Ethical AI: Upholding best practices in responsible mental wellness AI development

Ensuring ethical AI: Upholding best practices in student mental wellness and responsible AI development

In the evolving landscape of innovative mental wellness support, the integration of generative-AI-powered tools offers promising advancements. However, the ethical development and implementation of these tools are paramount to ensure that they provide effective and reliable support to users. At Wayhaven, we are dedicated to upholding best practices in college mental wellness and responsible AI development. Our approach encompasses stringent privacy policies, a clear informed consent process, transparent communication strategies, and responding appropriately to mental wellness risks. Additionally, we emphasize reducing bias, promoting cultural sensitivity, and enhancing equitable access. Wayhaven is committed to evidence-based methods and interdisciplinary collaboration as we move our work forward. 

‍

Privacy, consent, and communication

Privacy and confidentiality

Privacy and confidentiality are crucial when leveraging LLMs to support mental wellness, as emphasized by Lawrence et al. (2024). Wayhaven's approach to privacy is underscored by its detailed privacy policy, which outlines the specific handling of user data to ensure safety and reliability. Wayhaven maintains strict data privacy by reviewing user data solely to enhance service quality while sharing anonymous information with researchers and educational institutions.

Informed consent

Obtaining informed consent is critical when users engage with mental wellness LLMs (Lawrence et al., 2024). Wayhaven explicitly secures user consent for data collection and usage, in line with best practices advocated by Lai et al. (2023). During the onboarding process, Wayhaven captures user consent through a clear and understandable interface, confirming that users are fully aware of the terms and privacy conditions before proceeding.

Clear communication

Effective communication of privacy policies and informed consent is essential for fostering trust and openness. Wayhaven conveys these elements in simple, succinct language during our onboarding process, avoiding ‘legalese’. This approach helps reduce withholding behaviors and encourages increased trust and disclosure (Bucher, 2020; Lawrence et al., 2024; Song et al., 2024; Tong, 2023; Wutz et al., 2023). 

User autonomy and data protection

Wayhaven prioritizes user autonomy, data protection, and privacy by employing advanced data protection measures such as end-to-end encryption and secure data storage. Users are provided with control over their personal wellness data, including mechanisms to consent to data sharing or to delete their personal information. These practices align with the principles outlined by Golden & Aboujaoude (in press). Wayhaven presents these options clearly and simply during the onboarding process such that users understand their rights and choices regarding data handling.

‍

Ensuring transparency and explainability 

Mental wellness LLMs should be transparent and capable of explanation, as emphasized by Lawrence et al. (2024) and You et al. (2023). At Wayhaven, we believe that transparency is a cornerstone of building trust with our users, providing details such as the composition of our development team (Golden & Aboujaoude, in press). We explicitly inform users to exercise caution when interpreting or acting on LLM output and are clear about the bounds of our AI’s competence (You et al., 2023). We are also careful to convey to users that they are interacting with an AI, not a human. Wayhaven is constantly expanding its 'self-awareness,' an inventory of responses to common user questions. This allows users to understand how the AI works and what they can expect from their interactions with Wayhaven. Themes are particularly focused on explainability, such as how responses are developed, the AI’s capabilities, and its limitations.

‍

Commitment to evidence-based methods

Integrating proven approaches

At Wayhaven, we are dedicated to drawing wholly from evidence-based principles and practices, well-supported by current research in mental wellness interventions. We integrate a variety of well-established approaches, including Cognitive Behavioral Therapy (CBT), Dialectical Behavior Therapy (DBT), Acceptance and Commitment Therapy (ACT), positive psychology, and mindfulness-based cognitive therapies. This alignment with current evaluative frameworks for mental wellness LLMs (Golden & Aboujaoude, in press; Stade et al., 2024) optimizes the chance of creating meaningful impact (Stade et al., 2024). This principle is central to our development process, striving to provide reliable and effective support to users.

Prioritizing impactful improvement

While promoting engagement is critical for achieving an adequate amount of interaction with the mental wellness intervention, it is essential that engagement contribute to meaningful interactions with an app’s key active ingredients. At Wayhaven, we prioritize meaningful improvement, such as increases in well-being and functioning (Stade et al., 2024), focusing on making our interventions both engaging and beneficial to our users.

Rigorous evaluation and ongoing research

Wayhaven is committed to rigorous evaluation, prioritizing feasibility, acceptability, and efficacy in our research efforts. This approach aligns with existing recommendations for the evaluation of digital mental wellness applications (Stade et al., 2024). To further validate our approach, Wayhaven will undergo a feasibility and acceptability trial this fall, including preliminary indicators of efficacy. Following this, we plan to conduct a randomized controlled trial (RCT) to more rigorously validate Wayhaven’s effectiveness. This commitment to continuous improvement helps maintain our platform as a trusted resource for mental wellness support.

‍

Responding appropriately to mental wellness risks

Comprehensive safety protocols 

Mental wellness LLMs must respond appropriately to mental wellness risks, as underscored by Lawrence et al. (2024). At Wayhaven, we demonstrate comprehensive safety protocols and crisis management features designed to provide real-time crisis interventions, as recommended by Golden & Aboujaoude (in press).

