Custom, Targeted, Compact, and Neuro-symbolic AI Models for Health Applications

This talk explores the development of next-generation AI models for healthcare applications that go beyond current generative AI limitations. Learn why trustworthy AI requires user-level explainability, safety, privacy, attribution, reliability, consistency, and adherence to clinical guidelines. The 47-minute presentation demonstrates how neurosymbolic approaches can utilize diverse knowledge and engage domain experts in model validation, while explaining the benefits of custom, targeted AI that addresses specific health needs with high-quality data. Discover the advantages of compact, right-sized models that can be deployed on edge devices in clinics or consumer devices. The speaker illustrates these concepts through real-world applications in nutrition and mental health, showing how this new generation of AI models overcomes the deficiencies of purely data-driven and generative AI approaches to better support patients and clinicians through chatbots, copilots, and agents.