funding guidelines and optimize fee collection processes, addressing crucial operational
challenges in mental health care delivery.
However, MHLs expressed significant concerns about the implementation and use of AI
technologies. A primary concern centers on the potential rigidity of AI approaches and their lack
of nuanced understanding of the complexity of mental health work. They worry about biases and
fairness issues arising from limited diversity in training data, which could lead to inequitable or
inappropriate care recommendations for certain patient populations. Privacy, transparency, and
security emerge as critical concerns, particularly regarding whether patients have adequate

129
understanding and consent regarding how their data is being used. These concerns encompass
broader issues of informed consent, data security, shared responsibility among stakeholders, and
the potential for errors in AI systems that could impact patient care.
A fundamental concern expressed by MHLs is the potential impact on human-to-human
interactions in mental health care. The findings strongly emphasize that organizations should not
view AI as a replacement for human expertise but rather as a tool to enhance human capabilities.
Maintaining a balance between the use of AI and human judgment is critical to ensuring ethical
decision-making (Adelakun et al., 2024). Additionally, MHLs raised important technical and
regulatory concerns about data interoperability and compliance with healthcare regulations,
highlighting the need for robust systems that can meet stringent healthcare data management
requirements.
Given these far-reaching implications, the MHL participants believe mental health
organizations need a structured approach that addresses both the technical implementation of AI
and its broader impact on care delivery and patient well-being. While this paper cannot address
all potential implications of AI, it focuses specifically on addressing the primary concerns raised
by the MHLs in this study. The recommendations in this chapter provide a practical framework
for organizations to implement AI responsibly, addressing technological, organizational, and
environmental considerations. The technological recommendations focus on three key areas:
consumer-facing applications, administrative systems, and data security infrastructure. Each area
requires specific safeguards and protocols to ensure ethical implementation while maximizing
the benefits of AI integration in mental health services.

130
Overall AI Systems Design Approach
The interviewed MHLs expressed significant concerns about AI implementation,
including data integrity, algorithmic fairness, and security. While study participants lacked
technical expertise, my analysis identified potential technological solutions like systems design
frameworks and blockchain that may address their identified concerns, going beyond their nontechnical recommendations. Blockchain technology for data storage could enhance data security
and transparency, while careful systems design could help ensure algorithmic fairness and
appropriate AI integration. However, these technical approaches should be viewed as partial
solutions that require further research and validation rather than comprehensive answers to the
complex challenges of AI adoption in mental healthcare.
Inclusive AI Design Framework
A universal concern among the interviewees was whether AI training data sufficiently
represents diverse cultural backgrounds and mental health experiences. Beyond HIPAA-secured,
the foundation of responsible AI implementation in mental health care must be built upon
principles of inclusive design that consider all potential users’ diverse needs to mitigate data bias
and algorithm bias (Powell & Kleiner, 2023). Regardless of whether organizations are
implementing consumer-facing technology or administrative systems, they must prioritize
inclusive design principles throughout the development or selection process of AI products.
Mental health organizations must conduct comprehensive user research that deliberately includes
traditionally underrepresented populations in mental health care, including racial and ethnic
minorities, LGBTQIA+ individuals, neurodivergent users, people with disabilities, and
individuals from various socioeconomic backgrounds. Organizations should employ
participatory design methods where members of these communities are not merely subjects of

131
research but active participants in the design process, ensuring that AI systems reflect diverse
perspectives and experiences from the earliest stages of development.
The framework must extend beyond simple translation to incorporate true cultural
adaptation of content and interfaces through multi-language support that considers cultural
nuances, idioms, and mental health concepts that vary across different communities. This
includes adapting clinical terminology, assessment tools, and therapeutic approaches to align
with different cultural understandings of mental health. Organizations should implement rigorous
testing protocols using digital twin technology to prevent potential harm before deployment,
incorporating role-playing exercises and real-world testing with feedback from diverse user
groups. Comprehensive accessibility features must be integrated as core components rather than
additions, developing multiple access methods that accommodate different needs and
preferences, including text-based interfaces, voice interaction systems with multiple accent
recognition, video options with captioning in multiple languages, and adaptable interface designs
for various cognitive styles.
A diverse community advisory board should be established to provide ongoing oversight
and validation of AI features and functionality, including mental health professionals from
diverse backgrounds, cultural community representatives, disability advocates, and service users
from underrepresented groups. The board should have real authority to influence design
decisions and implementation strategies, ensuring that inclusive design principles are maintained
throughout the AI system’s life cycle. Organizations must establish continuous improvement
processes that regularly assess and update AI systems based on user feedback and emerging
needs, with clear metrics for measuring success across different demographic groups. Through
this comprehensive approach to inclusive design, mental health organizations can ensure that

132
their AI systems truly serve all members of their community while maintaining high standards of
cultural competence and accessibility, ultimately enhancing the effectiveness and reach of AIenabled services.
Blockchain for Responsible Data Sharing and Data Sovereignty
The majority of respondents expressed concerns about data transparency, security, and
interoperability in AI implementation. Leveraging permission-based and shared blockchain
technology as decentralized data storage can enhance data security, transparency, data
ownership, and accountability, addressing several challenges related to data management and
trustworthiness (Halamka et al., 2017).
Blockchain has had success in helping the country Estonia’s e-Health Foundation
launched blockchain technology to secure healthcare records in 2016 (Agbo et al., 2019; Andero,
2023). Blockchain technology emerges as a promising solution to address these challenges in
both consumer-facing and administrative applications, and when done correctly, blockchain
could enhance data security and ensure compliance with regulatory frameworks such as the
GDPR (Al-Abdullah et al., 2020). Blockchain’s decentralized architecture can provide a secure
and transparent framework for managing sensitive mental health data while meeting
organizational needs for efficient data management (Halamka et al., 2017).
Blockchain can create transparent, immutable audit trails of data usage through smart
contracts and zero-knowledge proofs. Zero-knowledge proofs can solve the AI transparency
problem by allowing one party, known as the prover, to convince another party, the verifier, that
a particular statement is true without revealing any additional information beyond the validity of
the statement itself (Bai et al., 2022). The essence of zero-knowledge proofs enables
organizations to verify AI model training practices without accessing sensitive training data, and

133
patients can verify how their data is being used in AI training without exposing their personal
information. Through blockchain-enabled smart contracts, health organizations can provide
patients the transparency and control over their health information with Dynamic Consent, where
patients can grant or revoke access permissions in real time (Appenzeller et al., 2022; Hasan et
al., 2021). This capability extends to AI implementations, allowing patients to make informed
decisions about whether their data can be used for model training while maintaining their privacy
through zero-knowledge proofs (Gaba et al., 2022).
Artificial intelligence systems in healthcare require interoperability to improve their
accuracy through access to diverse datasets and reduce biases that may develop when operating
in isolation. This is particularly crucial in mental health care, where patients often receive
treatment across multiple providers and settings (e.g., community mental health centers, private
therapists, and psychiatric hospitals) and where mental health conditions frequently co-occur
with physical health issues. Without interoperability, each provider’s AI system would need to
start fresh in understanding patient patterns, potentially missing insights from previous care
settings. Blockchain can standardize data exchange and enhance system interoperability,
ultimately leading to better patient care (Durneva et al., 2020). As patients move between
different providers or health systems, blockchain enables secure and seamless data portability
while maintaining complete audit trails of data access and usage. This interoperability, combined
with blockchain’s inherent security features, helps organizations meet regulatory requirements
for data protection while reducing administrative overhead. The result is a more efficient,
transparent, and secure system that balances patient privacy rights with organizational needs for
data access and AI development.

134
Although the participants did not explicitly bring up the issue of data colonization and
data sovereignty, those technical issues can be contributing factors to data colonization, as
discussed in Chapter Two. Blockchain can potentially help with decolonizing data by ensuring
communities maintain control over their data and helping patients maintain data sovereignty
(Martin et al., 2022). While data sovereignty is a critical consideration in AI implementation and
one that the researcher views as essential to ethical AI adoption, this topic did not emerge
organically from participant interviews. This absence may indicate an opportunity for future
research exploring MHLs’ perspectives on data sovereignty and its implications for AI adoption.
Multilingual and Multi-modal Disclosure for Transparency
The MHLs raised the question of how to effectively communicate AI use to patients
presents a significant challenge in healthcare settings. While informed consent is standard
practice for many AI consumer products, traditional disclosure methods often use complex
language that can be difficult for patients to understand, particularly for non-English speakers. In
the context of clinical decision support tools, the most common AI application in healthcare,
organizations face the critical question of how to effectively disclose AI use and what
information should be provided to patients (H. J. Park, 2024).
This study revealed the need for a more accessible and comprehensive approach to AI
disclosure to patients. Durneva et al. (2020) noted that under GDPR requirements, patients have
the right to understand how AI systems make automated decisions about their healthcare,
requiring organizations to provide meaningful explanations of their AI’s decision-making
processes. Despite the lack of clear guidelines, the MHLs emphasized that organizations should
implement clear, understandable disclosure protocols for both consumer-facing and
administrative AI applications. This information should be presented in simple, accessible

135
language that explains how AI systems use patient data, how this data is stored, and how it
impacts patient care. The informed consent form should describe the role of human oversight in
AI-assisted care, emphasizing that AI tools supplement rather than replace human clinical
judgment while clearly communicating patient rights, including their options to opt out of AIenhanced services. Organizations should maintain this transparency throughout the patient’s
entire journey, from initial marketing materials through intake processes and ongoing treatment
documentation.
As the participants noted, patients need to understand both the potential benefits and
limitations of AI-assisted care. To accommodate different cultural and language barriers, a
recommendation is for organizations to use different languages and mediums, such as interactive
videos or short-form videos, to explain the disclosure, making the information easy to
understand. The findings of H. J. Park (2024) indicated that patient preferences for information
regarding AI use vary significantly, necessitating tailored disclosure practices that reflect
individual needs and concerns. According to Pandey et al. (2021), employing a variety of
communication formats, such as interactive and short-form videos, is essential for organizations
aiming to bridge cultural and language gaps. The implementation of these protocols must remain
flexible and responsive to diverse patient needs. As Jack noted, different patient populations
have varying levels of receptivity to AI tools; therefore, communication approaches should take
technological literacy, cultural background, and individual preferences into consideration.
To keep up with the fast pace of AI technology, organizations should regularly review
and update their disclosure protocols to reflect changes in AI capabilities, regulatory
requirements, patient feedback, and emerging ethical considerations. Cao et al. (2023)
emphasized the need for organizations to regularly update their disclosure protocols to align with

136
the latest developments in AI and regulatory landscapes. This patient-centered approach,
supported by dynamic and comprehensive disclosure practices, helps maintain trust while
leveraging AI’s potential to enhance access to and quality of mental health care.
Consumer Technology: Safety and Human Support
Many MHLs conveyed the concerns that AI may impact human-to-human connections
during interventions; therefore, consumer technology user interface should be distinctly nonhuman looking to avoid misconceptions about AI’s “humanness” and capabilities while clearly
communicating that AI tools complement, rather than replace, human mental health providers.
According to Maddali et al. (2022), AI interfaces should avoid human-like characteristics and
instead focus on clear, functional design, especially for vulnerable populations such as
individuals with dementia or psychosis. This approach helps users maintain realistic expectations
about AI’s capabilities and mitigate disappointments (Gorre et al., 2023). One MHL’s
observation that “the connection between a human to a human … is the basis for therapy”
underscores the importance of maintaining the therapeutic alliance while integrating AI tools
(Battaglia, 2019).
Organizations should establish clear protocols for situations where AI interactions
indicate clinical risks or emergencies, ensuring seamless handoff to human providers when
necessary, maintaining patient safety, and ensuring that appropriate care is delivered in critical
situations. Di Sarno et al. (2024) discussed the integration of AI in pediatric emergency medicine
and emphasized the need for well-defined procedures for escalating care to human providers
when AI systems detect urgent clinical needs to ensure that the capabilities of AI are effectively
complemented by human expertise. A graduated approach to AI implementation is recommended
through a three-tier system. The first tier involves basic AI chatbots for initial screening and

137
psychoeducation. This progresses to the second tier, which incorporates AI-augmented self-help
tools with human monitoring. The third tier represents a hybrid care model combining AI
analytics with human therapy. This comprehensive system should include real-time risk
detection capabilities with automatic escalation to human providers through keyword monitoring
and easily accessible help features. Regular assessment of therapeutic outcomes and client
satisfaction ensures continuous improvement and accountability.
Mental health organizations must implement automatic handoff systems that trigger an
immediate transition from AI to human providers when certain risk indicators are detected based
on keywords identified. These transitions should occur when AI systems identify expressions of
suicidal ideation, potential harm to self or others, severe psychological distress, acute psychotic
symptoms, significant changes in mental status, complex trauma disclosures, medication-related
emergencies, or situations involving domestic violence or abuse. The handoff process must be
immediate and seamless from the patient’s perspective, with clear documentation requirements
and automatic escalation to appropriate clinical staff, followed by a thorough review of all
documentation by human providers.
Administrative Technology: Addressing Over-Reliant on AI
While AI technology is promising, MHLs are concerned about human clinical staff
becoming overly reliant on AI diagnostic evaluation and treatment suggestions. As AI becomes
increasingly integrated, there are growing concerns that clinicians’ dependency on it could
decrease their ability to think critically and make independent decisions (Ali et al., 2024). The
MHLs’ concerns are not unfound as clinicians will rapidly become less skilled, less attentive,
and less discerning as AI becomes a more ubiquitous component of their clinical work (AdlerMilstein et al., 2024). To maintain clinical accuracy, accountability, and transparency,

138
organizations can consider implementing AI watermarks similar to SynthID or scrambled
sentences at random intervals to mandate clinician vigilance to review texts to make proper
adjustments and corrections, require licensed clinicians to identify AI-generated text during
document review, and ensure proper oversight before EHR approval. Regular accuracy audits
and human override capabilities for all AI decisions provide additional safeguards against system
errors or inappropriate automated responses. Real-time vigilance tracking to track clinicians’ AI
suggestions acceptance rate can be a good measure and can prompt education, feedback,
coaching, and even turning off the AI for a period in serious cases (Adler-Milstein et al., 2024).
The quality assurance framework should track key performance indicators, including
documentation accuracy rates, system response times, and clinical efficiency metrics through
monthly audits of AI-generated administrative documents and quarterly comprehensive analyses.
Organizations should continuously evaluate the impact of AI administrative systems
through comprehensive monitoring that includes clinical operations metrics, patient satisfaction
assessments, and compliance verification. Real-time tracking of system interactions, automated
alerts for potential privacy breaches, and regular HIPAA compliance checks should be integrated
into daily operations. Annual external audits provide independent verification of system
performance, while quarterly risk assessments help identify and address potential vulnerabilities
before they impact operations.
Organization Component: Enabling Factors and Implementation Framework
The organizational factors enabling responsible AI adoption emerged as crucial
determinants of implementation success. Rangavittal (2022) found that organizational culture
and leadership play pivotal roles in AI adoption, with particular emphasis on ethical
considerations and change management. In this study, MHLs emphasized that successful

139
implementation requires more than technical capability—it demands systematic organizational
change that addresses both structural and cultural elements.
A finding centers on the role of organizational culture in fostering ethical AI adoption.
The research reveals that organizations lacking awareness of ethical implications may
inadvertently engage in practices that violate ethical standards. This situation creates potential
ethical dissonance, which Barkan et al. (2015) defined as the internal conflict experienced when
actions do not align with personal or professional values. In professional contexts, this
dissonance can occur when external forces, such as organizational policies or technological
requirements, conflict with ethical standards (Lauria & Long, 2019).
The influence of early adopters emerges as a significant factor in shaping organizational
attitudes. Early adopters can act as a positive force to promote and encourage the adoption of
technologies by the mainstream majority (Latif, 2017). This peer influence proves particularly
powerful in healthcare settings, where professional trust and credibility significantly impact
adoption decisions. The research demonstrates that successful implementations often begin with
a core group of enthusiastic practitioners who can effectively demonstrate the benefits and
address their colleagues’ concerns.
Education and training of the healthcare workforce represent another critical factor, often
emerging as a significant barrier to AI adoption (Ahmed et al., 2023). MHLs emphasized the
importance of comprehensive training programs that address both technical competency and
ethical awareness. Furthermore, psychological safety emerges as a crucial element that promotes
integrity and ethical behavior in the organization (Ferrère et al., 2022), creating an environment
where staff feel comfortable voicing concerns and seeking guidance about AI implementation.

140
Organizational Recommendations: ADKAR Framework Implementation
The research findings revealed that the participants emphasized the need for systematic
organizational change to implement AI responsibly. They identified critical components for
successful implementation, including the need for a clear understanding of AI’s implications,
building staff buy-in through early adopters, providing comprehensive training, and ensuring
ongoing support with feedback mechanisms. These identified needs align naturally with the
ADKAR change management framework of awareness, desire, knowledge, ability, and
reinforcement (Zine et al., 2023). The ADKAR framework is a tool that companies use to make a
change, increasing productivity and improving working conditions. The usage of Prosci’s
ADKAR model as the change management framework is ideal, as its main priority is people over
process (DePodesta, 2024). Therefore, ADKAR provides an appropriate structure for organizing
the organizational recommendations for AI implementation in mental health settings, addressing
both the technical and human aspects of change that MHLs identified as crucial.
Awareness: Creating an Ethics-First AI Vision
The awareness stage of ADKAR is crucial for laying the foundation for successful AI
adoption in mental health organizations. According to Ball and Hiatt (2024), Awareness
addresses fundamental questions about why change is needed and the risks of not changing.
MHLs emphasized that organizations must focus first on developing a deep understanding of
why change is necessary and how AI will impact care delivery. As one MHL noted, “First of all,
there needs to be fully immersed to understand the depth of our operations would change if it
was to come in.” The Awareness phase should create an ethics-first AI vision, focusing on
communicating how AI fits into the evolving landscape of mental health care, emphasizing its

141
role in enhancing rather than replacing human clinical judgment while demonstrating the
organization’s commitment to ethical implementation.
Organizations should establish regular communication channels through newsletters,
meetings, and forums to share successful mental health AI implementation case studies and
address ethical concerns using the principlism framework. This communication must highlight
specific ways AI will address current service delivery challenges while maintaining the crucial
human connection in mental health care. Leadership must maintain transparency about potential
challenges and their mitigation strategies, fostering an environment where staff feel informed
and included in the transformation process. To ensure effective awareness building,
organizations should implement clear metrics for measuring progress and engagement, as staff
understanding assessments and engagement levels provide insights into the effectiveness of
awareness campaigns.
Desire: Fostering Organizational Buy-In
This study found that MHLs recognize the importance of fostering genuine willingness
among staff to engage with AI technologies beyond mere compliance, and the findings revealed
the crucial role of early adopters in creating organizational desire. Desire, according to (Ball &
Hiatt, 2024), represents an individual’s personal choice to engage in and support the change
based on their understanding of its personal impact and motivating factors. As one MHL
observed,
There was a group in our clinic, early adopters where they were sort of excited to try it
out, the pilot that did it, they loved it. And then the best way is when you hear it from
someone else that you trust.

