Meet OpenAI's doctor-whisperer making ChatGPT better at talking health
Rebecca Soskin Hicks, a physician who joined OpenAI in 2024, coordinates a network of doctors who assess how ChatGPT handles health discussions. Their reviews look for confusing answers and responses that could endanger users.
The part-time contractors work across countries and specialties. By July, they had examined more than 700,000 ChatGPT responses; their assessments identify weaknesses but are not used directly as AI training data.
OpenAI’s research team uses the feedback to explore data and training approaches for better emergency triage, follow-up questions, communication of uncertainty and suitable levels of detail.
The work comes amid concerns about chatbot health advice. A pastor sued OpenAI in July, alleging GPT-4o mishandled information about his lung condition and seeking a pause in health-related products. Soskin Hicks said focusing solely on the chatbot can oversimplify medical decisions.
OpenAI has added connections to Apple Health and some medical records, and released a benchmark for mental health responses. Soskin Hicks regards evaluations as early evidence and hopes future studies will show whether health chatbots improve outcomes.
Key points
- A global network of contracted doctors reviews ChatGPT health responses for risks and weaknesses.
- The doctors had reviewed over 700,000 responses by July; their assessments are not direct training data.
- OpenAI hopes future studies will establish whether chatbot health advice improves health outcomes.