How AI Is Making Medical Expertise More Accessible for Everyone

Medical expertise has long been unevenly distributed — accessible to some, out of reach for many. Karan Singal, a researcher at OpenAI with roots in health AI development through Google's Med-PaLM project, is working to change that through AI systems designed for real clinical and patient use.

In this NEJM podcast discussion, Singal outlines how AI tools are already being used by both clinicians and patients to answer medical questions, inform care decisions, and help individuals better advocate for themselves within complex health systems. This democratization of expertise is central to his vision: AI not as a replacement for physicians, but as a force multiplier for medical knowledge.

Two key projects anchor his work. HealthBench is an evaluation framework designed to rigorously test whether health AI systems are safe, accurate, and genuinely useful — addressing a critical gap in how medical AI is assessed before deployment. ChatGPT for Clinicians is a purpose-built tool aimed at supporting healthcare professionals in real-world settings, integrating AI assistance directly into clinical workflows.

For health-optimizing individuals, the implications are significant. Better AI tools could mean faster access to evidence-based guidance, more informed patient-doctor conversations, and improved ability to interpret personal health data. Singal emphasizes that trustworthiness and clinical responsibility must remain central as these systems scale.

Caveats remain important. AI in medicine is still maturing, and even well-evaluated systems can produce errors or miss nuance that experienced clinicians would catch. The shift toward AI-assisted care requires careful oversight, strong evaluation standards like HealthBench, and ongoing collaboration between technologists and medical professionals to ensure patient safety is never compromised in the pursuit of accessibility.