142
Organizations can leverage early adopters as an AI champion network to engage with all
potential users, not just to address operational concerns, but to strengthen their commitment to
ethical principles in AI adoption. By aligning AI implementation with healthcare’s established
principlism framework (autonomy, beneficence, non-maleficence, and justice), organizations can
help staff envision how AI can enhance patient care while maintaining ethical standards. This
approach can be achieved by developing an AI champion network, which creates a foundation
where early adopters serve as both technical ambassadors and ethical champions.
Developing an AI champions network begins with cultivating a genuine desire among
staff to implement AI ethically and responsibly. The champions network’s success depends on
creating an environment that promotes both psychological safety and ethical innovation.
Organizations should establish safe spaces for staff to experiment with AI tools while
continuously evaluating their alignment with ethical principles. Mentorship programs should pair
AI champions with hesitant colleagues, focusing not just on technical competency but on ethical
decision-making. This requires clear selection criteria for AI champions that emphasize both
technical aptitude and ethical awareness, along with protected time for innovation and ethical
consideration. Regular champion network meetings should include discussions of ethical
implications, while anonymous feedback mechanisms ensure continuous improvement in both
technical and ethical aspects. Organizations should develop reward systems that specifically
recognize ethical AI use, encouraging staff to prioritize responsible implementation that upholds
mental health care’s core values.
Knowledge: Implementing Ethical AI Learning
The knowledge stage of ADKAR focuses on providing staff with the specific
information, training, and education needed to understand how to effectively implement AI in

143
mental health care practice. The knowledge stage addresses how to change during the transition
and how to perform effectively in the future state (Ball & Hiatt, 2024). The participants
emphasized this critical need for comprehensive training and understanding. One highlighted,
“There’s a lot of training that gets involved. … You have to have enough people to understand
what’s going on to train them to continuously monitor that change management.” To address
knowledge requirements, organizations should establish a comprehensive ethical AI learning
framework that provides two essential types of knowledge: the user aspects of AI
implementation and the ethical frameworks guiding its use. This dual focus aligns with the
emphasis on both tactical knowledge (how to use the technology) and behavioral knowledge
(how to work within new ethical guidelines). Organizations should develop tiered training
programs that progress logically from fundamental AI concepts to advanced applications in
mental health care, incorporating ethical decision-making protocols based on the principlism
framework’s principles of autonomy, beneficence, non-maleficence, and justice.
The successful transfer of knowledge requires structured support systems, including a
cross-functional ethics committee and regular workshops on AI ethics and governance.
Following ADKAR’s framework guidance on knowledge transfer, organizations must invest in
robust learning management systems to track progress, assess competency, and ensure that
training translates into practical application. This systematic approach to knowledge building
ensures staff acquire both the technical skills and ethical understanding needed to effectively
implement AI in mental health care settings, emphasizing comprehensive knowledge
development for successful change implementation.

144
Ability: Fostering Competence Through Psychological Safety and Resource Support
While the knowledge stage provides the foundation for understanding AI tools, the ability
stage addresses the actual execution and development of proficiency in using these tools
effectively (Ball & Hiatt, 2024). The findings in this research highlighted the importance of
fostering psychological safety for staff, enabling them to develop competence in AI use in
mental health settings. The MHLs emphasized creating an environment where staff feel
comfortable asking questions and voicing concerns without fear of judgment. As one MHL
explained, “There is a level of psychological safety where [staff] can feel like they can come
forward and ask those questions.” highlighting the importance of a supportive environment for
developing new capabilities.
To build ability effectively, organizations should implement a structured safe practice
environment that allows staff to develop and demonstrate competence in using AI tools. This
environment includes sandbox settings where staff can practice with AI applications without risk
to patient care, converting theoretical knowledge into practical ability. Following Prosci’s
emphasis on demonstrated capability, organizations should implement a graduated practice
system that moves from basic AI interactions to more complex scenarios specific to mental
health care. This includes real-time technical support, structured practice scenarios, and clear
performance metrics to measure progress toward required competency levels.
In addition to psychological safety, resource allocation plays a critical role in the ability
phase. Successful implementation requires access to technical support resources that can address
staff questions and resolve issues in real time after the knowledge stage. Practice scenarios and
simulations, specifically designed for mental health applications, provide structured
opportunities for staff to gain practical experience in a safe environment. Performance support

145
tools, such as step-by-step guides or quick-reference materials, should be readily accessible,
allowing staff to navigate AI applications with confidence. Finally, establishing clear success
metrics will help staff understand progress and competency goals, offering a concrete way to
measure their development and the overall success of the organization’s AI initiatives.
By emphasizing psychological safety and ensuring ample resources, mental health organizations
can foster the ability required for staff to work effectively with AI. This approach builds
technical proficiency and nurtures a culture of openness and support, empowering staff to engage
with AI tools confidently and responsibly.
Reinforcement: Cultural Sustainability and Ethical Governance
For a meaningful lasting change, reinforcement is essential to maintain new behaviors
and ensure they are consistently supported. According to Ball and Hiatt (2024), reinforcement
focuses on solidifying change in organizational culture by recognizing achievements, creating
feedback loops, and addressing resistance. The participants have recognized the need for lasting
change that includes ongoing support and feedback. As one MHL put it, “You can implement
something, and people will just do it because they think it has to be done, but they’re not really
wholeheartedly wanting to [follow ethical guidelines].” This statement highlights the importance
of true engagement over mere compliance for sustainable change.
To establish a sustainable AI culture in mental health organizations, reinforcing ethical
practices and governance is critical as AI becomes an integrated part of providing services.
MHLs emphasize the importance of regular ethics audits and structured reviews to keep AI tools
aligned with core organizational values and ethical standards. Continuous feedback loops enable
staff to voice their experiences and concerns with AI systems, fostering a proactive approach to
addressing any ethical or operational issues. Programs that recognize and reward ethical AI use

146
can further embed these values, creating incentives for thoughtful engagement with AI tools.
Specific measures, such as clear audit schedules, well-defined recognition programs, and updated
performance metrics that include AI ethics, help reinforce these values daily. Regular cultural
assessments ensure that commitment to ethical AI remains strong, allowing mental health
organizations to build a resilient, ethically grounded culture that aligns with the vision of MHLs.
ADKAR Critical Success Factor
This study identified four critical success factors for AI implementation: strong
leadership commitment, accountability to uphold principlism bioethics, psychological safety in
organizations, and sound resource management. These interconnected elements provide the
foundation for responsible AI adoption in mental healthcare organizations.
Leadership Commitment
According to Zona and Thaib (2020), a leader plays a very important role in managing
relationships, coordinating change, adapting operational activities to strategies, building
structures, and developing rewards. The success of AI implementation in mental health
organizations is deeply rooted in strong leadership commitment. Visible executive sponsorship is
essential, as it signals to staff that AI initiatives are both valued and prioritized at the highest
levels. Leaders must maintain consistent ethical messaging, reinforcing a commitment to using
AI in ways that respect the dignity, privacy, and autonomy of individuals. Furthermore, the
allocation of resources, including budget and personnel, must be assured from the outset to
sustain AI initiatives. Regular progress reviews by leadership provide accountability, ensuring
AI projects adhere to ethical standards and achieve their intended impact.

147
Ethical Framework Integration
Integrating principlism ethical framework into AI practices is crucial for maintaining
trust and accountability in the healthcare setting. Aligning AI initiatives with the ethical
principles of principlism is a healthcare ethical framework that guides decision-making through
four core principles: autonomy (patient’s right to self-determination), beneficence (promoting
patient good), non-maleficence (avoiding harm), and justice (ensuring fairness in care delivery)
(Pasricha, 2022). Regular ethics committee meetings serve as a forum for discussing potential
challenges and aligning AI use with organizational values. Documented decision-making
protocols enhance transparency, allowing stakeholders to understand how ethical considerations
influence AI-related decisions. Continuous ethical assessment ensures that AI applications
remain sensitive to ethical issues as they evolve, reinforcing a responsible and principled
approach to AI in mental health settings.
Psychological Safety Measures
Creating an environment of psychological safety is key to fostering open, honest
engagement with AI tools. Anonymous feedback channels enable staff to voice concerns without
fear of retribution, promoting transparency and early identification of potential issues (Awar et
al., 2023). Organizations should establish a no-blame learning environment that encourages
individuals to report and learn from mistakes or challenges in AI implementation. Clear
escalation procedures provide a structured way to address ethical or operational concerns as they
arise. Regular assessments of the organizational safety climate ensure that employees feel secure
and supported, which is vital for the successful and responsible adoption of AI in mental health
services.

148
Resource Management
Effective resource management is foundational for sustainable AI implementation. A
dedicated budget for AI initiatives is necessary to support ongoing development, updates, and
troubleshooting of AI systems (Marotta & Au, 2022). Organizations should protect time for
training to ensure staff have the opportunity to develop the skills needed to interact with and
utilize AI effectively. Organizations should also establish a strong technical support
infrastructure that further aids in resolving issues and minimizing downtime and frustration.
Finally, organizations should invest in ongoing AI-related education that allows the organization
to keep pace with advancements in AI, ensuring that the workforce remains informed and
equipped to use AI in safe and ethically sound ways.
Environment Component: External Influences and Strategic Responses
The adoption of AI in mental health care is significantly influenced by a complex web of
external factors that shape organizational decision-making and implementation strategies. This
study found that MHLs must navigate a multifaceted environment encompassing regulatory
requirements, funding constraints, and stakeholder expectations. In this environment, several key
factors emerge as particularly influential in AI adoption decisions.
Government policies and professional guidelines serve as critical environmental factors
shaping AI adoption (Neumann et al., 2022). MHLs emphasized the importance of aligning AI
initiatives with regulatory frameworks, even as these frameworks continue to evolve. The current
lack of specific AI guidelines has led many organizations to look to professional associations for
guidance, creating a situation where implementation strategies must remain flexible enough to
adapt to emerging regulatory requirements.

149
Funding and economic considerations emerged as significant environmental factors in
program and technology implementation, particularly for community mental health organizations
(Razzouk, 2023). Historical precedent suggests potential challenges in obtaining funding for
technological implementation, as evidenced by one participant’s observation about EHRs: “EHR
[and electronic medical record] technology was conceived around the 1960s, yet we didn’t get
funding to implement it until the 2010s.” This lag in funding access raises significant equity
concerns (Zabelski et al., 2024), as organizations with greater financial resources can invest in
more advanced AI tools, potentially widening the gap in service quality and accessibility
between well-funded and resource-constrained organizations.
The influence of public managed care organizations, particularly Medicare, emerged as a
crucial environmental factor. Medicare’s role as the largest healthcare payer in the United States
creates a ripple effect throughout the industry. As private insurers increasingly use AI for claims
processing (Jaffe, 2023), community mental health organizations face both pressure and
opportunities in their AI adoption journey. The growing prevalence of private insurance plans in
providing Medicare services through programs like Medicare Advantage adds another layer of
complexity to this dynamic (Thomson et al., 2020).
Patient perspectives and public perception also emerged as significant environmental
factors influencing AI adoption decisions. MHLs observed varying levels of receptivity among
different patient groups, with factors such as technological literacy, cultural background, and
generational differences playing important roles in acceptance and engagement with AIenhanced services.

150
Environmental Strategy Recommendations
Based on these findings, I recommend a comprehensive approach to addressing
environmental challenges that focuses on strategic engagement with external stakeholders and
systematic resource management. Mental health organizations must develop adaptive strategies
that balance AI implementation with regulatory compliance, resource limitations, and ethical
care standards in an evolving healthcare environment. We identified patterns suggesting that
successful AI adoption requires organizations to simultaneously navigate relationships with
multiple stakeholders, including government agencies, funding sources, insurance providers, and
patient communities. This approach encompasses three key areas: strategic funding and policy
engagement, collaborative networks, and patient-centered implementation strategies. Each area
addresses specific environmental challenges identified in this study while recognizing the
interconnected nature of these challenges in mental health care settings. These recommendations
are particularly relevant for community mental health organizations that must balance innovation
with fiscal constraints and regulatory requirements. By addressing these three key areas
comprehensively, organizations can build resilient systems that support responsible AI adoption
while maintaining their commitment to accessible, high-quality mental health care.
Strategic Funding and Regulatory Compliance
Organizations can consider establishing a public healthcare-aligned AI documentation
initiative to address the challenges posed by Medicare and other public healthcare systems. This
initiative should focus on creating standardized documentation processes that align with current
requirements while maintaining flexibility to adapt to future changes. As major public healthcare
systems like Medicare and Medicaid increasingly utilize AI for risk assessment and claims
processing, community mental health organizations must align their AI implementation

151
strategies accordingly. The research findings revealed that MHLs are particularly concerned
about maintaining compliance while maximizing reimbursement opportunities.
This initiative begins with forming a dedicated Medicare compliance team to review
current AI documentation requirements and create standardized templates that align with
Medicare documentation rules. Organizations may want to systematically track common
documentation challenges that could be addressed through AI solutions, providing data-driven
justification for technology investments. Additionally, proactive discussions with EHR vendors
about AI integration capabilities can help organizations prepare for future implementation
requirements while ensuring system compatibility.
The MHLs in this study noted that historical delays in technology funding, such as with
EHR implementation, require organizations to find creative ways to maximize limited resources
while maintaining regulatory compliance. The implementation of blockchain technology may
offer a promising solution for managing the complex requirements of multiple stakeholders in
the mental healthcare environment (Mercer & Khurshid, 2021). This decentralized approach can
help organizations meet Medicare and Medicaid documentation requirements while maintaining
efficient operations. Through blockchain’s transparent audit trails (Regueiro et al., 2021),
organizations can demonstrate compliance with regulatory requirements while reducing
administrative overhead. The technology’s ability to create immutable records particularly
benefits community mental health organizations dealing with strict public healthcare
documentation requirements and multiple funding sources (Ettaloui et al., 2023). The benefits of
this approach include reduced documentation errors and improved Medicare reimbursement rates
through better compliance. Organizations will gain a clearer understanding of current and
emerging compliance requirements, positioning them to adapt more quickly to changes in the

152
public healthcare landscape. This proactive stance on documentation and compliance creates a
strong foundation for future AI implementations while addressing one of the most significant
environmental pressures community mental health organizations face.
Community Mental Health Alliance Network
The participants recognize that collaboration, not competition, is critical to successful AI
implementation in mental health organizations. To facilitate this collaboration, one
recommendation is to create a community mental health alliance network with two main
objectives: fostering structured collaborative relationships between organizations and building
diverse, unbiased AI training datasets through collective zero-knowledge proof data sharing.
The implementation begins by partnering with two to three organizations of different
sizes. Organizations with existing alliances should establish dedicated AI subcommittees.
Through monthly virtual meetings, partners share experiences and solutions, while a shared
repository captures AI resources and lessons learned. This systematic documentation of
challenges and solutions builds a collective knowledge base serving all participants.
The network delivers immediate benefits through shared learning experiences and cost
reduction through resource sharing. It also creates a stronger collective voice for advocacy in the
mental health sector. This collaborative structure proves particularly valuable for resourcelimited community mental health organizations, enabling them to leverage shared expertise for
more effective and responsible AI adoption. By working together, organizations avoid
duplicating efforts and strengthen their overall approach to AI implementation.
Patient Engagement and Public Trust-Building
Patient engagement and public trust-building initiatives should form the third pillar of
environmental strategy, requiring a thoughtful approach that acknowledges the sensitive nature

153
of mental health care and the varying levels of AI receptivity among different patient
populations. This study found that successful AI adoption depends significantly on patient
acceptance and trust, particularly given the intimate nature of mental health treatment and the
heightened privacy concerns in this field.
Organizations should implement a multilayered communication and engagement strategy
that begins before AI implementation and continues throughout the service delivery process. At
the pre-implementation stage, organizations should partner with community mental health
advocacy groups like the National Alliance on Mental Illness to conduct community forums and
focus groups that bring together diverse patient populations to understand their concerns,
preferences, and needs regarding AI in mental health care. These discussions should specifically
address cultural perspectives on mental health treatment, varying comfort levels with technology,
and specific privacy concerns that may affect different communities.
Cultural competency must be embedded in AI implementation through ongoing dialogue
with different community groups and adaptations to align with cultural norms and values.
Organizations should provide tiered support systems offering varying levels of assistance in
engaging with AI-enhanced services, ensuring that technological barriers do not create new
disparities in access to mental health care. Through these engagement strategies, organizations
can build the trust necessary for successful AI adoption while maintaining their commitment to
accessible, high-quality mental health care for all patients.
The research suggests that successful navigation of environmental challenges requires
organizations to maintain flexibility while building strong collaborative relationships. By
developing robust documentation systems, fostering inter-organizational collaboration, and

154
maintaining focus on patient needs, mental health organizations can better position themselves to
leverage AI technologies while managing external constraints and pressures.
Directions for Future Research
As the integration of AI in mental health care continues to evolve, several critical areas
emerge as priorities for future research. These directions address the current gaps in
understanding and pave the way for more effective and ethical implementation of AI
technologies in mental health services.
Long-Term Effects of AI Adoption on Patient Outcomes
A pressing need in the field is for longitudinal studies that assess the impact of AIenhanced mental health care on patient outcomes over an extended period. While current
research provides insights into the initial stages of AI implementation, the true value and
potential risks of these technologies can only be fully understood through long-term observation
and analysis. Future studies should track patients receiving AI-augmented care over several
years, comparing their outcomes with those receiving traditional care. These longitudinal studies
could examine various factors, including symptom reduction, recovery rates, relapse prevention,
and overall quality of life improvements. Additionally, researchers should investigate potential
unintended consequences or side effects of long-term exposure to AI-driven interventions. Such
comprehensive, long-term studies would validate the presumed benefits of AI in mental health
care and help in refining and optimizing AI systems for better patient care over time. The
findings from these studies could guide policy decisions, inform best practices, and ensure that
the integration of AI truly serves the best interests of patients in the long run.

