JAMA+ AI Conversations Designing AI for Uncertainty: A Conversation With Eric Horvitz
Nov 20, 2025
Eric Horvitz, Chief Scientific Officer at Microsoft and a pioneer in AI research, dives into the future of AI in medicine with Yulin Hswen. They explore how AI can safely reason in clinical settings, emphasizing the need for integrity and safety. Horvitz shares insights on the evolution of AI, from Bayesian methods to the marvels of modern generative models. The conversation also tackles public fears about AI, cognitive complacency, and the importance of calibration and reproducibility in AI systems to ensure scientific integrity. Horvitz underscores the interdisciplinary approach necessary for tackling AI's long-term societal impacts.
AI Snips
Chapters
Transcript
Episode notes
From Neurons To Computation
- Eric Horvitz recounts shifting from neuroscience to AI after studying single neurons and asking how sparse spikes create rich minds.
- He left medical school courses to pursue graduate work in AI while keeping medicine as an application domain.
Two Major AI Inflections
- Horvitz highlights two AI inflections: Bayesian/statistical methods and the neural-net renaissance fueled by data and self-supervision.
- Modern large generative models show unexpected, polymathic abilities that echo deep questions about mind.
Dual-Edged Social Impact
- AI brings both threats to democracy via realistic disinformation and opportunities to solve grand challenges like climate and disease.
- The societal impacts span identity, agency, jobs, and education, requiring multidisciplinary attention.