Our Core Commitment
This research exists to improve healthcare AI, not to attack it. We believe AI can be a powerful force for healthcare equity—but only if we honestly examine and address its limitations.
Ethical Principles
Synthetic Data Only
All symptom profiles are completely fictional. No real patient data is used. Every "patient" in our study is a synthetic construct designed to test system behavior.
Established Methodology
Our methods are based on peer-reviewed research published in top journals (Science, PNAS, NEJM). We're applying proven techniques to a new domain.
Constructive Intent
Our goal is to help AI developers improve their systems. We want healthcare AI to work well for everyone—bias detection is the first step toward bias correction.
Responsible Disclosure
We share findings with AI developers before public release. They get 90 days to respond. Their responses are included in our publication.
What We Do NOT Do
- We do not use real patient data
- We do not attempt to break or exploit AI systems
- We do not publish findings without responsible disclosure
- We do not exaggerate or sensationalize results
- We do not name systems without giving them opportunity to respond
- We do not use this research to attack companies or individuals
Human Subjects Considerations
No Human Subjects Involved
This research does not involve human subjects. All data consists of:
- Synthetic symptom profiles (fictional)
- Name pairs selected from published research
- AI system outputs (text responses)
No real patients are queried. No personal health information is collected or used.
Terms of Service Compliance
Using Systems as Intended
Our research involves using consumer AI systems exactly as they're designed to be used—entering symptoms and receiving recommendations. We:
- Do not circumvent access controls
- Do not overload systems with automated requests
- Do not scrape or extract proprietary data
- Use only publicly available interfaces
Responsible Disclosure Process
Complete Analysis
Finish all data collection and statistical analysis. Ensure findings are robust and reproducible.
Developer Notification
Contact each AI system developer with specific findings related to their system. Provide full technical details.
Response Period
Allow 90 days for developers to respond, investigate, and potentially address issues.
Include Responses
Incorporate developer responses into the final publication. Their perspective matters.
Public Release
Publish findings with full methodology, data, and developer responses.
Why This Research Matters
The Stakes Are High
Healthcare AI is being deployed at scale. If these systems exhibit bias:
- Millions of users receive inconsistent recommendations
- Existing healthcare disparities may be amplified
- Trust in AI-assisted healthcare erodes
- Underserved populations bear disproportionate burden
Identifying bias is the prerequisite to fixing it.
Our Hopes
For AI Developers
We hope our findings help you build fairer systems. Bias often exists unintentionally—awareness enables improvement.
For Healthcare Providers
We hope this research helps you understand the limitations of AI tools and when to apply clinical judgment.
For Patients
We hope this contributes to a future where AI-assisted healthcare serves everyone equally.
For Researchers
We hope our methodology enables continued scrutiny and improvement of healthcare AI fairness.
Contact
Questions about our ethical framework? Concerns about our methodology? We welcome dialogue.
For AI developers: If you believe your system is included in our research and wish to engage, please contact us. We are committed to collaborative improvement.
Learn More
Explore our research protocol or see the methodology in detail.