155
Comparative Studies Across Different Mental Health Settings
The effectiveness and challenges of AI adoption likely vary significantly across different
mental health care settings. Future research should focus on comparative studies that examine AI
implementation and outcomes across various contexts, such as inpatient facilities, outpatient
clinics, community mental health centers, and private practices. These studies should investigate
how factors like resource availability, patient populations, organizational structures, and
technological infrastructure influence the adoption and efficacy of AI technologies. For instance,
researchers could explore how AI-driven diagnostic tools perform in high-pressure emergency
psychiatric settings compared to routine outpatient visits. Similarly, studies could examine the
effectiveness of AI-assisted therapy in community clinics serving diverse populations versus
specialized private practices. Such comparative research would provide crucial insights into the
adaptability and scalability of AI solutions across different mental health care environments. The
findings would be instrumental in developing tailored implementation strategies that account for
each setting’s challenges and opportunities. This nuanced understanding would enable MHLs
and policymakers to make more informed decisions about AI adoption, ensuring that these
technologies are deployed in ways that maximize benefits while minimizing risks across the
entire spectrum of mental health care services.
Development of AI-Specific Ethical Frameworks for Mental Health Care
As AI becomes more prevalent in mental health care, there is an urgent need for research
focused on developing and refining ethical frameworks specifically tailored to this context.
While general ethical guidelines for AI and healthcare exist, the sensitive nature of mental health
data and the potential vulnerability of mental health patients necessitate specialized ethical
considerations. Future studies should explore the ethical implications of AI use in areas such as

156
patient privacy, informed consent in AI-assisted diagnoses, the potential for AI bias in mental
health assessments, and the impact of AI on the therapeutic relationship. Researchers should
collaborate with ethicists, mental health professionals, AI developers, legal experts, and patient
advocates to create comprehensive ethical guidelines that address the nuances of AI applications
in mental health. This research could involve case studies of ethical dilemmas in AI
implementation, surveys of stakeholder perspectives on ethical issues, and the development and
testing of ethical decision-making models for AI use in mental health care. The resulting
frameworks should provide clear guidance on issues such as data handling, algorithm
transparency, accountability for AI-driven decisions, and safeguarding patient autonomy. By
establishing robust, context-specific ethical frameworks, this research would help ensure that the
advancement of AI in mental health care aligns with core ethical principles and prioritizes patient
well-being.
Exploration of Patient Experiences With AI-Augmented Mental Health Services
A critical area for future research lies in conducting in-depth studies of patient
perspectives and experiences with AI in mental health care. While much of the current focus has
been on the viewpoints of healthcare providers and the technical aspects of AI implementation,
understanding the patient’s experience is crucial for developing truly patient-centered AI
solutions. Future studies should employ a mix of qualitative and quantitative methods to explore
how patients perceive, interact with, and benefit from AI technologies in their mental health care
journey. Researchers could conduct interviews and focus groups to gather rich, detailed accounts
of patient experiences with AI-assisted therapy, diagnostic tools, and monitoring systems.
Additionally, large-scale surveys could be developed to collect quantitative data on
patient satisfaction, perceived effectiveness, and comfort levels with various AI applications in

157
mental health care. These studies should investigate patient preferences regarding the balance
between AI and human interaction in their care, their trust in AI-generated insights, and any
concerns they may have about privacy or the depersonalization of care. Researchers should also
examine how different patient demographics—considering factors such as age, cultural
background, and type of mental health condition—might influence attitudes toward and
experiences with AI in mental health care. The insights gained from these studies would be
invaluable in informing the design and implementation of more acceptable, effective, and
patient-centered AI interventions. By prioritizing the patient voice in AI development, this
research would help ensure that technological advancements in mental health care truly align
with patient needs, preferences, and values.
To summarize, these directions for future research—examining long-term patient
outcomes, comparing AI adoption across different settings, developing specialized ethical
frameworks, and exploring patient experiences—are critical for advancing our understanding of
AI in mental health care. By pursuing these research avenues, the field can work toward more
effective, ethical, and patient-centered integration of AI technologies in mental health services.
As AI continues to evolve and shape the landscape of mental health care, ongoing research in
these areas will be essential for realizing the full potential of AI to improve mental health
outcomes while addressing challenges and ethical concerns. The findings from these future
studies will contribute to the body of scientific knowledge and provide guidance for
policymakers, healthcare providers, and technology developers in shaping the future of AIaugmented mental health care.

158
Conclusion
The integration of AI in mental health care represents a transformative shift in the field,
offering unprecedented opportunities to enhance patient care, improve accessibility, and
potentially revolutionize treatment approaches. This study has explored the multifaceted
landscape of AI adoption in mental health organizations through the lens of the TOE framework,
providing insights into the perceptions, challenges, and potential pathways for responsible
implementation.
Our examination of the MHLs’ perceptions of AI technologies revealed a nuanced
landscape characterized by cautious optimism. Leaders recognize the immense potential of AI to
personalize treatment strategies, enhance predictive analytics for early intervention, and improve
operational efficiency. The promise of AI in democratizing access to mental health support and
potentially reducing societal stigma associated with mental health issues is particularly
noteworthy. However, this optimism is tempered by significant concerns regarding the accuracy,
reliability, and potential biases of AI systems. The complexity of mental health work, with its
inherent nuances and the critical importance of human empathy, poses unique challenges for AI
integration that cannot be overlooked.
The organizational factors enabling responsible AI adoption emerged as critical elements
in this study. Our findings underscore the importance of cultivating an organizational culture that
prioritizes ethical considerations, fosters psychological safety, and promotes transparent
communication about AI initiatives. The role of leadership in setting a clear vision for AI
adoption, allocating resources for ethical governance, and nurturing a culture of innovation
cannot be overstated. The influence of early adopters in shaping organizational attitudes toward

159
AI highlights the potential for strategically leveraging these champions to facilitate broader
acceptance and responsible implementation.
External factors significantly influence the decision-making process for AI adoption in
mental health care. The evolving regulatory landscape, funding constraints, and the influence of
managed care organizations all play crucial roles in shaping the environment in which AI is
adopted. The competitive pressures and collaborative opportunities in the healthcare ecosystem
create a complex backdrop against which organizations must navigate their AI strategies. Patient
perspectives and public trust emerge as critical factors, emphasizing the need for transparent,
ethical, and patient-centered approaches to AI implementation.
Looking to the future, several key areas demand further research and attention.
Longitudinal studies examining the long-term effects of AI adoption on patient outcomes are
essential to validate the presumed benefits and identify any unforeseen consequences.
Comparative studies across different mental health settings will provide insights for tailoring
implementation strategies to diverse contexts. Developing AI-specific ethical frameworks for
mental health care remains a critical priority to ensure that AI adoption aligns with the core
values and ethical standards of the mental health profession.
The practical recommendations outlined in this study, structured around the ADKAR
change management framework, offer a roadmap for mental health organizations embarking on
the journey of AI adoption. From creating awareness and fostering a desire for change to
building knowledge, developing abilities, and reinforcing new practices, these guidelines provide
a comprehensive approach to responsible AI implementation. The emphasis on maintaining
human-centered care while leveraging AI’s capabilities reflects a balanced approach that respects
the nature of mental health work.

160
The adoption of AI in mental health care stands at a critical juncture. The potential
benefits are immense, offering hope for more effective, accessible, and personalized mental
health services. However, the challenges and ethical considerations are equally significant.
Responsible AI adoption in mental health care is not merely a technological challenge but a
multifaceted endeavor that requires careful navigation of organizational, ethical, and societal
considerations.
Moving forward, it is imperative that the mental health community approaches AI
adoption with a commitment to ethical practice, patient-centered care, and continuous learning.
By fostering collaboration between mental health professionals, AI developers, policymakers,
and patients can work toward a future where AI enhances and supports high-quality, ethical, and
accessible mental health care. The journey ahead is complex, but with thoughtful implementation
and ongoing research, AI could significantly improve mental health outcomes and contribute to a
more effective and compassionate mental health care system.
The future of mental health care lies not in choosing between human expertise and AI but
in finding the optimal synergy between the two. Continuing to explore and refine the role of AI
in mental health care will require remaining grounded in the core values of the profession:
empathy, ethical practice, and the unwavering commitment to improving the lives of those
struggling with mental health challenges. Doing so will harness the power of AI to create a more
responsive, effective, and inclusive mental health care system for all (Figure 2).

161
Figure 2
Model for Responsible AI Adoption

162
References
Aali, G., Kariotis, T., & Shokraneh, F. (2020). Avatar therapy for people with schizophrenia or
related disorders. The Cochrane Library, 2020(5), Article CD011898
https://doi.org/10.1002/14651858.CD011898.pub2
Abd-Alrazaq, A., Alajlani, M., Alalwan, A. A., Bewick, B. M., Gardner, P., & Househ, M.
(2019). An overview of the features of chatbots in mental health: A scoping review.
International Journal of Medical Informatics, 132, Article 103978.
https://doi.org/10.1016/j.ijmedinf.2019.103978
Ackerman, S. J., & Hilsenroth, M. J. (2003). A review of therapist characteristics and techniques
positively impacting the therapeutic alliance. Clinical Psychology Review, 23(1), 1–33.
https://doi.org/10.1016/s0272-7358(02)00146-0
Adams, D. (2024, January 30). Updating HIPAA security to respond to artificial intelligence.
Journal of AHIMA. https://journal.ahima.org/page/updating-hipaa-security-to-respond-toartificial-intelligence
Adelakun, N. B. O., Majekodunmi, N. T. G., & Akintoye, N. O. S. (2024). AI and ethical
accounting: Navigating challenges and opportunities. International Journal of Advanced
Economics, 6(6), 224–241. https://doi.org/10.51594/ijae.v6i6.1230
Adigozel, O., Awad, N., Nayak, S., & Rabelo, I. (2023). Are payers and providers ready for
digital change in health care? BCG Global.
https://www.bcg.com/publications/2023/navigating-digital-healthcare-change
Adler-Milstein, J., Redelmeier, D. A., & Wachter, R. M. (2024). The limits of clinician vigilance
as an AI safety bulwark. Journal of the American Medical Association, 331(14), 1173–
1174. https://doi.org/10.1001/jama.2024.3620

163
Agbo, C. C., Mahmoud, Q. H., & Eklund, J. M. (2019). Blockchain technology in healthcare: A
systematic review. Healthcare, 7(2), 56. https://doi.org/10.3390/healthcare7020056
Ahmed, M. I., Spooner, B., Isherwood, J., Lane, M., Orrock, E., & Dennison, A. (2023). A
systematic review of the barriers to the implementation of artificial intelligence in
healthcare. Cureus, 15(10), Article e46454. https://doi.org/10.7759/cureus.46454
Ahsan, S. M. M., Dhungel, A., Chowdhury, M., Hasan, M. S., & Hoque, T. (2024). Hardware
accelerators for artificial intelligence. Cornell University.
https://doi.org/10.48550/arxiv.2411.13717
Akbarighatar, P. (2024). Operationalizing responsible AI principles through responsible AI
capabilities. AI and Ethics. https://doi.org/10.1007/s43681-024-00524-4
Al-Abdullah, M., Alsmadi, I., AlAbdullah, R., & Farkas, B. (2020). Designing privacy-friendly
data repositories: a framework for a blockchain that follows the GDPR. Digital Policy
Regulation and Governance, 22(5/6), 389–411. https://doi.org/10.1108/dprg-04-2020-
0050
Alam, A., & Prybutok, V. R. (2024). Use of responsible artificial intelligence to predict health
insurance claims in the USA using machine learning algorithms. Exploration of Digital
Health Technologies, 2024(2), 30–45. https://doi.org/10.37349/edht.2024.00009
Alami, H., Lehoux, P., Denis, J., Motulsky, A., Petitgand, C., Savoldelli, M., Rouquet, R.,
Gagnon, M., Roy, D., & Fortin, J. (2020). Organizational readiness for artificial
intelligence in health care: Insights for decision-making and practice. Journal of Health
Organisation and Management, 35(1), 106–114. https://doi.org/10.1108/jhom-03-2020-
0074

164
Alhur, A. (2024). Overcoming electronic medical records adoption challenges in Saudi Arabia.
Cureus, 16(2), Article e53827. https://doi.org/10.7759/cureus.53827
Alhuwaydi, A. (2024). Exploring the role of artificial intelligence in mental healthcare: Current
trends and future directions – A narrative review for a comprehensive insight. Risk
Management and Healthcare Policy, 17, 1339–1348.
https://doi.org/10.2147/rmhp.s461562
Ali, K., Garcia, A., & Vadsariya, A. (2024). Impact of the AI dependency revolution on both
physical and mental health. Journal of Strategic Innovation and Sustainability, 19(2).
https://doi.org/10.33423/jsis.v19i2.7006
Allyn, B. (2023, July 2). Hollywood actors are pushing back against studios using AI to clone
them. NPR. https://www.npr.org/2023/07/02/1185684635/hollywood-actors-are-pushingback-against-studios-using-ai-to-clone-them
Alonso, S. G., De La Torre Díez, I., Hamrioui, S., López-Coronado, M., Barreno, D. C.,
Nozaleda, L. M., & Franco, M. (2018). Data mining algorithms and techniques in mental
health: A systematic review. Journal of Medical Systems, 42(9), Article 161.
https://doi.org/10.1007/s10916-018-1018-2
Alto, V. (2023). Modern generative AI with ChatGPT and OpenAI models: Leverage the
capabilities of OpenAI’s LLM for productivity and innovation with GPT3 and GPT4.
Packt Publishing.
American Psychological Association. (2024, November 21). Artificial intelligence in mental
health care. https://www.apa.org/practice/artificial-intelligence-mental-health-care
Andero. (2023, July 5). How the world’s leading blockchain company Guardtime turns trust into
digital truth from its Tallinn office. Invest in Estonia. https://investinestonia.com/how-

165
the-worlds-leading-blockchain-company-guardtime-turns-trust-into-digital-truth-from-itstallinn-office/
Anyanwu, E. C., Okongwu, C. C., Olorunsogo, T. O., Ayo-Farai, O., Osasona, F., &
Daraojimba, O. D. (2024). Artificial intelligence in healthcare: A review of ethical
dilemmas and practical applications. International Medical Science Research Journal,
4(2), 126–140. https://doi.org/10.51594/imsrj.v4i2.755
Appenzeller, A., Hornung, M., Kadow, T., Krempel, E., & Beyerer, J. (2022). Sovereign digital
consent through privacy impact quantification and dynamic consent. Technologies, 10,
35. https://mdpi-res.com/technologies/technologies-10-
00035/article_deploy/technologies-10-00035-v2.pdf?version=1645519830
Arigbabu, A. T., Olaniyi, O. O., Adigwe, C. S., Adebiyi, O. O., & Ajayi, S. A. (2024). Data
governance in AI-enabled healthcare systems: A case of the project Nightingale. Asian
Journal of Research in Computer Science, 17(5), 85–107.
https://doi.org/10.9734/ajrcos/2024/v17i5441
Atlam, H. F., Shafik, M., Kurugollu, F., & Elkelany, Y. (2022). Emotions in mental healthcare
and psychological interventions: Towards an inventive emotions recognition framework
using AI. In Advances in Transdisciplinary Engineering, 25, 317–322
https://doi.org/10.3233/atde220609
Awar, D. T. A., Abdulla, F. I. M., Bakhamis, S. a. A., Rashid, M. a. A., Saleh, A. A., Mamalac,
A. D., & Laja, N. (2023). Fostering a safe psychological environment and encouraging
speak-up culture in primary care setups. International Journal of Research in Medical
Sciences, 11(12), 4583–4589. https://doi.org/10.18203/2320-6012.ijrms20233740

166
Bai, T., Hu, Y., He, J., Fan, H., & An, Z. (2022). Health-ZKIDM: a healthcare identity system
based on fabric blockchain and Zero-Knowledge Proof. Sensors, 22(20), Article 7716.
https://doi.org/10.3390/s22207716
Baker, J. (2011). The technology–organization–environment framework. In Y. K. Dwivedi, M.
R. Wade, & S. L. Schneberger (Eds.), Information systems theory: Explaining and
predicting our digital society (Vol. 1, pp. 231–245). Springer.
https://doi.org/10.1007/978-1-4419-6108-2_12
Bal, S. B. (2008). An introduction to medical malpractice in the United States. Clinical
Orthopaedics and Related Research, 467(2), 339–347. https://doi.org/10.1007/s11999-
008-0636-2
Ball, K., & Hiatt, J. (2024). The ADKAR advantage: Your new lens for successful change. Prosci
Publications.
Barkan, R., Ayal, S., & Ariely, D. (2015). Ethical dissonance, justifications, and moral behavior.
Current Opinion in Psychology, 6, 157–161.
https://doi.org/10.1016/j.copsyc.2015.08.001
Battaglia, J. (2019). Supportive psychotherapy and importance of therapeutic alliance.
Psychiatric News. https://doi.org/10.1176/appi.pn.2019.10b24
Baum, S. D. (2016). On the promotion of safe and socially beneficial artificial intelligence. AI &
Society, 32(4), 543–551. https://doi.org/10.1007/s00146-016-0677-0
Beauchamp, T. L., & Childress, J. F. (1979). Principles of biomedical ethics. Oxford University
Press, USA.
Beauchamp, T. L., & Childress, J. F. (2012). Principles of biomedical ethics (7th ed.). Oxford
University Press

