Building Shared Conceptual Grounding for Interacting with Generative AI

Explore how researchers are addressing the fundamental challenge of human-AI collaboration in generative AI systems through this 47-minute Stanford University research presentation. Discover the core problem of "theory of mind" gaps between users and black-box AI systems, where AI often misinterprets user intent and users lack predictive models for AI behavior, leading to inefficient trial-and-error workflows. Learn about innovative approaches to building shared conceptual grounding that enables both humans and AI to better simulate and predict each other's operations when working with text-to-visual content generation tasks. Examine research methodologies and findings that aim to transform the current paradigm of repeated prompt experimentation into more intuitive and effective human-AI partnerships. The session includes a comprehensive lecture followed by an interactive Q&A segment, providing insights into cutting-edge work supported by Stanford's Hoffman-Yee Research Grant program on improving the collaborative potential of generative AI tools.