To develop a practical, disclosure-oriented ethical framework for AI in digital mental health, emphasizing honesty about the limitations and biases of AI systems, particularly concerning neurodivergent populations and their ethical implications.
Approach:
Feynman's Principle: Translate Feynman's principle of scientific integrity into actionable criteria for AI integrity in digital mental health, focusing on implementation strategies.
Neurodiversity Stress Test: Employ the neurodiversity paradigm to examine empirical evidence of AI bias against neurodivergent populations, providing specific examples.
Feynman Honesty Standard: Propose a five-criterion Feynman Honesty Standard and a forward research agenda, detailing how each criterion can be applied.
Key Findings:
AI tools trained on narrow behavioral and linguistic norms risk misrepresenting neurodivergent users.
Existing ethical frameworks have not adequately addressed the need for candor in AI deployment.
The field of digital mental health must prioritize transparency about what AI systems know and their limitations.
Interpretation:
The analysis highlights the need for an honesty standard in AI development to ensure ethical practices, especially for vulnerable populations.
Limitations:
The empirical evidence primarily addresses autism and ADHD, limiting generalizability to other neurodevelopmental conditions and their implications.
Conclusion:
The field will be evaluated based on both algorithmic performance and transparency regarding the capabilities and limitations of AI systems.