167
Belani, S., Tiarks, G. C., Mookerjee, N., & Rajput, V. (2021). “I agree to disagree”: Comparative
ethical and legal analysis of big data and genomics for privacy, consent, and ownership.
Cureus, 13(10), Article e18736. https://doi.org/10.7759/cureus.18736
Bench‐Capon, T., Araszkiewicz, M., Ashley, K. D., Atkinson, K., Bex, F., Borges, F., Bourcier,
D., Bourgine, P., Conrad, J. G., Francesconi, E., Gordon, T. F., Governatori, G., Leidner,
J. L., Lewis, D., Loui, R. P., McCarty, L. T., Prakken, H., Schilder, F., Schweighofer, E.,
. . . Wyner, A. (2012). A history of AI and Law in 50 papers: 25 years of the international
conference on AI and Law. Artificial Intelligence and Law, 20(3), 215–319.
https://doi.org/10.1007/s10506-012-9131-x
Bernardone, L. (2024, November). ‘Please die’: Google’s AI abuses grad student. Information
Age. https://ia.acs.org.au/article/2024/-please-die---google-s-ai-abuses-grad-student.html
Berre, A. J., Tsalgatidou, A., Francalanci, C., Ivanov, T., Pariente-Lobo, T., Ruiz-Saiz, R.,
Novalija, I., & Grobelnik, M. (2022). Big data and AI pipeline framework: Technology
analysis from a benchmarking perspective. In E. Curry, S. Auer, A. J. Berre, A. Metzger,
M. S. Perez & S. Zillner (Eds.), Technologies and applications for big data value (pp.
63–88). Springer https://doi.org/10.1007/978-3-030-78307-5_4
Bhargava, C., & Sharma, P. K. (2021). Artificial intelligence: Fundamentals and Applications.
CRC Press.
Bhatia, K., Tanch, J., Chen, E. S., & Sarkar, I. N. (2020). Applying FAIR principles to improve
data searchability of emergency department datasets: A case study for HCUP-SEDD.
Methods of Information in Medicine, 59(01), 048–056. https://doi.org/10.1055/s-0040-
1712510

168
Bickman, L. (2020). Improving mental health services: A 50-year journey from randomized
experiments to artificial intelligence and precision mental health. Administration and
Policy in Mental Health, 47(5), 795–843. https://doi.org/10.1007/s10488-020-01065-8
Biswas, A., & Talukdar, W. (2024). Intelligent clinical documentation: Harnessing generative AI
for patient-centric clinical note generation. International Journal of Innovative Science
and Research Technology, 9(5), 994–1008.
https://doi.org/10.38124/ijisrt/ijisrt24may1483
BlackDeer, A. A., & Beeler, S. (2024). Decolonizing big data: addressing data colonialism in
social work’s grand challenges. Journal of Ethnic & Cultural Diversity in Social Work,
1–7. https://doi.org/10.1080/15313204.2024.2321440
Bordelon, B. (2023, June 21). Schumer launches new phase in push for AI bill. POLITICO.
https://www.politico.com/news/2023/06/21/schumer-launches-new-phase-in-push-for-aibill-00102871
Borghouts, J., Eikey, E. V., Mark, G., De Leon, C., Schueller, S. M., Schneider, M., Stadnick, N.
A., Zheng, K., Mukamel, D. B., & Sorkin, D. H. (2021). Barriers to and facilitators of
user engagement with digital mental health interventions: Systematic review. Journal of
Medical Internet Research, 23(3), Article e24387. https://doi.org/10.2196/24387
Bostrom, N. (2015). Superintelligence: Paths, dangers, strategies. Oxford University Press.
Bradley, S. (2023, August 24). I tried the Woebot AI therapy app to see if it would help my
anxiety and OCD. Verywell Mind. https://www.verywellmind.com/i-tried-woebot-aitherapy-app-review-7569025

169
Bright, R. A. (2000). What is FDA doing about medical errors and adverse events related to
medical devices? Pharmacoepidemiology and Drug Safety, 9(5), 437–440.
https://doi.org/10.1002/1099-1557(200009/10)9:5
Browne, G. (2022, October 1). The problem with mental health bots. WIRED UK.
https://www.wired.co.uk/article/mental-health-chatbots
Buckwalter, J. G., & California Psychological Association. (2023, July 21). Artificial intelligence
(AI) and psychology: The ethics of the future are needed now [Webinar]. Friday Webinar
Series.
Bucky, S. F., Callan, J. E., & Stricker, G. (2013). Ethical and legal issues for mental health
professionals. Routledge. https://doi.org/10.4324/9781315821160
Bukaty, P. (2019). The California Consumer Privacy Act (CCPA).
https://doi.org/10.2307/j.ctvjghvnn
Buruk, B., Ekmekçi, P. E., & Arda, B. (2020). A critical perspective on guidelines for
responsible and trustworthy artificial intelligence. Medicine Health Care and Philosophy,
23(3), 387–399. https://doi.org/10.1007/s11019-020-09948-1
Caetano, R. (2011). There is potential for cultural and social bias in DSM-V. Addiction, 106(5),
885–887. https://doi.org/10.1111/j.1360-0443.2010.03308.x
Cahn, S. M., & Markie, P. J. (2011). Ethics: History, theory, and contemporary issues. Oxford
University Press. http://ci.nii.ac.jp/ncid/BB07954360
Caldarini, G., Jaf, S., & Mcgarry, K. (2022). A literature survey of recent advances in chatbots.
Information, 13(1), Article 41. https://doi.org/10.3390/info13010041

170
California Legislative Information. (2024a, September 28). Bill Text - AB-3030 Health care
services: artificial intelligence. Retrieved December 9, 2024, from
https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240AB3030
California Legislative Information. (2024b). SB-1120 Health care coverage: utilization review.
https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202320240SB1120
Cannarsa, M. (2021). Ethics guidelines for trustworthy AI. In L. A. DiMatteo, A. Janssen, P.
Ortolani, F. de Elizalde, M. Cannarsa, & M. Dukovic (Eds.), The Cambridge handbook of
lawyering in the digital age (pp. 283–297). Cambridge University Press.
https://doi.org/10.1017/9781108936040.022
Cao, S., Jiang, W., Yang, B., & Zhang, A. L. (2023). How to talk when a machine is listening:
Corporate disclosure in the age of AI. Review of Financial Studies, 36(9), 3603–3642.
https://doi.org/10.1093/rfs/hhad021
Carr, S. (2020). ‘AI gone mental’: Engagement and ethics in data-driven technology for mental
health. Journal of Mental Health, 29(2), 125–130.
https://doi.org/10.1080/09638237.2020.1714011
Carta, S., Podda, A. S., Recupero, D. R., & Stanciu, M. M. (2022, October 4–8). Explainable AI
for financial forecasting [Paper presentation]. Machine Learning, Optimization, and Data
Science: 7th International Conference, Grasmere, UK. https://doi.org/10.1007/978-3-030-
95470-3_5
Cecula, P., Yu, J., Dawoodbhoy, F. M., Delaney, J., Tan, J., Peacock, I., & Cox, B. (2021).
Applications of artificial intelligence to improve patient flow on mental health inpatient
units - Narrative literature review. Heliyon, 7(4), Article e06626.
https://doi.org/10.1016/j.heliyon.2021.e06626

171
Cesar, L. B., Callejo, M. Á. M., & Cira, C. (2023). BERT (Bidirectional Encoder
Representations from Transformers) for Missing Data Imputation in Solar Irradiance
Time Series. Engineering Proceedings, 39(1), Article 26.
https://doi.org/10.3390/engproc2023039026
Chatterjee, S., Rana, N. P., Dwivedi, Y. K., & Baabdullah, A. M. (2021). Understanding AI
adoption in manufacturing and production firms using an integrated TAM-TOE model.
Technological Forecasting and Social Change, 170, Article 120880.
https://doi.org/10.1016/j.techfore.2021.120880
Chen, Y., Huerta, E. A., Duarte, J., Harris, P., Katz, D. S., Neubauer, M. S., Diaz, D., Mokhtar,
F., Kansal, R., Park, S. E., Kindratenko, V. V., Zhao, Z., & Rusack, R. (2022). A FAIR
and AI-ready Higgs boson decay dataset. Scientific Data, 9(1).
https://doi.org/10.1038/s41597-021-01109-0
Childress, J. F. (1983). Principles of biomedical ethics. Oxford University Press, USA.
Chiruvella, V., & Guddati, A. K. (2021). Ethical issues in patient data ownership. Interactive
Journal of Medical Research, 10(2), Article e22269. https://doi.org/10.2196/22269
Cirqueira, D., Helfert, M., & Bezbradica, M. (2021, July 24–29). Towards design principles for
user-centric explainable AI in fraud detection [Paper presentation]. Artificial Intelligence
in HCI: Second International Conference, online. https://doi.org/10.1007/978-3-030-
77772-2_2
Clark, T. R. (2020). The 4 stages of psychological safety: Defining the path to inclusion and
innovation. Berrett-Koehler Publishers.
Cogito Corporation. (2022, December 19). Customer experience improvement.
https://cogitocorp.com/solutions/improve-my-cx/

172
Coglianese, C. (2002). Empirical analysis and administrative law. SSRN Electronic Journal.
https://doi.org/10.2139/ssrn.327520
CompanionMx. (2019, September 26). The U.S. Department of Defense selects
CompanionMXTM for suicide prevention study in active duty naval personnel. Business
Wire. https://www.businesswire.com/news/home/20190926005130/en/The-U.S.-
Department-of-Defense-Selects-CompanionMx%E2%84%A2-for-Suicide-PreventionStudy-in-Active-Duty-Naval-Personnel
Cooper, M. (2023). Open Source is good for AI, but is AI good for Open Source? Itnow, 65(2),
50–51. https://doi.org/10.1093/combul/bwad062
Cordina, J., Gilbert, G., Mph, N. G., & Kumar, R. (2019). Next-generation member engagement
during the care journey. McKinsey & Company.
https://www.mckinsey.com/industries/healthcare/our-insights/next-generation-memberengagement-during-the-care-journey
Couldry, N., & Mejias, U. A. (2018). Data colonialism: Rethinking big data’s relation to the
contemporary subject. Television & New Media, 20(4), 336–349.
https://doi.org/10.1177/1527476418796632
Creswell, J. W., & Creswell, J. D. (2017). Research design: Qualitative, Quantitative, and mixed
methods approaches. SAGE Publications.
Dakanalis, A., Wiederhold, B. K., & Riva, G. (2023). Artificial intelligence: A game-changer for
mental health care. Cornell University. https://doi.org/10.31234/osf.io/ubdxz
Dale, S. (2023). A critique of principlism. Voices in Bioethics, 9.
https://doi.org/10.52214/vib.v9i.10522

173
Dalvi, C., Rathod, M., Patil, S., Gite, S., & Kotecha, K. (2021). A survey of AI-based facial
emotion recognition: Features, ML & DL techniques, age-wise datasets and future
directions. IEEE Access, 9, 165806–165840.
https://doi.org/10.1109/access.2021.3131733
Dawoodbhoy, F. M., Delaney, J., Cecula, P., Yu, J., Peacock, I., Tan, J., & Cox, B. (2021). AI in
patient flow: Applications of artificial intelligence to improve patient flow in NHS acute
mental health inpatient units. Heliyon, 7(5), Article e06993.
https://doi.org/10.1016/j.heliyon.2021.e06993
DeAngelis, T. (2021, September 1). Can real-world data lead to better interventions. American
Psychological Association. https://www.apa.org.
https://www.apa.org/monitor/2021/09/news-real-world-data
De Carvalho, M. R., Freire, R. C., & Nardi, A. E. (2010). Virtual reality as a mechanism for
exposure therapy. World Journal of Biological Psychiatry, 11(2–2), 220–230.
https://doi.org/10.3109/15622970802575985
Delahanty, R., Kaufman, D., & Jones, S. S. (2018). Development and evaluation of an automated
machine learning algorithm for in-hospital mortality risk adjustment among critical care
patients. Critical Care Medicine, 46(6), e481–e488.
https://doi.org/10.1097/ccm.0000000000003011
Del Re, A., Flückiger, C., Horvath, A. O., Symonds, D., & Wampold, B. E. (2012). Therapist
effects in the therapeutic alliance–outcome relationship: A restricted-maximum
likelihood meta-analysis. Clinical Psychology Review, 32(7), 642–649.
https://doi.org/10.1016/j.cpr.2012.07.002

174
Denecke, K., Gabarron, E., Grainger, R., Konstantinidis, S. T., Lau, A., Rivera-Romero, O.,
Miron-Shatz, T., & Merolli, M. (2019). Artificial intelligence for participatory health:
Applications, impact, and future implications. Yearbook of Medical Informatics, 28(01),
165–173. https://doi.org/10.1055/s-0039-1677902
DePodesta, M. (2024). The development of leadership communities of practice. Nursing
Administration Quarterly, 48(3), 225–233.
https://doi.org/10.1097/naq.0000000000000648
DeSouza, D. D., Tang, S. X., & Danilewitz, M. (2022). The burgeoning role of speech and
language assessment in schizophrenia spectrum disorders. Psychological Medicine,
53(10), 4825–4826. https://doi.org/10.1017/s0033291722001325
DeSouza, D. D., Xu, M., Fidalgo, C., Robin, J., & Simpson, W. S. (2021). Psychometric
approach to speech feature analysis as an objective measure of anxiety. Biological
Psychiatry, 89(9), S130. https://doi.org/10.1016/j.biopsych.2021.02.335
Diprose, W. K., Buist, N., Hua, N., Thurier, Q., Shand, G. B., & Robinson, R. (2020). Physician
understanding, explainability, and trust in a hypothetical machine learning risk calculator.
Journal of the American Medical Informatics Association, 27(4), 592–600.
https://doi.org/10.1093/jamia/ocz229
Di Sarno, L., Caroselli, A., Tonin, G., Graglia, B., Pansini, V., Causio, F. A., Gatto, A., &
Chiaretti, A. (2024). Artificial intelligence in pediatric emergency medicine:
Applications, challenges, and future perspectives. Biomedicines, 12(6), Article 1220.
https://doi.org/10.3390/biomedicines12061220
Došilović, F. K., Brčić, M., & Hlupić, N. (2018, May 21–25). Explainable artificial intelligence:
A survey [Paper presentation]. 41st International Convention on Information and

175
Communication Technology, Electronics and Microelectronics, Opatija, Croatia.
https://doi.org/10.23919/MIPRO.2018.8400040
Dos Santos, P. I. G., & Rosinhas, A. C. (2023). Artificial intelligence and mental health. Seven
Editora Academica. https://doi.org/10.56238/innovhealthknow-020
Durneva, P., Cousins, K., & Chen, M. (2020). The current state of research, challenges, and
future research directions of blockchain technology in patient care: Systematic review
Journal of Medical Internet Research, 22(7), Article e18619.
https://doi.org/10.2196/18619
Ebers, M. (2021). Regulating explainable AI in the European Union. An overview of the current
legal framework(s). SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3901732
Ebert, C., & Louridas, P. (2023). Generative AI for software practitioners. IEEE Software, 40(4),
30–38. https://doi.org/10.1109/MS.2023.3265877
Edlich, A., Ip, F., & Whiteman, R. (2018). How bots, algorithms, and artificial intelligence are
reshaping the future of corporate support functions. McKinsey & Company.
https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/how-botsalgorithms-and-artificial-intelligence-are-reshaping-the-future-of-corporate-supportfunctions
Edwards, B. (2023, January 11). Controversy erupts over non-consensual AI mental health
experiment [Updated]. Ars Technica. https://arstechnica.com/informationtechnology/2023/01/contoversy-erupts-over-non-consensual-ai-mental-healthexperiment/
Ejjami, R. (2024). AI-driven healthcare in France. International Journal for Multidisciplinary
Research, 6(3). https://doi.org/10.36948/ijfmr.2024.v06i03.22936

176
Elliott, R., Bohart, A. C., Watson, J. C., & Greenberg, L. S. (2011). Empathy. Psychotherapy,
48(1), 43–49. https://doi.org/10.1037/a0022187
Engle, R. L., Mohr, D. C., Holmes, S. K., Seibert, M. N., Afable, M. K., Leyson, J., & Meterko,
M. (2019). Evidence-based practice and patient-centered care: Doing both well. Health
Care Management Review, 46(3), 174–184.
https://doi.org/10.1097/hmr.0000000000000254
Enqvist, L. (2024). Rule-based versus AI-driven benefits allocation: GDPR and AIA legal
implications and challenges for automation in public social security administration.
Information & Communications Technology Law, 33(2), 222–246.
https://doi.org/10.1080/13600834.2024.2349835
Ettaloui, N., Arezki, S., & Gadi, T. (2023). An overview of Blockchain-based electronic health
records and compliance with GDPR and HIPAA. Data & Metadata, 2, Article 166.
https://doi.org/10.56294/dm2023166
European Commission. (2021). Laying down harmonised rules on artificial intelligence
(Artificial Intelligence Act) and amending certain union legislative acts. https://eurlex.europa.eu/resource.html?uri=cellar:e0649735-a372-11eb-9585-
01aa75ed71a1.0001.02/DOC_1&format=PDF
European Commission, Joint Research Centre, Delipetrev, B., Chrysi, T., & Uros, K. (2020).
Historical evolution of artificial intelligence: Analysis of the three main paradigm shifts
in AI. Publications Office. https://doi.org/10.2760/801580
Evans, D. R., Cowey, H. D., Gliksman, L., Csapo, K., & Heseltine, G. F. D. (1976). The
automated psychological evaluation system (APES). Behavior Research Methods &
Instrumentation, 8(2), 108–111. https://doi.org/10.3758/bf03201755