Wayhaven has detailed protocols for multiple risk scenarios, ensuring that our AI is trained to detect various potential risk scenarios in user language and to respond in appropriately tailored ways. Importantly, Wayhaven does not attempt to perform a risk assessment, which we believe is currently beyond the capacity of an LLM, nor does it ask the user to perform a risk assessment on themselves.

Encouraging engagement and reducing overload

Golden and Aboujaoude (in press) highlight the need for mental wellness AIs to display not only comprehensive risk management protocols, but also to include strategies to optimize user engagement with resources surfaced. Wayhaven provides direct links to relevant crisis and emergency services to minimize friction and limits the resources provided to three or fewer to prevent cognitive overload. It also consistently includes a chat option for students who prefer text over phone. 

There is limited research on ways that crisis response protocols may be designed to empower users and to increase follow-through with the presentation of support options (Cohen et al., 2023). Existing research indicates that simply providing more resources does not necessarily help and may even harm users if the resources do not meet their needs (Jacobson et al., 2022). Therefore, we carefully select and limit the resources presented to ensure that they are accessible and effective.

We draw on single-session intervention (SSI) principles, such as offering users choices to boost autonomy and including testimonials from others facing similar challenges to enhance message credibility to improve the uptake of mental wellness crisis resources (Cohen et al., 2023). When a potential crisis is detected, our system surfaces both local and national resources, preserving user autonomy by always listing both types and providing students a choice.

Wayhaven’s comprehensive safety protocols, crisis management features, and adherence to ethical AI principles enable us to respond appropriately to mental wellness risks, preserving user autonomy and providing compassionate support during moments of utmost need.

‍

Reducing bias, promoting cultural sensitivity, and enhancing equitable access

There is a risk that LLMs could perpetuate inequities and stigma, further widening mental wellness disparities (Koutsouleris et al., 2022). Mental wellness concerns are highly stigmatized (Sickel et al., 2014), and there are notable inequities in who receives mental wellness care (McGuire & Miranda, 2008). At Wayhaven, we recognize these challenges and actively work to mitigate them. 

Wayhaven is committed to advancing mental wellness equity, as emphasized by Lawrence et al. (2024) and Golden & Aboujaoude (in press). We seek to expand access to mental wellness support for underrepresented populations, such as BIPOC, first-generation, low-income, LGBTQ+, and international students. These groups are more likely to experience negative mental wellness outcomes and less likely to receive support from traditional college counseling centers (Abelson et al., 2024). By leveraging AI, we aim to deliver mental wellness interventions in regions where access to mental wellness providers is limited and where significant barriers, such as cost, may exist (Lawrence et al., 2024). 

We proactively work to reduce bias, focusing our efforts on incorporating cultural diversity and inclusivity in all interactions and ensuring that our AI is designed and implemented with a deep understanding of these critical principles (Golden & Aboujaoude, in press). By integrating preventative strategies and considering the unique needs of underrepresented populations from the outset, we strive to provide effective, equitable support. Our approach includes creating a diverse student advisory board offering insights and feedback throughout the product development process to ensure that our platform considers various perspectives and experiences. We train our AI to use respectful language and acknowledge diverse experiences, fostering an environment of understanding and respect. From project design to evaluation, we actively consider racial and cultural equity in our processes. This commitment extends to our content creation, where we strive to reflect and respect diverse cultural backgrounds, identities, and socioeconomic situations. Regular reviews of our AI’s responses help us identify and address any potential biases, optimizing the fair and unbiased nature of conversations.

‍

Actively promoting user empowerment

Ethical development in AI-powered mental wellness tools entails actively promoting user empowerment, minimizing potential dependency by offering tools and resources that encourage self-efficacy in managing mental wellness (Golden & Aboujaoude, 2024). Wayhaven routinely engages users in generating action plans, determining when they will practice certain skills beyond the app, and anticipating and problem-solving potential obstacles. We also surface opportunities for users to exercise informed choice over their mental wellness journeys, aligning with best practices in mental wellness AI (Golden & Aboujaoude, in press).

Wayhaven allows users to make choices at multiple points during their interaction with the platform. For instance, users can select the approach or modality with which they prefer to work. This ensures that users can choose the method that resonates most with them, whether it be CBT, DBT, ACT, or another evidence-based practice. Additionally, users have the flexibility to pick a different tool or opt out of a tool at any time, allowing them to switch between tools based on their evolving needs and preferences.

Furthermore, users have control over the pacing and frequency of their conversations, enabling them to engage with the platform at a rate that suits their wants, needs, and schedules. By offering these choices, Wayhaven supports user autonomy and fosters a sense of ownership over their mental wellness journey.

