Responsible AI

Responsible AI can transform sickle cell care — but only if its risks are measured, not merely acknowledged. Every system we build is designed against five commitments.

• Accuracy we can measure. We hold our systems to the highest accuracy we can achieve, and we test for it rather than assume it.
• Bias we actively look for. Sickle cell care is shaped by long-standing inequities. AI trained on that record inherits them unless someone goes looking.
• Accessible by design. Outputs and interactions are written for the people who will actually read them — patients, caregivers, and clinicians alike.
• A complement to clinical judgment, never a replacement. Our tools are decision support. The decision stays with the human.
• Humans in the loop. Expert reviewers evaluate how our systems perform and give us semantic, accessibility, and ethical feedback that we fold back into the product.

We shared this vision at Warrior Con 2026, the 13th Annual Sickle Cell Warriors Convention in Los Angeles — view the presentation.

Our systems draw on a curated knowledge base: current reference textbooks, seminal peer-reviewed articles, and indexes of trusted medical information such as the NIH. If you would like to review our sources — or propose one we have missed — please get in touch.

Nothing SickleCellPedia provides is medical advice. It is a medical information tool, not a substitute for care from a qualified professional. Medical advice requires clinical judgment applied to a specific person in a specific context, and that is not something a single-turn AI tool is equipped to provide. Always consult your care team before acting on anything you read here.

Pro is a decision-support tool built for clinicians, wherever they practice. It is designed to be more robust than the public version:

• Multi-agent architecture
• Mandatory citation of consensus guidelines
• The ability to enter case-specific information
• Chain-of-thought reasoning
• Credential-based access

Learn more on the SickleCellPedia Pro page, where clinicians can register their interest.

SickleCellPedia does not currently cover mental health needs, nor does it help patients navigate access to care within their own health system. Both matter. Both also fall outside the scope of this version, because both depend heavily on local and personal context. Our focus for now is consensus-based medical information that holds across settings.

We will present benchmark results for SickleCellPedia at ASCAT in October 2026.

Human-in-the-loop surveys. To benchmark SickleCellPedia against general-purpose AI tools (ChatGPT, Claude, Gemini), we ask our expert collaborators to evaluate the responses each tool produces. Our method follows the approach used in HealthBench (Arora et al., 2025).

We never sell your information, and we do not use it without your express permission. Read our privacy policy.