In this episode of the a16z podcast, host Jay Rugani is joined by Karin Singhal, who leads health AI research at OpenAI, to discuss the transformative role of artificial intelligence in healthcare.
The conversation begins by examining how OpenAI integrates specialized healthcare data into its general-purpose models across every stage of training. Rather than relying on smaller, isolated specialized models that quickly fall behind the technological frontier, OpenAI hitches a ride on the scale and reasoning improvements of general frontier models. This approach allows ChatGPT to serve massive numbers of users—over 300 million people weekly for health and wellness questions—while offering tailored behavior for lay users, expert patients, and clinicians alike.
Rugani and Singhal discuss the harsh realities of the current healthcare system, characterized by massive financial spend, severe workforce scarcity, and a chronic lack of primary care access. To combat this, OpenAI's mission is to raise the floor of human health by providing abundant, trustworthy health intelligence, and to raise the ceiling by making the impossible achievable. Singhal shares a touching anecdote about Nathan LeBenz, who used AI to help navigate his young son’s serious leukemia diagnosis. They also explore the integration of AI into complex workflows, such as OpenAI's collaboration with Epic, enabling providers to seamlessly utilize ChatGPT for chart review and safety-net support directly inside their everyday EHR software.
Trust, correctness, and medical consensus form a core pillar of the discussion. Singhal explains that OpenAI relies on a global network of hundreds of medical experts across dozens of specialties to evaluate model performance and establish behavioral guidelines for tricky, uncertain, or nuanced medical scenarios. This process ensures that the models prioritize patient safety, even when a user might prefer comforting reassurance over medical reality. They also tackle the evolving intersection between provider and payer systems, administrative friction, and the potential for AI agents to ultimately compress back-office costs and simplify care journeys.
Looking toward the future, Singhal envisions a shift from mere medical information to proactive, personal health assistants that act as a "quarterback" for chronic care, flag medication adherence issues, and help users navigate logistics. While AI will not replace the physical and empathetic aspects of traditional medical care, it will act as a powerful catalyst for the healthcare ecosystem. Finally, Singhal expresses strong optimism for the next five to ten years, predicting that AI-guided target generation, molecule discovery, and clinical trial prioritization will dramatically accelerate the development of personalized medicines.