From Single Neurons to AI Systems: The Evolution of Decision Sciences in Medicine with Dr. Eric Horvitz
Mar 20, 2024
auto_awesome
Dr. Eric Horvitz shares his journey from neurobiology to AI in medicine. Discusses importance of probabilistic models and decision theory in AI. Reflects on responsible AI development. Explores applications of GPT-4 in medicine and legacy of ethical AI leadership. Navigating complex terrain of AI ethics.
Dr. Eric Horvitz's transition from neurobiology to AI highlights the evolution of decision sciences in medicine.
Probabilistic models and decision theory play a crucial role in AI for unraveling complex human decision-making processes.
Balancing excitement and caution, responsible AI development is essential for integrating AI technologies into various aspects of human life.
Deep dives
Founding of Microsoft Research
Microsoft Research was founded in 1991 following a discussion between Bill Gates and Nathan Mervold. The focus on AI was emphasized, with a small initial team working on probabilistic and Bayesian inference. The lab grew rapidly and established principles like researcher autonomy in publishing, fostering a unique environment. Over time, the lab expanded to cover various areas of computer science, offering a blend of applications and foundational work.
Dr. Eric Horvitz's Background and Work at Microsoft
Dr. Eric Horvitz, the Chief Scientific Officer of Microsoft, shared his journey from a biophysics undergrad degree to delving into neurobiology and artificial intelligence. His unique MD and PhD background led to significant contributions, such as developing a Bayesian approach for identifying spam emails. Dr. Horvitz highlighted his work on AI in medical diagnosis and the 100-year AI study at Stanford. His involvement in advising on AI policy at a White House level showcased his diverse impact across computer science, medicine, and public policy.
Application of Decision Theory in AI and Healthcare
Dr. Horvitz's deep dive into decision theory and its application in AI and healthcare was discussed. Decision theory involves assessing uncertainties in medical decisions, calculating expected values, and factoring in patient preferences. Dr. Horvitz explored using large language models like GPT-4 to improve Bayesian diagnosis and decision making, emphasizing the need to blend traditional decision theory with modern AI capabilities. The MedPrompt technique showcased inferences and the importance of crafting effective prompts for language models in medical scenarios.
AI Models Can Play Multiple Roles in Problem Solving
Models like mid-prompt have shown the ability to generate few-shot reasoning chains and improve problem-solving. The use of AI models in various roles, such as critiquer and generator, shows versatility in problem-solving approaches. These models can handle multiple phases of problem-solving and prompting, leading to the development of multi-agent solutions like auto-gen.
Microsoft's Focus on Responsible AI and The 100-Year Study on AI
Microsoft has established AI principles and the Ether Committee to address responsibilities in AI technology development. The 100-year study on AI, initiated by Stanford with Microsoft's endowment, aims to provide proactive guidance over the long term. This study focuses on advising government, civil society, and academia every five years to address evolving AI challenges and opportunities.
In this episode of the AI Grand Rounds podcast, Dr. Eric Horvitz describes his career evolution from an interest in neurobiology to significant contributions in AI, particularly in understanding complex systems and applying AI in medicine. He discusses the shift from studying neurobiology to embracing AI and computational methods as tools for unraveling the complexities of the human mind and broader decision-making processes. Horvitz emphasizes the importance of probabilistic models and decision theory in AI, highlighting his work on bounded rationality and the challenges of interpretability in AI systems. He also reflects on the potential of AI in medicine, the necessity of responsible AI development, and the future of AI research. He suggests a blend of excitement and caution as AI technologies become increasingly integrated into various aspects of human life and decision making.