177
Familoni, N. B. T. (2024). Ethical frameworks for AI in healthcare entrepreneurship: A
theoretical examination of challenges and approaches. International Journal of Frontiers
in Biology and Pharmacy Research, 5(1), 057–065.
https://doi.org/10.53294/ijfbpr.2024.5.1.0032
Federal Trade Commission. (2021, June 22). FTC finalizes order with Flo Health, a fertilitytracking app that shared sensitive health data with Facebook, Google, and others.
https://www.ftc.gov/news-events/news/press-releases/2021/06/ftc-finalizes-order-flohealth-fertility-tracking-app-shared-sensitive-health-data-facebook-google
Federal Trade Commission. (2023, October 5). Protecting the privacy of health information: A
baker’s dozen takeaways from FTC cases. https://www.ftc.gov/businessguidance/blog/2023/07/protecting-privacy-health-information-bakers-dozen-takeawaysftc-cases
Federal Trade Commission. (2024a, May 6). FTC gives final approval to order banning
BetterHelp from sharing sensitive health data for advertising, requiring it to pay $7.8
million. https://www.ftc.gov/news-events/news/press-releases/2023/07/ftc-gives-finalapproval-order-banning-betterhelp-sharing-sensitive-health-data-advertising
Federal Trade Commission. (2024b, July 17). Proposed FTC order will prohibit telehealth firm
cerebral from using or disclosing sensitive data for advertising purposes, and require it
to pay $7 million. https://www.ftc.gov/news-events/news/pressreleases/2024/04/proposed-ftc-order-will-prohibit-telehealth-firm-cerebral-using-ordisclosing-sensitive-data
Feijóo-García, P. G., Wrenn, C. G., Stuart, J., De Siqueira, A. G., & Lok, B. (2023).
Participatory design of virtual humans for mental health support among North American

178
computer science students: Voice, appearance, and the similarity-attraction Effect. ACM
Transactions on Applied Perception, 20(3), 1–27. https://doi.org/10.1145/3613961
Ferrère, A., Rider, C., Renerte, B., & Edmondson, A. (2022, June 7). Fostering ethical conduct
through psychological safety. MIT Sloan Management Review.
https://sloanreview.mit.edu/article/fostering-ethical-conduct-through-psychologicalsafety/
Ferryman, K. (2021). The dangers of data colonialism in precision Public Health. Global Policy,
12(S6), 90–92. https://doi.org/10.1111/1758-5899.12953
Finkelstein, A., & McKnight, R. (2005). What did Medicare do (and was it worth it)?
https://doi.org/10.3386/w11609
Fiske, A., Henningsen, P., & Buyx, A. (2019). Your robot therapist will see you now: Ethical
Implications of Embodied Artificial intelligence in Psychiatry, Psychology, and
Psychotherapy. Journal of Medical Internet Research, 21(5), Article e13216.
https://doi.org/10.2196/13216
Fjelland, R. (2020). Why general artificial intelligence will not be realized. Humanities & Social
Sciences Communications, 7(1), 10. https://doi.org/10.1057/s41599-020-0494-4
Food and Drugs Administration. (2021, January 12). FDA releases artificial
intelligence/machine learning action plan. https://www.fda.gov/. Retrieved December 10,
2024, from https://www.fda.gov/news-events/press-announcements/fda-releasesartificial-intelligencemachine-learning-action-plan
Fox Rothschild LLP. (2019, October 3). Data privacy and bias concerns in AI health tech.
https://hipaahealthlaw.foxrothschild.com/2019/10/articles/hit-health-informationtechnol/data-privacy-and-bias-concerns-in-ai-health-tech

179
Franco, M. P., Monfort, C., Piñas-Mesa, A., & Rincón, E. (2021). Could avatar therapy enhance
mental health in chronic patients? A systematic review. Electronics, 10(18), Article 2212.
https://doi.org/10.3390/electronics10182212
Furnham, A., & Sjokvist, P. (2017). Empathy and mental health literacy. HLRP Health Literacy
Research and Practice, 1(2). https://doi.org/10.3928/24748307-20170328-01
Gaba, G. S., Hedabou, M., Kumar, P., Braeken, A., Liyanage, M., & Alazab, M. (2022). Zero
knowledge proofs based authenticated key agreement protocol for sustainable healthcare.
Sustainable Cities and Society, 80, 103766. https://doi.org/10.1016/j.scs.2022.103766
Gangwar, H., Date, H., & Ramaswamy, R. (2015). Understanding determinants of cloud
computing adoption using an integrated TAM-TOE model. Journal of Enterprise
Information Management, 28(1), 107–130. https://doi.org/10.1108/jeim-08-2013-0065
Gao, M. (2023, March 10–12). Research on the application of artificial intelligence in night
tourism [Paper presentation]. 3rd International Conference on Public Management and
Intelligent Society, Shanghai, China. https://doi.org/10.2991/978-94-6463-200-2_57
Geertz, C. (1973). Thick description: Towards an interpretive theory of culture. Basic Books.
https://doi.org/10.4324/9780203931950-11
Gega, L. (2017). The virtues of virtual reality in exposure therapy. The British Journal of
Psychiatry, 210(4), 245–246. https://doi.org/10.1192/bjp.bp.116.193300
Ghosh, A., Chakraborty, D., & Law, A. (2018). Artificial intelligence in Internet of things. CAAI
Transactions on Intelligence Technology, 3(4), 208–218.
https://doi.org/10.1049/trit.2018.1008

180
Gianfrancesco, M. A., Tamang, S., Yazdany, J., & Schmajuk, G. (2018). Potential biases in
machine learning algorithms using electronic health record data. JAMA Internal
Medicine, 178(11), 1544–1547. https://doi.org/10.1001/jamainternmed.2018.3763
Giovanola, B., & Tiribelli, S. (2022). Beyond bias and discrimination: redefining the AI ethics
principle of fairness in healthcare machine-learning algorithms. AI & Society, 38(2), 549–
563. https://doi.org/10.1007/s00146-022-01455-6
Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep learning. MIT Press.
https://dl.acm.org/citation.cfm?id=3086952
Gorre, N., Fuhrman, J., Carranza, E., Li, H., Madduri, R. K., Giger, M. L., & El Naqa, I. (2023).
User experience evaluation for MIDRC AI interface [Paper presentation]. SPIE Medical
Imaging, San Diego, CA, United States. https://doi.org/10.1117/12.2651812
Götzl, C., Hiller, S., Rauschenberg, C., Schick, A., Fechtelpeter, J., Abaigar, U. F., Koppe, G.,
Durstewitz, D., Reininghaus, U., & Krumm, S. (2022). Artificial intelligence-informed
mobile mental health apps for young people: A mixed-methods approach on users’ and
stakeholders’ perspectives. Child and Adolescent Psychiatry and Mental Health, 16(1).
https://doi.org/10.1186/s13034-022-00522-6
Graham, S., Depp, C. A., Lee, E., Nebeker, C., Tu, X., Kim, H., & Jeste, D. V. (2019). Artificial
Intelligence for mental health and mental illnesses: An overview. Current Psychiatry
Reports, 21(11), Article 116. https://doi.org/10.1007/s11920-019-1094-0
Grundy, Q., Chiu, K., Held, F., Continella, A., Bero, L., & Holz, R. (2019). Data sharing
practices of medicines related apps and the mobile ecosystem: traffic, content, and
network analysis. BMJ, 364, l920–l920. https://doi.org/10.1136/bmj.l920

181
Gugerty, L. (2006). Newell and Simon’s Logic Theorist: Historical Background and Impact on
Cognitive Modeling. Proceedings of the Human Factors and Ergonomics Society Annual
Meeting, 50(9), 880–884. https://doi.org/10.1177/154193120605000904
Guo, B., Lou, Y., & Pérez‐Castrillo, D. (2015). Investment, duration, and exit strategies for
corporate and independent venture capital‐backed start‐ups. Journal of Economics &
Management Strategy, 24(2), 415–455. https://doi.org/10.1111/jems.12097
Gupta, N. (2023). Artificial intelligence ethics and fairness: A study to address bias and fairness
issues in AI systems, and the ethical implications of AI applications. Revista Review
Index Journal of Multidisciplinary, 3(2), 24–35.
https://doi.org/10.31305/rrijm2023.v03.n02.004
Haber, Y., Levkovich, I., Hadar-Shoval, D., & Elyoseph, Z. (2024). The artificial third: A broad
view of the effects of introducing generative artificial intelligence on psychotherapy
(Preprint). JMIR Mental Health, 11, e54781. https://doi.org/10.2196/54781
Hadar-Shoval, D., Asraf, K., Mizrachi, Y., Haber, Y., & Elyoseph, Z. (2024). Assessing the
alignment of large language models with human values for mental health integration:
Cross-sectional study using Schwartz’s theory of basic values. JMIR Mental Health, 11,
Article e55988. https://doi.org/10.2196/55988
Haenlein, M., & Kaplan, A. (2019). A Brief History of artificial intelligence: on the past, present,
and future of artificial intelligence. California Management Review, 61(4), 5–14.
https://doi.org/10.1177/0008125619864925
Hagendorff, T. (2024). Deception abilities emerged in large language models. Proceedings of the
National Academy of Sciences, 121(24). https://doi.org/10.1073/pnas.2317967121

182
Halamka, J. D., Lippman, A., & Ekblaw, A. (2017, May 18). The potential for blockchain to
transform electronic health records. Harvard Business Review.
https://hbr.org/2017/03/the-potential-for-blockchain-to-transform-electronic-healthrecords?ab=at_art_art_1x4_s03
Hardy, A., & Allnutt, H. (2023, March 23). Replika, a ‘virtual friendship’ AI chatbot, receives
GDPR ban and threatened fine from Italian regulator over child safety concerns.
Lexology. https://www.lexology.com/library/detail.aspx?g=d028a39d-e65b-4d20-b5a5-
9cfa75a7addf
Hasan, H. R., Salah, K., Jayaraman, R., Yaqoob, I., Omar, M., & Ellahham, S. (2021).
Blockchain-Enabled telehealth services using smart contracts. IEEE Access, 9, 151944–
151959. https://doi.org/10.1109/access.2021.3126025
Hatem, R., Simmons, B., & Thornton, J. E. (2023). A call to address AI “hallucinations” and
how healthcare professionals can mitigate their risks. Cureus, 15(9), Artcile e44720.
https://doi.org/10.7759/cureus.44720
Haupt, C. E., & Marks, M. (2024). FTC regulation of AI-generated medical disinformation.
JAMA: The Journal of the American Medical Association, 332(23), 1975–1976.
https://doi.org/10.1001/jama.2024.19971
Heart, T., Ben-Assuli, O., & Shabtai, I. (2016). A review of PHR, EMR and EHR integration: A
more personalized healthcare and public health policy. Health Policy and Technology,
6(1), 20–25. https://doi.org/10.1016/j.hlpt.2016.08.002
Helpline Associates United [@HLAUnited]. (2023, May 26). Helpline Associates United
Statement on the Termination of the NEDA Helpline. Twitter.
https://twitter.com/HLAUnited/status/1662075335385526273/photo/1

183
Hendler, J. A. (2008). Avoiding another AI winter. IEEE Intelligent Systems, 23(2), 2–4.
https://doi.org/10.1109/mis.2008.20
Henry, S., Yetisgen, M., & Uzuner, O. (2021). Natural language processing in Mental Health
research and practice. In Computers in health care (pp. 317–353).
https://doi.org/10.1007/978-3-030-70558-9_13
Hilliard, A. (2024, December 3). The state of healthcare AI regulations in the US. Holistic AI.
Retrieved December 11, 2024, from https://www.holisticai.com/blog/healthcare-lawsus#:~:text=Healthtech%20laws%20already%20in%20effect,assist%20users%20with%20
basic%20tasks.
Hodge, J. G., & Gostin, K. G. (2004). Challenging themes in American Health information
privacy and the ’public’s health: Historical and modern assessments. The Journal of Law
Medicine & Ethics, 32(4), 670–679. https://doi.org/10.1111/j.1748-720x.2004.tb01972.x
Hodson, H. (2019, March 21). DeepMind and Google: The battle to control artificial intelligence.
The Economist. https://www.economist.com/1843/2019/03/01/deepmind-and-google-thebattle-to-control-artificial-intelligence
Hoffman, S., & Podgurski, A. (2020). Artificial intelligence and discrimination in health care
(Report Paper 2020-29). Case Western Reserve University School of Law.
https://case.edu/law/sites/default/files/2021-01/Sharona%20CLE%202-2021_0.pdf
Hollis, C., Morriss, R., Martin, J., Amani, S., Cotton, R., Denis, M., & Lewis, S. (2015).
Technological innovations in mental healthcare: harnessing the digital revolution. The
British Journal of Psychiatry, 206(4), 263–265.
https://doi.org/10.1192/bjp.bp.113.142612

184
Holm, S. (2002). Principles of Biomedical Ethics, 5th edn.: Beauchamp T L, Childress J F.
Oxford University Press, 2001, pound19.95, pp 454. ISBN 0-19-514332-9. Journal of
Medical Ethics, 28(5), 332. https://doi.org/10.1136/jme.28.5.332-a
Hummel, P., Braun, M., Tretter, M., & Dabrock, P. (2021). Data sovereignty: A review. Big
Data & Society, 8(1). https://doi.org/10.1177/2053951720982012
Hutson, E., & Melnyk, B. M. (2022). An adaptation of the COPE intervention for adolescent
bullying victimization improved mental and physical health symptoms. Journal of the
American Psychiatric Nurses Association, 28(6), 433–443.
https://doi.org/10.1177/10783903221127687
IEEE. (2019). Artificial intelligence. https://globalpolicy.ieee.org/wpcontent/uploads/2019/06/IEEE18029.pdf
Inkarbekov, M., Monahan, R., & Pearlmutter, B. A. (2023). Visualization of AI systems in
virtual reality: A comprehensive review. International Journal of Advanced Computer
Science and Applications, 14(8). https://doi.org/10.14569/ijacsa.2023.0140805
Ive, J. (2022). Leveraging the potential of synthetic text for AI in mental healthcare. Frontiers in
Digital Health, 4, Article 1010202. https://doi.org/10.3389/fdgth.2022.1010202
Jaffe, S. (2023, October 9). Denied by AI: Medicare Advantage’s “predictive” software cuts off
care, say feds. BenefitsPRO. https://www.benefitspro.com/2023/10/09/feds-rein-in-useof-predictive-software-that-limits-care-for-medicare-advantagepatients/?slreturn=20250120142917
Joerin, A., Rauws, M., Fulmer, R., & Black, V. (2020). Ethical artificial intelligence for digital
health organizations. Cureus. https://doi.org/10.7759/cureus.7202

185
Jones, A. (2019, September 26). The U.S. Department of Defense Selects CompanionMxTM for
suicide prevention study in active duty naval personnel. Business Wire.
https://www.businesswire.com/news/home/20190926005130/en/The-U.S.-Departmentof-Defense-Selects-CompanionMx%E2%84%A2-for-Suicide-Prevention-Study-inActive-Duty-Naval-Personnel
Joseph, A. P., & Babu, A. (2024). The unseen dilemma of AI in mental healthcare. AI & Society.
https://doi.org/10.1007/s00146-024-01937-9
Kandula, S., Ranga, S., & Moda, V. (2023). Hardware strategies for network optimization
supporting AI workloads. International Research Journal of Modernization in
Engineering Technology and Science. https://doi.org/10.56726/irjmets45389
Kargbo, R. B. (2023). Pioneering changes in psychiatry: biomarkers, psychedelics, and AI. ACS
Medicinal Chemistry Letters, 14(9), 1134–1137.
https://doi.org/10.1021/acsmedchemlett.3c00333
Kaul, V., Enslin, S., & Gross, S. A. (2020). History of artificial intelligence in medicine.
Gastrointestinal Endoscopy, 92(4), 807–812. https://doi.org/10.1016/j.gie.2020.06.040
Keaton, C. C. (2022). The health in mental health: Exploring individualized holistic health
practices as a process to transform mental health (Publication No. 29331360) [Doctoral
dissertation, The University of Texas at Arlington]. ProQuest Dissertations and Theses
Global.
Kelly, L. (2024, December 27). Third party data marketplaces: All you need to know 2024.
https://www.monda.ai/blog/third-party-data-marketplaces

186
Kerasidou, A. (2020). Artificial intelligence and the ongoing need for empathy, compassion and
trust in healthcare. Bulletin of the World Health Organization, 98(4), 245–250.
https://doi.org/10.2471/blt.19.237198
Khante, N., & Hande, K. N. (2019). A survey to chatbot system with knowledge base database
by using artificial intelligence & expert systems. International Research Journal of
Engineering and Technology, 6(05). https://www.irjet.net/archives/V6/i5/IRJETV6I5153.pdf
Kimiagari, S., & Baei, F. (2021). Promoting e-banking actual usage: Mix of technology
acceptance model and technology-organisation-environment framework. Enterprise
Information Systems, 16(8-9), Article 1894356.
https://doi.org/10.1080/17517575.2021.1894356
Klang, E., Apakama, D., Abbott, E. E., Vaid, A., Lampert, J., Sakhuja, A., Freeman, R.,
Charney, A. W., Reich, D., Kraft, M., Nadkarni, G. N., & Glicksberg, B. S. (2024). A
strategy for cost-effective large language model use at health system-scale. NPJ Digital
Medicine, 7(1), Article 30. https://doi.org/10.1038/s41746-024-01315-1
Klein, N. (2023, May 12). AI machines aren’t ‘hallucinating’. But their makers are. The
Guardian. https://www.theguardian.com/commentisfree/2023/may/08/ai-machineshallucinating-naomi-klein?CMP=share_btn_tw
Kordzadeh, N., & Ghasemaghaei, M. (2021). Algorithmic bias: Review, synthesis, and future
research directions. European Journal of Information Systems, 31(3), 388–409.
https://doi.org/10.1080/0960085x.2021.1927212
Krager, D., & Krager, C. (2016). HIPAA for health care professionals. Cengage Learning.