‍

The power of interdisciplinary collaboration

The importance of interdisciplinary teamwork

In developing mental wellness LLMs, interdisciplinary collaboration between mental wellness experts, engineers, and technologists is essential. While engineers might theoretically use manuals to build these models, this method falls short without the nuanced expertise of mental wellness experts. Manuals provide a foundation but lack the detailed guidance necessary for applying interventions to specific individuals and managing other issues that arise (Stade et al., 2024). Mental wellness professionals offer crucial insights, such as identifying limitations and risks, addressing the complexity of emotional phenomena, ensuring ethical development, preventing negative effects, and fine-tuning models for personalized interventions. For example, mental wellness experts can play a key role in prompt engineering by designing and testing prompts that lend the LLM the proper context for delivering specific wellness techniques. This ensures that the app not only performs technically but also meets the wellness needs of users effectively and ethically. 

Wayhaven's unique approach 

Wayhaven was co-designed by technologists and mental wellness professionals from its inception. This cross-functional collaboration is a cornerstone of our development process. From the beginning, prompt engineering and testing have represented a partnered effort between technologists and psychologists, contributing to a platform that is both technically robust and tailored to users’ mental wellness needs. This collaborative approach allows Wayhaven to harness the strengths of both disciplines, resulting in a more powerful, user-centered mental wellness tool. 

‍

Conclusion

Wayhaven’s dedication to ethical AI development in mental wellness is demonstrated through our comprehensive safety protocols and commitment to rigorous ongoing research efforts. By actively promoting user empowerment and fostering a collaborative approach between technologists and mental wellness professionals, we maintain a platform that is both technically robust and user-centered. As we continue to advance our AI-powered tools, we remain devoted to providing reliable, effective, and compassionate support tailored to the diverse needs of our users. Through these efforts, Wayhaven strives to lead the responsible transformation of college mental wellness support, making it accessible and equitable for all students.

‍

Sources

Abelson, S., Eisenberg, D., Lipson, S. K., & Schueller, S. M. (2024, June 24). Digital Mental Health Programs at Colleges and Universities Webinar [Webinar].

Bucher, A. (2020). Engaged: designing for behavior change. Rosenfeld Media.

Cohen, K., Dobias, M. L., Morris, R., & Schleider, J. L. (2023). Improving uptake of mental health crisis resources: Randomized test of a single-session intervention embedded in social media. Journal of Behavioral and Cognitive Therapy, 33, 24-34.

Golden, A., & Aboujaoude, E. The Framework for AI Tool Assessment in Mental Health (FAITA - Mental Health): a scale for evaluating AI-powered mental health tools. World Psychiatry (in press).

Jacobson, N. C., Yom-Tov, E., Lekkas, D., Heinz, M., Liu, L., & Barr, P. J. (2022). Impact of online mental health screening tools on help-seeking, care receipt, and suicidal ideation and suicidal intent: Evidence from internet search behavior in a large US cohort. Journal of psychiatric research, 145, 276-283.

Koutsouleris, N., Hauser, T. U., Skvortsova, V., & De Choudhury, M. (2022). From promise to practice: towards the realisation of AI-informed mental health care. The Lancet Digital Health, 4(11), e829-e840.

Lai, T., Shi, Y., Du, Z., Wu, J., Fu, K., Dou, Y., & Wang, Z. (2023). Supporting the Demand on Mental Health Services with AI-Based Conversational Large Language Models (LLMs). BioMedInformatics, 4(1), 8-33.

Liao, Y. (2021). Design and evaluation of a conversational agent for mental health support: forming human-agent sociotechnical and therapeutic relationships (Doctoral dissertation, University of Maryland, College Park).

McGuire, T. G., & Miranda, J. (2008). New evidence regarding racial and ethnic disparities in mental health: Policy implications. Health affairs, 27(2), 393-403.

Sickel, A. E., Seacat, J. D., & Nabors, N. A. (2014). Mental health stigma update: A review of consequences. Advances in Mental Health, 12(3), 202-215.

Song, I., Pendse, S. R., Kumar, N., & De Choudhury, M. (2024). The typing cure: Experiences with large language model chatbots for mental health support. arXiv preprint arXiv:2401.14362.

Stade, E. C., Stirman, S. W., Ungar, L. H., Boland, C. L., Schwartz, H. A., Yaden, D. B., ... & Eichstaedt, J. C. (2024). Large language models could change the future of behavioral healthcare: a proposal for responsible development and evaluation. NPJ Mental Health Research, 3(1), 12.

Tong, F., Lederman, R., D'Alfonso, S., Berry, K., & Bucci, S. (2023). Conceptualizing the digital therapeutic alliance in the context of fully automated mental health apps: A thematic analysis. Clinical Psychology & Psychotherapy, 30(5), 998-1012.

Wutz, M., Hermes, M., Winter, V., & Köberlein-Neu, J. (2023). Factors influencing the acceptability, acceptance, and adoption of conversational agents in health care: integrative review. Journal of Medical Internet Research, 25, e46548.

You, Y., Tsai, C. H., Li, Y., Ma, F., Heron, C., & Gui, X. (2023). Beyond self-diagnosis: how a chatbot-based symptom checker should respond. ACM Transactions on Computer-Human Interaction, 30(4), 1-44.

‍