The Role of Theory in Trustworthy & Interpretable AI

Overview

Recent progress in generative AI—Large Language Models, Diffusion Models, and beyond—has been transformative, yet has largely proceeded without engaging with the theoretical foundations developed by the TCS community. At the same time, the deployment of these models has created pressing challenges around trustworthiness: How do we attribute model behavior to training data? How do we verify whether content is AI-generated? How do we align AI systems with the diverse preferences of society? How do we ensure safety and interpretability?

We believe theory has a key role to play in addressing these questions. Indeed, a growing body of recent work demonstrates that ideas from cryptography, social choice theory, learning theory, and statistics yield concrete, practical tools for modern AI systems. This workshop will showcase these successes and make the case that trustworthy AI is a natural and fertile area for harnessing theoretical perspective for practical impact.

Organizers

Schedule