187
Kreps, S. E., & Kriner, D. L. (2023). How AI threatens democracy. Journal of Democracy,
34(4), 122–131. https://doi.org/10.1353/jod.2023.a907693
Kretzschmar, K., Tyroll, H., Pavarini, G., Manzini, A., & Singh, I. (2019). Can your phone be
your therapist? Young people’s ethical perspectives on the use of fully automated
conversational agents (Chatbots) in mental health support. Biomedical Informatics
Insights, 11, 117822261982908. https://doi.org/10.1177/1178222619829083
Kufel, J., Bargieł-Łączek, K., Kocot, S., Koźlik, M., Bartnikowska, W., Janik, M., Czogalik, Ł.,
Dudek, P., Magiera, M., Lis, A. E., Paszkiewicz, I., Nawrat, Z., Cebula, M., &
Gruszczyńska, K. (2023). What is machine learning, artificial neural networks and deep
learning?—Examples of practical applications in medicine. Diagnostics, 13(15), Article
2582. https://doi.org/10.3390/diagnostics13152582
Kuran, C. H. A., Morsut, C., Kruke, B. I., Krüger, M., Segnestam, L., Orru, K., Nævestad, T. O.,
Airola, M., Keränen, J., Gabel, F., Hansson, S., & Torpan, S. (2020). Vulnerability and
vulnerable groups from an intersectionality perspective. International Journal of Disaster
Risk Reduction, 50, Article 101826. https://doi.org/10.1016/j.ijdrr.2020.101826
Laijawala, V., Aachaliya, A., Jatta, H., & Pinjarkar, V. (2020). Mental health prediction using
data mining: A systematic review. SSRN Electronic Journal.
https://doi.org/10.2139/ssrn.3561661
Lamarche-Toloza, A. (2020). Digital humans, virtual humans, digital doubles. . . what’s the
difference? Virtuals - Creative Collective. https://virtuals.co/digital-humans-virtualhumans-differences-overview/

188
Laney, D. B. (2024, May 9). AI ethics essentials: Lawsuit over AI denial of healthcare. Forbes.
https://www.forbes.com/sites/douglaslaney/2023/11/16/ai-ethics-essentials-lawsuit-overai-denial-of-healthcare/
LaRose, C. (2024, August 23). 1557 Final rule protects against bias in health care algorithms -
National Health Law Program. National Health Law Program.
https://healthlaw.org/1557-final-rule-protects-against-bias-in-health-care-algorithms/
Latif, F. (2017). TELFest: an approach to encouraging the adoption of educational technologies.
Research in Learning Technology, 25(0). https://doi.org/10.25304/rlt.v25.1869
Lauria, M., & Long, M. F. (2019). Ethical dilemmas in professional planning practice in the
United States. Journal of the American Planning Association, 85(4), 393–404.
https://doi.org/10.1080/01944363.2019.1627238
Lee, E., Torous, J., De Choudhury, M., Depp, C. A., Graham, S., Kim, H., Paulus, M. P.,
Krystal, J. H., & Jeste, D. V. (2021). Artificial intelligence for mental health care:
Clinical applications, barriers, facilitators, and artificial wisdom. Biological Psychiatry:
Cognitive Neuroscience and Neuroimaging, 6(9), 856–864.
https://doi.org/10.1016/j.bpsc.2021.02.001
Lee, S., Kang, J., Kim, H., Chung, K., Lee, D., & Yeo, J. (2024). COCOA: CBT-based
conversational counseling agent using memory specialized in cognitive distortions and
dynamic prompt. Cornell University. https://doi.org/10.48550/arxiv.2402.17546
Lee, Y., & Park, W. (2022). Diagnosis of depressive disorder model on facial expression based
on Fast R-CNN. Diagnostics, 12(2), 317. https://doi.org/10.3390/diagnostics12020317

189
Lenert, L., Lane, S., & Wehbe, R. M. (2023). Could an artificial intelligence approach to prior
authorization be more human? Journal of the American Medical Informatics Association,
30(5), 989–994. https://doi.org/10.1093/jamia/ocad016
Leone, D. (2021). Data Colonialism in Canada: Decolonizing Data Through Indigenous data
governance. https://doi.org/10.22215/etd/2021-14697
Liehner, G. L., Hick, A., Biermann, H., Brauner, P., & Ziefle, M. (2023). Perceptions, attitudes
and trust toward artificial intelligence — An assessment of the public opinion. AHFE
International. https://doi.org/10.54941/ahfe1003271
Liu, D. (2021). machinesMemory: Malleability of AI technique, the data generated by machine
learning algorithms. Electronic Workshops in Computing.
https://doi.org/10.14236/ewic/eva2021.31
Long, K. M., & Meadows, G. N. (2017). Simulation modelling in mental health: A systematic
review. Journal of Simulation, 12(1), 76–85. https://doi.org/10.1057/s41273-017-0062-0
Lu, T., Liu, X., Sun, J., Bao, Y., Schuller, B. W., Han, Y., & Lu, L. (2023). Bridging the gap
between artificial intelligence and mental health. Science Bulletin, 68(15), 1606–1610.
https://doi.org/10.1016/j.scib.2023.07.015
Luxton, D. D. (2016). An introduction to artificial intelligence in behavioral and mental health
care. In D. D. Luxton (Ed.), Artificial intelligence in behavioral and mental health care
(pp. 1–26). Elsevier https://doi.org/10.1016/b978-0-12-420248-1.00001-5
Ma, X., & Jiang, C. (2023). On the ethical risks of artificial intelligence applications in education
and its avoidance strategies. Journal of Education Humanities and Social Sciences, 14,
354–359. https://doi.org/10.54097/ehss.v14i.8868

190
Maddali, H. T., Dixon, E., Pradhan, A., & Lazar, A. (2022, April 28). Investigating the potential
of artificial intelligence powered interfaces to support different types of memory for
people with dementia [Paper presentation]. CHI Conference on Human Factors in
Computing Systems, New Orleans, LA, United States.
Margam, R. (2023). ChatGPT: the silent partner in healthcare. The Review of Contemporary
Scientific and Academic Studies, 3(10). https://doi.org/10.55454/rcsas.3.10.2023.005
Marks, M., & Haupt, C. E. (2023). AI chatbots, health privacy, and challenges to HIPAA
compliance. JAMA, 330(4), 309–310. https://doi.org/10.1001/jama.2023.9458
Marotta, G., & Au, C. (2022). Budgeting in the age of artificial intelligence – new approaches,
old challenges? International Journal of Artificial Intelligence and Machine Learning,
2(2), 1–11. https://doi.org/10.51483/ijaiml.2.2.2022.1-11
Marr, B. (2023, July 6). AI in mental health: Opportunities and challenges in developing
intelligent digital therapies. Forbes.
https://www.forbes.com/sites/bernardmarr/2023/07/06/ai-in-mental-health-opportunitiesand-challenges-in-developing-intelligent-digital-therapies/?sh=7e5dc2c45e10
Martin, A., Sharma, G., De Souza, S. P., Taylor, L., Van Eerd, B., McDonald, S. M., Marelli, M.,
Cheesman, M., Scheel, S., & Dijstelbloem, H. (2022). Digitisation and sovereignty in
humanitarian space: Technologies, territories and tensions. Geopolitics, 28(3), 1362–
1397. https://doi.org/10.1080/14650045.2022.2047468
Martínez-Castaño, R., Htait, A., Azzopardi, L., & Moshfeghi, Y. (2021, September 21–24).
BERT-based transformers for early detection of mental health illnesses [Paper
presentation] 12th International Conference of the CLEF Association, online.
https://doi.org/10.1007/978-3-030-85251-1_15

191
Masche, J., & Le, N. (2017, June 30–July 1). A review of technologies for conversational systems
[Paper presentation]. 5th International Conference on Computer Science, Applied
Mathematics and Applications. https://doi.org/10.1007/978-3-319-61911-8_19
Mautang, T. W. E., & Suarjana, I. W. G. (2023). The global and cultural context of using AI for
mental health. Journal of Public Health, 46(2), e343.
https://doi.org/10.1093/pubmed/fdad262
May, J. (2018). Regard for reason in the moral mind. Oxford University Press.
Mayover, T. (2024, October 2). When AI technology and HIPAA collide. The HIPAA Journal.
https://www.hipaajournal.com/when-ai-technology-and-hipaa-collide/
Mazziotti, R., & Rutigliano, G. (2020). Tele–Mental Health for Reaching Out to Patients in a
Time of Pandemic: Provider Survey and Meta-analysis of Patient Satisfaction (Preprint).
JMIR Mental Health, 8(7). https://doi.org/10.2196/preprints.26187
McCradden, M. D., Joshi, S., Anderson, J. A., Mazwi, M., Goldenberg, A., & Shaul, R. Z.
(2020). Patient safety and quality improvement: Ethical principles for a regulatory
approach to bias in healthcare machine learning. Journal of the American Medical
Informatics Association, 27(12), 2024–2027. https://doi.org/10.1093/jamia/ocaa085
McDougall, R. (2018). Computer knows best? The need for value-flexibility in medical AI.
Journal of Medical Ethics, 45(3), 156–160. https://doi.org/10.1136/medethics-2018-
105118
McGraw, D. (2012). Building public trust in uses of Health Insurance Portability and
Accountability Act de-identified data. Journal of the American Medical Informatics
Association, 20(1), 29–34. https://doi.org/10.1136/amiajnl-2012-000936

192
McGregor, B., Belton, A., Henry, T. L., Wrenn, G., & Holden, K. B. (2019). Improving
behavioral health equity through cultural competence training of health care providers.
Ethnicity & Disease, 29(Supp2), 359–364. https://doi.org/10.18865/ed.29.s2.359
McLellan, E., MacQueen, K. M., & Neidig, J. L. (2003). Beyond the qualitative interview: Data
preparation and transcription. Field Methods, 15(1), 63–84.
https://doi.org/10.1177/1525822x02239573
Meinke, A., Schoen, B., Scheurer, J., Balesni, M., Shah, R., & Hobbhahn, M. (2024). Frontier
Models are Capable of In-context Scheming. arXiv.org. https://arxiv.org/abs/2412.04984
Melcher, J. R., & Torous, J. (2020). Smartphone apps for college mental health: A concern for
privacy and quality of current offerings. Psychiatric Services, 71(11), 1114–1119.
https://doi.org/10.1176/appi.ps.202000098
Mello, M. M., & Rose, S. (2024). Denial—Artificial intelligence tools and health insurance
coverage decisions. JAMA Health Forum, 5(3), e240622.
https://doi.org/10.1001/jamahealthforum.2024.0622
Melnyk, B. M., Hoying, J., & Tan, A. (2020). Effects of the MINDSTRONG© CBT-based
program on depression, anxiety and healthy lifestyle behaviors in graduate health
sciences students. Journal of American College Health, 70(4), 1001–1009.
https://doi.org/10.1080/07448481.2020.1782922
Memishi, B., Appuswamy, R., & Paradies, M. (2019). Cold storage data archives: more than
just a bunch of tapes. Cornell University. https://doi.org/10.48550/arxiv.1904.04736
Mendes-Santos, C., Nunes, F., Weiderpass, E., Santana, R., & Andersson, G. (2022).
Understanding mental health professionals’ perspectives and practices regarding the

193
implementation of digital mental health: Qualitative study. JMIR Formative Research,
6(4), Article e32558. https://doi.org/10.2196/32558
Mercer, T., & Khurshid, A. (2021). Advancing health equity for people experiencing
homelessness using blockchain technology for identity management: A research agenda.
Journal of Health Care for the Poor and Underserved, 32(2S), 262–277.
https://doi.org/10.1353/hpu.2021.0062
Merriam, S. B., & Tisdell, E. J. (2015). Qualitative research: A guide to design and
implementation. John Wiley & Sons.
Merriam, S. B., & Tisdell, E. J. (2016). Qualitative research: A guide to design and
implementation (4th ed.). John Wiley & Sons.
Misra, S., & Mondal, A. (2011). Identification of a company’s suitability for the adoption of
cloud computing and modelling its corresponding Return on Investment. Mathematical
and Computer Modelling, 53(3–4), 504–521. https://doi.org/10.1016/j.mcm.2010.03.037
Mittelstadt, B. (2019). Principles alone cannot guarantee ethical AI. Nature Machine
Intelligence, 1(11), 501–507. https://doi.org/10.1038/s42256-019-0114-4
Moczuk, E., & Płoszajczak, B. (2020). Artificial intelligence – Benefits and threats for society.
Humanities and Social Sciences Quarterly. https://doi.org/10.7862/rz.2020.hss.22
Mody, V., & Mody, V. (2019). Mental health monitoring system using artificial intelligence: A
review. IEEE. https://doi.org/10.1109/i2ct45611.2019.9033652
Moggia, D., Lutz, W., Brakemeier, E., & Bickman, L. (2024). Treatment personalization and
precision mental health care: Where are we and where do we want to go? Administration
and Policy in Mental Health and Mental Health Services Research, 51(5), 611–616.
https://doi.org/10.1007/s10488-024-01407-w

194
Molnár-Gábor, F. (2019). Artificial intelligence in healthcare: Doctors, patients, and liabilities.
In T. Wischmeyer & T. Rademacher (Ed.), Regulating artificial intelligence (pp. 337–
360). https://doi.org/10.1007/978-3-030-32361-5_15
Moradi, P., & Levy, K. (2020). The future of work in the age of AI: Displacement or riskshifting? In M. D. Dubber, F. Pasquale, & S. Das (Eds.), Oxford handbook of ethics of AI
(pp. 271–287). Oxford University Press.
Moreau, J. T., Baillet, S., & Dudley, R. W. (2020). Biased intelligence: on the subjectivity of
digital objectivity. BMJ Health & Care Informatics, 27(3), Article e100146.
https://doi.org/10.1136/bmjhci-2020-100146
Morris, C. (2023, May 26). National Eating Disorder Association replaces human helpline staff
with an AI chatbot. Fortune Well. https://fortune.com/well/2023/05/26/national-eatingdisorder-association-ai-chatbot-tessa/
Na, S., Heo, S., Han, S., Shin, Y., & Roh, Y. (2022). Acceptance model of artificial intelligence
(AI)-based technologies in construction firms: Applying the technology acceptance
model (TAM) in combination with the technology–organisation–environment (TOE)
framework. Buildings, 12(2), Article 90. https://doi.org/10.3390/buildings12020090
Nakao, M., Shirotsuki, K., & Sugaya, N. (2021). Cognitive–behavioral therapy for management
of mental health and stress-related disorders: Recent advances in techniques and
technologies. BioPsychoSocial Medicine, 15(1), 16. https://doi.org/10.1186/s13030-021-
00219-w
Napolitano, E. (2023, November 21). UnitedHealth uses faulty AI to deny elderly patients
medically necessary coverage, lawsuit claims. CBS News.

195
https://www.cbsnews.com/news/unitedhealth-lawsuit-ai-deny-claims-medicareadvantage-health-insurance-denials/
National Eating Disorders Association. (2023, September 22). Our work.
https://www.nationaleatingdisorders.org/about-us/our-work
Neumann, O., Guirguis, K., & Steiner, R. (2022). Exploring artificial intelligence adoption in
public organizations: A comparative case study. Public Management Review, 26(1), 1–
28. https://doi.org/10.1080/14719037.2022.2048685
Newhouse, J. P., & Sinaiko, A. D. (2007). Productivity adjustment in the Medicare physician fee
schedule update. Health Care Financing Review, 29(2), 5–14.
https://pmc.ncbi.nlm.nih.gov/articles/PMC4195023/
Nguyen, N., Labonté-LeMoyne, É., Grégoire, Y., Radanielina-Hita, M. L., & Sénécal, S. (2022,
June 26–July 1). Understanding the ’patients’ adoption and usage of AI solution in
mental health: A scoping review [Paper presentation]. 24th International Conference on
Human-Computer Interaction, online. https://doi.org/10.1007/978-3-031-19682-9_85
Niebel, C. (2021). The impact of the general data protection regulation on innovation and the
global political economy. Computer Law & Security Report, 40, Article 105523.
https://doi.org/10.1016/j.clsr.2020.105523
Nilsson, N. J. (2009). The quest for artificial intelligence. Cambridge University Press.
https://doi.org/10.1017/CBO9780511819346
Noguchi, Y. (2023, January 19). Therapy by chatbot? The promise and challenges in using AI
for mental health. NPR. https://www.npr.org/sections/healthshots/2023/01/19/1147081115/therapy-by-chatbot-the-promise-and-challenges-in-usingai-for-mental-health

196
Norcross, J. C., & Lambert, M. J. (2018). Psychotherapy relationships that work III.
Psychotherapy, 55(4), 303–315. https://doi.org/10.1037/pst0000193
Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an
algorithm used to manage the health of populations. Science, 366(6464), 447–453.
https://doi.org/10.1126/science.aax2342
O’Connor, J., & Matthews, G. (2011). Informational privacy, public health, and state laws.
American Journal of Public Health, 101(10), 1845–1850.
https://doi.org/10.2105/ajph.2011.300206
Ofulue, J., & Benyoucef, M. (2022). Data monetization: insights from a technology-enabled
literature review and research agenda. Management Review Quarterly, 74(2), 521–565.
https://doi.org/10.1007/s11301-022-00309-1
Oguamanam, C. (2020). Indigenous peoples, data sovereignty and self-determination: Current
realities and imperatives. The African Journal of Information and Communication,
2020(26), 31–50. https://doi.org/10.23962/10539/30360
Oliveira, T., & Martins, M. F. O. (2010). Understanding e‐business adoption across industries in
European countries. Industrial Management and Data Systems, 110(9), 1337–1354.
https://doi.org/10.1108/02635571011087428
Oliveira, T., Thomas, M. A., & Espadanal, M. (2014). Assessing the determinants of cloud
computing adoption: An analysis of the manufacturing and services sectors. Information
& Management, 51(5), 497–510. https://doi.org/10.1016/j.im.2014.03.006
Ooijen, P., Darzidehkalani, E., & Dekker, A. (2022). AI technical considerations: Data storage,
cloud usage and AI pipeline. Cornell University.
https://doi.org/10.48550/arxiv.2201.08356

197
Pandey, M., Kamrul, R., Michaels, C. R., & McCarron, M. (2021). Identifying barriers to
healthcare access for new immigrants: a qualitative study in Regina, Saskatchewan,
Canada. Journal of Immigrant and Minority Health, 24(1), 188–198.
https://doi.org/10.1007/s10903-021-01262-z
Paniagua, F. A. (2018). ICD-10 versus DSM-5 on cultural issues. SAGE Open, 8(1),
215824401875616. https://doi.org/10.1177/2158244018756165
Park, H. J. (2024). Patient perspectives on informed consent for medical AI: A web-based
experiment. Digital Health, 2024(10). https://doi.org/10.1177/20552076241247938
Park, P. S., Goldstein, S., O’Gara, A., Chen, M., & Hendrycks, D. (2024). AI deception: A
survey of examples, risks, and potential solutions. Patterns, 5(5), Article 100988.
https://doi.org/10.1016/j.patter.2024.100988
Parsons, R. D., & Dickinson, K. L. (2016). Ethical practice in the human services: From
Knowing to Being. SAGE Publications.
Pasricha, S. (2022). AI ethics in smart Healthcare. IEEE Consumer Electronics Magazine, 12(4),
12–20. https://doi.org/10.1109/mce.2022.3220001
Patton, M. Q. (1999). Enhancing the quality and credibility of qualitative analysis. PubMed, 34(5
Pt 2), 1189–1208. https://pubmed.ncbi.nlm.nih.gov/10591279
Patton-López, M. (2022). Communities in action: Pathways to health equity. Journal of Nutrition
Education and Behavior, 54(1), 94–95. https://doi.org/10.1016/j.jneb.2021.09.012
Payne, K. (2024, October 25). AI chatbot pushed teen to kill himself, lawsuit alleges. AP News.
https://apnews.com/article/chatbot-ai-lawsuit-suicide-teen-artificial-intelligence9d48adc572100822fdbc3c90d1456bd0

198
Percy, S. (2024, August 8). Do the right thing: 4 ways to navigate the ethics of AI. Forbes.
https://www.forbes.com/sites/sallypercy/2024/08/05/do-the-right-thing-4-ways-tonavigate-the-ethics-of-ai/
Perlis, R. (2023, February 6). Mindstrong’s demise and the future of mental health care. STAT.
https://www.statnews.com/2023/02/06/mindstrong-demise-future-mental-health-care/
Petersson, L., Larsson, I., Nygren, J. M., Nilsen, P., Neher, M., Reed, J., Tyskbo, D., &
Svedberg, P. (2022). Challenges to implementing artificial intelligence in healthcare: A
qualitative interview study with healthcare leaders in Sweden. BMC Health Services
Research, 22(1), Article 850. https://doi.org/10.1186/s12913-022-08215-8
Place, S., Blanch‐Hartigan, D., Smith, V., Erb, J. L., Marci, C. D., & Ahern, D. K. (2020). Effect
of a mobile monitoring system vs usual care on depression symptoms and psychological
health. JAMA Network Open, 3(1), Article e1919403.
https://doi.org/10.1001/jamanetworkopen.2019.19403
Polzer, A., Fleiß, J., Ebner, T., Kainz, P., Koeth, C., & Thalmann, S. (2022). Validation of AIbased information systems for sensitive use cases: using an XAI approach in
pharmaceutical engineering. Proceedings of the . . . Annual Hawaii International
Conference on System Sciences/Proceedings of the Annual Hawaii International
Conference on System Sciences. https://doi.org/10.24251/hicss.2022.186
Poufinas, T., Gogas, P., Papadimitriou, T., & Zaganidis, E. (2023). Machine learning in
forecasting motor insurance claims. Risks, 11(9), Article 164.
https://doi.org/10.3390/risks11090164
Powell, J., & Kleiner, A. (2023). The AI dilemma: 7 principles for responsible technology.
Berrett-Koehler Publishers.

199
Powles, J., & Hodson, H. (2017). Google DeepMind and healthcare in an age of algorithms.
Health and Technology, 7(4), 351–367. https://doi.org/10.1007/s12553-017-0179-1
Pratap, A., Homiar, A., Waninger, L., Herd, C., Suver, C., Volponi, J., Anguera, J. A., & Areán,
P. (2022). Real-world behavioral dataset from two fully remote smartphone-based
randomized clinical trials for depression. Scientific Data, 9(1).
https://doi.org/10.1038/s41597-022-01633-7
Prochaska, J. J., Vogel, E. A., Chieng, A., Kendra, M. S., Baiocchi, M., Pajarito, S., & Robinson,
A. (2021). A therapeutic relational agent for reducing problematic substance use
(WOEBOT): Development and usability study. Journal of Medical Internet Research,
23(3), Article e24850. https://doi.org/10.2196/24850
Psychiatrist.com. (2023, June 5). NEDA suspends AI chatbot for giving harmful eating disorder
advice. https://www.psychiatrist.com/news/neda-suspends-ai-chatbot-for-giving-harmfuleating-disorder-advice/
Pujari, S., Reis, A., Zhao, Y., Alsalamah, S., Serhan, F., Reeder, J. C., & Labrique, A. (2023).
Artificial intelligence for global health: cautious optimism with safeguards. Bulletin of
the World Health Organization, 101(06), 364–364A.
https://doi.org/10.2471/blt.23.290215
Rangavittal, P. B. (2022). Evolving role of AI in enhancing patient care within digital health
platforms. Journal of Artificial Intelligence & Cloud Computing, 1–6.
https://doi.org/10.47363/jaicc/2022(1)241
Räsänen, M., & Nyce, J. M. (2013). The raw is cooked. Science Technology & Human Values,
38(5), 655–677. https://doi.org/10.1177/0162243913480049

200
Razzouk, D. (2023). Editorial: Mental health economics and public mental health policy: mental
health services costs, quality and its impact on reducing the burden of mental illness.
Frontiers in Health Services, 3, Article 1267580.
https://doi.org/10.3389/frhs.2023.1267580
Regueiro, C., Seco, I., Gutiérrez-Agüero, I., Urquizu, B., & Mansell, J. (2021). A blockchainbased audit trail mechanism: Design and implementation. Algorithms, 14(12), Article
341. https://doi.org/10.3390/a14120341
Rezaeikhonakdar, D. (2023). AI chatbots and challenges of HIPAA compliance for AI
developers and vendors. The Journal of Law Medicine & Ethics, 51(4), 988–995.
https://doi.org/10.1017/jme.2024.15
Rizzo, A., Hartholt, A., & Mozgai, S. (2021). From combat to COVID-19 – Managing the
impact of trauma using virtual reality. Journal of Technology in Human Services, 39(3),
314–347. https://doi.org/10.1080/15228835.2021.1915931
Rizzo, A., Lange, B., Buckwalter, J. G., Forbell, E., Kim, J., Sagae, K., Williams, J., Difede, J.,
Rothbaum, B. O., Reger, G., Parsons, T., & Kenny, P. (2011). SimCoach: an intelligent
virtual human system for providing healthcare information and support. International
Journal on Disability and Human Development, 10(4), 277–281.
https://doi.org/10.1515/IJDHD.2011.046
Rizzo, A., & Shilling, R. (2017). Clinical virtual reality tools to advance the prevention,
assessment, and treatment of PTSD. European Journal of Psychotraumatology, 8(sup5).
https://doi.org/10.1080/20008198.2017.1414560
Rizzo, A. S., & Koenig, S. T. (2017). Is clinical virtual reality ready for primetime?
Neuropsychology, 31(8), 877–899. https://doi.org/10.1037/neu0000405

201
Robin, J., Xu, M., Kaufman, L. D., & Simpson, W. S. (2021). Using digital speech assessments
to detect early signs of cognitive impairment. Frontiers in Digital Health, 3, Article
749758. https://doi.org/10.3389/fdgth.2021.749758
Rocher, L., Hendrickx, J. M., & De Montjoye, Y. (2019). Estimating the success of reidentifications in incomplete datasets using generative models. Nature Communications,
10(1). https://doi.org/10.1038/s41467-019-10933-3
Rosenfeld, S., Bernasek, C., & Mendelson, D. (2005). Medicare’s next voyage: encouraging
physicians to adopt health information technology. Health Affairs, 24(5), 1138–1146.
https://doi.org/10.1377/hlthaff.24.5.1138
Russell, C. (2023, February 8). Health care bias is dangerous. but so are ‘fairness’ algorithms.
WIRED. https://www.wired.com/story/bias-statistics-artificial-intelligence-healthcare/
Russell, S., & Norvig, P. (2016). Artificial intelligence: A modern approach. Pearson Education
Limited.
Russell, S. J., & Norvig, P. (1995). Artificial intelligence: a modern approach. Choice Reviews
Online, 33(03), 33–1577. https://doi.org/10.5860/choice.33-1577
Sandu, I., Wiersma, M., & Manichand, D. (2022). Time to audit your AI algorithms. Maandblad
Voor Accountancy En Bedrijfseconomie, 96(7/8), 253–265.
https://doi.org/10.5117/mab.96.90108
Sarker, I. H. (2022). AI-based modeling: Techniques, applications and research issues towards
automation, intelligent and smart systems. SN Computer Science, 3(2).
https://doi.org/10.1007/s42979-022-01043-x
Scheuch, K. (2024, December 2). News Release: Utah Department of Commerce’s Office of
Artificial Intelligence announces first regulatory mitigation agreement [Press release].

202
https://blog.commerce.utah.gov/2024/12/02/news-release-utah-department-ofcommerces-office-of-artificial-intelligence-announces-first-regulatory-mitigationagreement/
Schueller, S. M., Hunter, J. F., Figueroa, C., & Aguilera, A. (2019). Use of digital mental health
for marginalized and underserved populations. Current Treatment Options in Psychiatry,
6(3), 243–255. https://doi.org/10.1007/s40501-019-00181-z
Seger, E. (2022). In defence of principlism in AI ethics and governance. Philosophy &
Technology, 35(2). https://doi.org/10.1007/s13347-022-00538-y
Shahriari, K., & Shahriari, M. (2017). IEEE standard review — Ethically aligned design: A
vision for prioritizing human wellbeing with artificial intelligence and autonomous
systems. IEEE. https://doi.org/10.1109/ihtc.2017.8058187
Sharfstein, S. S. (2019). Community mental health and deinstitutionalization. Psychiatric News,
54(24). https://doi.org/10.1176/appi.pn.2019.12b28
Sharma, R. (2020). Artificial Intelligence in Healthcare: a review. Türk Bilgisayar Ve Matematik
Eğitimi Dergisi, 11(1), 1663–1667. https://doi.org/10.61841/turcomat.v11i1.14628
Shi, L., & Singh, D. A. (2004). Essentials of the U.S. health care system.
https://openlibrary.org/books/OL22552128M/Essentials_of_the_U.S._health_care_syste
m
Shilton, K., Moss, E., Gilbert, S. A., Bietz, M. J., Fiesler, C., Metcalf, J., Vitak, J., & Zimmer,
M. (2021). Excavating awareness and power in data science: A manifesto for trustworthy
pervasive data research. Big Data & Society, 8(2).
https://doi.org/10.1177/20539517211040759
Shu, J. (2024). Data Storage Architectures and Technologies. Springer.

203
Shum, H., He, X., & Li, D. (2018). From Eliza to XiaoIce: challenges and opportunities with
social chatbots. Frontiers of Informaion Technology & Electronic Engineering, 19(1),
10–26. https://doi.org/10.1631/fitee.1700826
Siala, H., & Wang, Y. (2022). SHIFTing artificial intelligence to be responsible in healthcare: A
systematic review. Social Science & Medicine, 296, 114782.
https://doi.org/10.1016/j.socscimed.2022.114782
Silverman, J. J., Galanter, M., Jackson-Triche, M., Jacobs, D. G., Lomax, J. W., Riba, M., Tong,
L., Watkins, K. E., Fochtmann, L. J., Rhoads, R. S., & Yager, J. (2015). The American
Psychiatric Association practice guidelines for the psychiatric evaluation of adults.
American Journal of Psychiatry, 172(8), 798–802.
https://doi.org/10.1176/appi.ajp.2015.1720501
Singh, R., & Gill, S. S. (2023). Edge AI: A survey. Internet of Things and Cyber-Physical
Systems, 3, 71–92. https://doi.org/10.1016/j.iotcps.2023.02.004
Singh, S., & Thakur, H. K. (2020). Survey of Various AI Chatbots Based on Technology Used.
IEEE. https://doi.org/10.1109/icrito48877.2020.9197943
Sinha, C., Meheli, S., & Kadaba, M. (2023). Understanding digital mental health needs and
usage with an artificial intelligence–led mental health app (WYSA) during the COVID19 pandemic: Retrospective analysis. JMIR Formative Research, 7, e41913.
https://doi.org/10.2196/41913
Sipola, T., Alatalo, J., Kokkonen, T., & Rantonen, M. (2022). Artificial intelligence in the IoT
Era: A review of edge AI hardware and software. FRUCT Oy, 320–331.
https://doi.org/10.23919/fruct54823.2022.9770931

204
Šlepecký, M., Škobrtal, P., Novotný, M., Bazinková, E., & Kotianová, A. (2018). Exposure
therapy by virtual reality and its monitoring by biofeedback. Cognitive Remediation
Journal, 7(1), 4–9. https://doi.org/10.5507/crj.2018.001
Sofia, S., Montserrat, L. C., Emilia, G. G., Giuditta, D. P., Martínez-Plumed, F., & Delipetrev, B.
(2020). AI WATCH. Defining Artificial Intelligence. Publications Office of the
European Union. https://doi.org/10.2760/382730
State of Colorado. (2024). Consumer protections for artificial intelligence. Colorado General
Assembly. Retrieved December 11, 2024, from https://leg.colorado.gov/bills/sb24-205
Stix, C. (2018). 3 ways AI could help our mental health. World Economic Forum.
http://www.weforum.org/agenda/2018/03/3-ways-ai-could-could-be-used-in-mentalhealth/
Stoeklé, H., Mamzer-Bruneel, M., Vogt, G., & Hervé, C. (2016). 23andMe: A new two-sided
data-banking market model. BMC Medical Ethics, 17(1). https://doi.org/10.1186/s12910-
016-0101-9
Straw, I., & Callison-Burch, C. (2020). Artificial Intelligence in mental health and the biases of
language based models. PLoS ONE, 15(12), e0240376.
https://doi.org/10.1371/journal.pone.0240376
Stringer, H. (2025, January). Technology is reshaping practice to expand psychology’s reach.
https://www.apa.org. https://www.apa.org/monitor/2025/01/trends-technology-shapingpractice
Sutton, J., & Austin, Z. (2015). Qualitative research: data collection, analysis, and management.
The Canadian Journal of Hospital Pharmacy, 68(3).
https://doi.org/10.4212/cjhp.v68i3.1456

205
Suver, C., Harper, J., Loomba, J., Saltz, M., Solway, J., Anzalone, A. J., Walters, K., Pfaff, E.,
Walden, A., McMurry, J., Chute, C. G., & Haendel, M. (2023). The N3C governance
ecosystem: A model socio-technical partnership for the future of collaborative analytics
at scale. Journal of Clinical and Translational Science, 7(1).
https://doi.org/10.1017/cts.2023.681
Terranova, C., Cestonaro, C., Fava, L., & Cinquetti, A. (2024). AI and professional liability
assessment in healthcare. A revolution in legal medicine? Frontiers in Medicine, 10.
https://doi.org/10.3389/fmed.2023.1337335
Thomson, S., Sagan, A., & Mossialos, E. (2020). Private health insurance: History, politics and
performance. Cambridge University Press.
Timmons, A. C., Duong, J. B., Fiallo, N. S., Lee, T., Vo, H. P. Q., Ahle, M. W., Comer, J. S.,
Brewer, L. C., Frazier, S. L., & Chaspari, T. (2022). A call to action on assessing and
mitigating bias in artificial intelligence applications for mental health. Perspectives on
Psychological Science. https://doi.org/10.1177/17456916221134490
Timofeyev, Y., & Jakovljević, M. (2020). Fraudster’s and victims’ profiles and loss predictors’
hierarchy in the mental healthcare industry in the US. Journal of Medical Economics,
23(10), 1111–1122. https://doi.org/10.1080/13696998.2020.1801454
Toosi, A., Bottino, A., Saboury, B., Siegel, E. L., & Rahmim, A. (2021). A brief history of AI:
How to prevent another winter (A critical review). PET Clinics, 16(4), 449–469.
https://doi.org/10.1016/j.cpet.2021.07.001
Tornatzky, L. G., Fleischer, M., & Chakrabarti, A. (1990). Processes of technological
innovation. Lexington Books.

206
Tornero-Costa, R., Martinez-Millana, A., Azzopardi-Muscat, N., Lazeri, L., Traver, V., &
Novillo-Ortiz, D. (2023). Methodological and quality flaws in the use of artificial
intelligence in mental health research: Systematic review. JMIR Mental Health, 10,
Article e42045. https://doi.org/10.2196/42045
Trang, B. (2024, December 21). As Congress waffles on AI, state legislatures step up to fill the
void. STAT. https://www.statnews.com/2024/12/23/regulating-health-ai-state-lawsincrease-compliance-burden/
Tseng, E., Meza, K., Marsteller, J. A., Clark, J. M., Maruthur, N. M., & Smith, K. (2022).
Engaging payors and primary care physicians together in improving diabetes prevention.
Journal of General Internal Medicine, 38(2), 309–314. https://doi.org/10.1007/s11606-
022-07788-8
Turner Lee, N., Resnick, P., & Barton, G. (2023, June 27). Algorithmic bias detection and
mitigation: Best practices and policies to reduce consumer harms. The Brookings
Institution. https://www.brookings.edu/articles/algorithmic-bias-detection-andmitigation-best-practices-and-policies-to-reduce-consumer-harms/
Tutun, S., Johnson, M., Ahmed, A., Albizri, A., Irgil, S., Yesilkaya, I., Ucar, E. N., Sengun, T.,
& Harfouche, A. (2022). An AI-based decision support system for predicting mental
health disorders. Information Systems Frontiers, 25(3), 1261–1276.
https://doi.org/10.1007/s10796-022-10282-5
U.S. Department of Health and Human Services. (2022, October 9). Summary of the HIPAA
Privacy Rule. https://www.hhs.gov. Retrieved December 8, 2024, from
https://www.hhs.gov/hipaa/for-professionals/privacy/laws-regulations/index.html

207
Utah State Legislature. (2024). S.B. 149 Artificial Intelligence Amendments.
https://le.utah.gov/~2024/bills/static/SB0149.html
Vayena, E., Blasimme, A., & Cohen, I. G. (2018). Machine learning in medicine: Addressing
ethical challenges. PLoS Medicine, 15(11), Article e1002689.
https://doi.org/10.1371/journal.pmed.1002689
Virginia’s Legislative Information System. (2021, March 18). 2021 Special Session I.
https://legacylis.virginia.gov/cgi-bin/legp604.exe?212+ful+CHAP0219
Wang, J. (2023). The power of AI-Assisted diagnosis. ICST Transactions on e-Education and eLearning, 8(4), e3. https://doi.org/10.4108/eetel.3772
Wang, T., & Bashir, M. (2020). Privacy considerations when predicting mental health using
social media. Proceedings of the Association for Information Science and Technology,
57(1). https://doi.org/10.1002/pra2.244
Warraich, H. J., Tazbaz, T., & Califf, R. M. (2024, October 15). FDA perspective on the
regulation of artificial intelligence in health care and biomedicine. JAMA.
https://doi.org/10.1001/jama.2024.21451
Weber, M., Engert, M., Schaffer, N., Weking, J., & Krcmar, H. (2022). Organizational
capabilities for AI implementation—Coping with inscrutability and data dependency in
AI. Information Systems Frontiers, 25(4), 1549–1569. https://doi.org/10.1007/s10796-
022-10297-y
Wells, K. (2023, June 9). An eating disorders chatbot offered dieting advice, raising fears about
AI in health. NPR. https://www.npr.org/sections/healthshots/2023/06/08/1180838096/an-eating-disorders-chatbot-offered-dieting-adviceraising-fears-about-ai-in-hea

208
The White House. (2023). Blueprint for an AI Bill of Rights. Retrieved March 16, 2023, from
https://www.whitehouse.gov/ostp/ai-bill-of-rights/
Wieder, P., & Nolte, H. (2022). Toward data lakes as central building blocks for data
management and analysis. Frontiers in Big Data, 5, Article 945720.
https://doi.org/10.3389/fdata.2022.945720
Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Appleton, G., Axton, M., Baak, A.,
Blomberg, N., Boiten, J., Da Silva Santos, L. B., Bourne, P. E., Bouwman, J., Brookes,
A. J., Clark, T., Crosas, M., Dillo, I., Dumon, O., Edmunds, S., Evelo, C. T., Finkers, R.,
. . . Mons, B. (2016). The FAIR guiding principles for scientific data management and
stewardship. Scientific Data, 3(1). https://doi.org/10.1038/sdata.2016.18
Williams, R. (2023, May 24). Innovative app uses AI to battle rising veteran suicide rates.
Spectrum News 1. https://spectrumnews1.com/ca/la-west/mentalhealth/2023/05/24/innovative-app-uses-ai-to-battle-rising-veteran-suicide-rates#
Wimbarti, S., Kairupan, B. H. R., & Tallei, T. E. (2024). Critical review of self‐diagnosis of
mental health conditions using artificial intelligence. International Journal of Mental
Health Nursing, 33(2), 344–358. https://doi.org/10.1111/inm.13303
World Health Organization. (2021). Ethics and governance of artificial intelligence for health:
WHO guidance. https://www.who.int/publications/i/item/9789240029200
Zabelski, S., Hollander, M., & Alexander, A. (2024). Addressing inequities in access to mental
healthcare: A policy analysis of community mental health systems serving minoritized
populations in North Carolina. Administration and Policy in Mental Health and Mental
Health Services Research, 51(4), 543–553. https://doi.org/10.1007/s10488-024-01344-8

209
Zanna, K., Sridhar, K., Yu, H., & Sano, A. (2022). Bias reducing multitask learning on mental
health prediction. IEEE. https://doi.org/10.1109/acii55700.2022.9953850
Zeberga, K., Attique, M., Shah, B., Ali, F., Jembre, Y. Z., & Chung, T. S. (2022). A novel text
mining approach for mental health prediction using Bi-LSTM and BERT model.
Computational Intelligence and Neuroscience, 2022, 1–18.
https://doi.org/10.1155/2022/7893775
Zhang, M., Scandiffio, J., Younus, S., Jeyakumar, T., Karsan, I., Charow, R., Salhia, M., &
Wiljer, D. (2023). The Adoption of AI in Mental Health Care–Perspectives from Mental
Health Professionals: Qualitative Descriptive study. JMIR Formative Research, 7,
e47847. https://doi.org/10.2196/47847
Zhao, J., Chang, Y., Li, D., Xia, C., Cui, H., Zhang, K., & Feng, X. (2018). On retargeting the AI
programming framework to new hardwares. In Lecture notes in computer science (pp.
39–51). https://doi.org/10.1007/978-3-030-05677-3_4
Zidaru, T., Morrow, E., & Stockley, R. (2021). Ensuring patient and public involvement in the
transition to AI‐assisted mental health care: A systematic scoping review and agenda for
design justice. Health Expectations, 24(4), 1072–1124. https://doi.org/10.1111/hex.13299
Zine, M., Harrou, F., Terbeche, M., Bellahcene, M., Dairi, A., & Sun, Y. (2023). E-Learning
readiness assessment using machine learning methods. Sustainability, 15(11), 8924.
https://doi.org/10.3390/su15118924
Zona, M. A., & Thaib, I. (2020, November 14–15). Leadership and commitment to
organizational change of government employee in Padang, West Sumatra [Paper
presentation]. Sixth Padang International Conference on Economics Education,

210
Economics, Business and Management, Accounting and Entrepreneurship, Padang,
Indonesia. https://doi.org/10.2991/aebmr.k.210616.074
Zou, C. (2024). Revolutionizing machine learning: Harnessing hardware accelerators for
enhanced AI efficiency. Applied and Computational Engineering, 47(1), 141–146.
https://doi.org/10.54254/2755-2721/47/20241256

211
Appendix A: Interview Protocol
This section outlines the research questions, conceptual foundations, questions,
respondent criteria, an introduction, structured interview questions and probes, and an
introduction and conclusion for the subjects. The interview questions align with the research aim
to elicit insights into perceptions, organizational elements, and external factors of AI adoption in
mental healthcare settings.
Brief Statement of the Problem
While artificial intelligence promises advancements in mental health assessments,
interventions, and accessibility, adopting AI risks displacing human relationships and creating
harm if implemented without proper precautions (Fiske et al., 2019; Sinha et al., 2023). This
research explores organizational barriers and facilitators to ethically integrating AI in a way that
complements clinicians’ irreplaceable therapeutic role. Rather than focusing on technical details,
it investigates MHLs’ perceptions, organizational factors, and environmental pressures to
identify considerations for responsibly adopting AI to augment human-centered practice. This
exploratory research aims to help MHLs, ethicists, and technologists strategize how to leverage
AI’s potential in a way that preserves the humanistic core of mental healthcare.
Brief Statement of the Conceptual Framework
This study applied two complementary frameworks—the TOE framework—to develop a
comprehensive understanding of the factors influencing the ethical integration of AI in mental
healthcare settings. The TOE framework (Tornatzky et al., 1990) provides a lens to analyze the
technological, organizational, and environmental dimensions that shape AI adoption, including
organizational culture, resources, regulations, and ethical landscape.

212
Research Questions
1. What are the MHLs’ perceptions of adopting AI technologies in their organizations?
2. What organizational factors influence MHLs to adopt AI technologies responsibly in
their organization?
3. What external factors are MHLs concerned about the most when evaluating their
organization’s readiness to adopt AI technologies?
Respondent Type
This research utilized purposive sampling to recruit 11 U.S.-based leaders at various
levels of responsibility from community mental health and group practice organizations to
participate in interviews. Eligible leadership roles included frontline supervisors overseeing
employees, middle managers supervising frontline supervisors, senior executives supervising
middle managers, and thought leaders in advisory/consultative roles providing expertise on
technology, ethics, and mental healthcare policy. Participants were MHLs with a background in
community mental health or group practice settings and at least 1 year of experience in a
leadership position related to clinical operations and clinical technology, which may intersect
with a clinical license, though licensure is not mandatory for inclusion.
The goal is to target leaders with extensive mental health experience who are currently in
organizational decision-making roles related to administration and care delivery workflows.
Exclusions focus the sample on those in community-based organizations and group practice
settings rather than academics or solo private practices. Exclusion criteria are as follows:
• mental health professionals whose leadership experience is exclusive to self-owned,
single-provider practices.
• entry-level practitioners without leadership or management duties

213
Introduction to the Interview
Thank you for taking the time to participate in this interview. I am Mabel Yiu, and I am a
leadership and organizational change doctoral student from USC. I am conducting a study
exploring perspectives on adopting artificial intelligence technologies in mental healthcare
organizations. You have been selected for this interview because of your valuable insights as a
leader overseeing technology decisions that shape clinical practice.
The aim of this research was to uncover factors that influence AI adoption to develop
best practices for its ethical and effective implementation. My goal is to gather mental health
leaders’ views, motivations, and concerns to guide appropriate integration of AI that thoughtfully
augments human expertise and therapeutic relationships.
This semi-structured interview will take around 60 minutes and will be recorded for
analysis if you consent. All responses will remain confidential, and your identity will be
protected. There are no right or wrong answers; I am simply interested in understanding your
honest perspectives. I would like to emphasize that you are in control of this interview. If there
are any questions you do not wish to answer or topics you do not wish to discuss, please let me
know. The interview may also be terminated at any time. Our conversation will be kept
confidential; no identifiable information is recorded and any information you provide will be
used only for this research.
If you have any other questions, please let me know before we begin. May I have your
permission to record our discussion? Thank you, I truly appreciate your participation today.

214
Table A1
Interview Protocol
Interview questions Potential probes RQ
addressed
Key concept
addressed
What is your current
understanding of how AI is
being applied in mental
healthcare settings?
Can you give me an
example?
RQ1 TOE,
technology
How would you describe your
general perceptions and attitudes
toward adopting AI technologies
in your organization?
What factors shape your
overall opinion of AI
tools?
RQ1 TOE,
technology
In what ways could AI
technologies benefit your
organizations?
What opportunities excite
you the most about AI
tools?
Can you give me an
example?
RQ1 TOE,
technology
What potential risks do you
associate with AI adoption?
Can you give me an
example?
RQ1 TOE,
technology
What are some of the main
workflow or process needs you
feel AI could address in your
organization?
Where are current pain
points that technology
could help?
RQ2 TOE,
organizational
In your view, what barriers
currently exist that may inhibit
effective AI adoption?
What infrastructure gaps
would need to be
addressed?
RQ2 TOE,
organizational
How could leadership better
support adoption of new
technologies like AI?
Can you give me an
example?
RQ2 TOE,
organizational
How might introducing AI tools
impact the culture of clinical
teams and their openness to
integrating new technologies
into practice?
What cultural shifts or
training would be
needed?
RQ2 TOE,
organizational
Is there any reluctance or
resistance within your
organization to adopt AI?
Can you give me an
example?
RQ2 TOE,
organizational
How would your organization
respond if your employee
secretly used AI technology
such as using ChatGPT to
complete clinical notes?
Are there existing
policies in place to
detect and address
unauthorized use of AI
technology?
RQ2 TOE,
organizational
Could you share some insight into
who takes the lead on
What are the roles and
backgrounds of those
RQ2 TOE,
organizational

215
Interview questions Potential probes RQ
addressed
Key concept
addressed
establishing guidelines for the
ethical use of AI technologies
within your organization?
helping craft
responsible AI
policies?
What best practices in change
management would you
emphasize to successfully
integrate AI innovations into
your mental health services?
What training or support
would leadership need
to provide?
RQ2 TOE,
organizational
What external pressures or
competitive forces may
influence the need to adopt AI
technologies in your
organization?
Which environmental
factors most
significantly impact
technology decisions?
RQ3 TOE,
environmental
How does patient demand shape
your approach to AI tools?
Can you give me an
example?
RQ3 TOE,
environmental
How significant are regulatory
requirements and ethical
obligations in evaluating
whether to adopt an AI
technology?
Which regulations or
ethical codes are most
relevant?
RQ3 TOE,
environmental
How could AI tools impact
clinical effectiveness compared
to current human-delivered
practices?
What tasks or roles do
you feel should remain
strictly humandelivered?
RQ3 TOE,
environmental
What challenges, risks or
unintended consequences are
you most concerned about in
adopting AI technologies?
How might AI negatively
impact workplace
culture and human
roles?
RQ3 TOE,
environmental
Have you or your organization
seen vendors marketing their AI
tools for mental health?
Can you give me some
examples?
RQ3 TOE,
environmental
On an industry-wide level, what
groups or organizations do you
trust to lead the guidelines
development for implementing
AI in mental healthcare?
Who are the major
players influencing
those conversations?
RQ3 TOE,
environmental
Is there anything I didn’t ask that
you think is important?
What is your thought on
AI in the future of
mental health?
Closing Closing

216
Conclusion to the Interview
Thank you so much for taking the time to speak with me today and share your valuable
perspectives on AI adoption in mental healthcare organizations. I truly appreciate you providing
such thoughtful insights into the opportunities and challenges associated with integrating
emerging technologies into clinical practice and workflows. Your opinions and experiences will
be incredibly helpful for developing best practices to guide leaders in ethically and effectively
leveraging AI tools to augment human expertise. If any other reflections or considerations come
to mind after our discussion, please feel free to contact me. I sincerely thank you for contributing
your first-hand knowledge to this research and participating in this interview. It was a pleasure
speaking with you, and I wish you the very best.

217
Appendix B: Institutional Review Board Approval

Abstract (if available)

Abstract This research examined the technological, organizational, and environmental factors impacting how mental healthcare leaders responsibly adopt artificial intelligence (AI) technologies. Drawing from interviews with mental health leaders, the paper identifies key factors, including leaders’ perceptions of AI’s capabilities and limitations, the organization’s technological readiness and change management processes, and broader industry pressures. While ethical AI principles focused on patient autonomy, avoiding harm, and promoting well-being are essential, this paper explored the meaning of “responsible” in AI adoption that requires mental health leaders to meet ethical standards and take direct accountability for outcomes, anticipate potential challenges, obtain stakeholders’ buy-in, and maintain the adaptability as healthcare needs evolve. The findings reveal that while most interviewed mental health leaders are not technical experts in AI, they share a cautiously optimistic outlook about AI’s potential to enhance treatment effectiveness and reduce provider burnout, tempered by concerns about the absence of comprehensive guidelines, data integrity, and algorithmic accuracy. Leaders specifically expressed apprehension about AI’s cultural competency limitations, the risk of over-reliance and ethical dissonance, funding inequities, and potential impacts on therapeutic rapport in mental healthcare delivery. By understanding the multifaceted influences on leaders, the research aims to provide a practical guide for harnessing the benefits of AI while preserving the essential human elements of compassionate, patient-centered care, ensuring AI augments rather than replaces clinical expertise.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Ambient anxiety within leadership teams and its impact on organizational efficiency in mental health organizations

PDF

Reducing misdiagnosis in mental health professions: a study of a promising practice

PDF

The barriers and challenges associated with mental health help-seeking behaviors of police officers in the United States: a descriptive study

PDF

Evaluating the pursuit of advanced degrees in a health information profession: a gap analysis study

PDF

Adolescent mental health services in Nevada: assessing the perspectives of families

PDF

Health system mergers: the significance of leaders

PDF

Collaboration, capacity, and communication: Leaders’ perceptions of innovative work behavior across hybrid and remote work environments

PDF

Exploring generative AI in business education: faculty perceptions and usage

PDF

Gaps in organizational mental health mitigation within the United States Fire Service

PDF

Students with disabilities in higher education: examining factors of mental health, psychological well-being, and resiliency

PDF

The experience of Eritrean refugee women in addressing their mental health needs

PDF

Aligning digital technology to support motivation, physical activity, and sports

PDF

The first-time manager journey: a study to inform a smoother leadership transition

PDF

Evaluation of mental health support and needs for emergency department providers

PDF

Improving student-athlete mental health services: addressing the mental health needs of college student-athletes

PDF

Digital transformation in the resources industry: an exploration of promising management practices

PDF

Failing toward success: a learn-from-failure mindset and its impact on institutional effectiveness in higher education

PDF

The relationship between Latinx undergraduate students’ mental health and college graduation rates

PDF

U.S. Army Reserve: the journey to psychological health resources

PDF

Transgender patients’ perceptions of healthcare: A study of gender minority stress and resilience factors in predicting healthcare behavioral intentions

Asset Metadata

Creator Yiu, Mabel (author)

Core Title Responsible AI adoption in community mental health organizations: a study of leaders’ perceptions and decisions

School Rossier School of Education

Degree Doctor of Education

Degree Program Organizational Change and Leadership (On Line)

Degree Conferral Date 2025-05

Publication Date 02/05/2025

Defense Date 01/15/2025

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag artificial intelligence,change management,community mental health,digital transformation,ethics,leadership,mental health,responsible AI,technology-organization-environment

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Tobey, Patricia (committee chair), Chung, Ruth (committee member), Rizzo, Albert (committee member)

Creator Email mabelyiu@gmail.com,mabelyiu@usc.edu

Unique identifier UC11399GTHB

Identifier etd-YiuMabel-13815.pdf (filename)

Legacy Identifier etd-YiuMabel-13815

Document Type Dissertation

Format theses (aat)

Rights Yiu, Mabel

Internet Media Type application/pdf

Type texts

Source 20250211-usctheses-batch-1241 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu