Submission Link Coming Soon
This workshop addresses the critical challenge of verifying autonomous systems that incorporate artificial intelligence, with particular attention to recent advances in foundation models, large language models (LLMs), and generative AI. Our objectives are to:
Assess the current state: Document verified successes where AI-enabled autonomy has been rigorously validated, and candidly examine failures where verification approaches proved insufficient or where AI introduced unforeseen risks.
Identify open problems: Characterize gaps in current verification methodologies, including challenges unique to neural network-based perception, LLM-driven planning, and hybrid classical-learning architectures.
Develop best practices: Establish preliminary guidelines for practitioners deploying AI in safety-critical autonomous systems, drawing from formal methods, runtime monitoring, and empirical validation approaches.
Map the solution space: Determine when traditional non-AI autonomy remains advantageous from a verification perspective, and clarify how AI verification techniques (adversarial robustness, uncertainty quantification, interpretability) integrate with systems-level safety assurance.
Reframe uncertainty: Explore whether AI constitutes a fundamentally new category of uncertainty requiring novel theoretical frameworks, or whether existing stochastic and epistemic uncertainty models can be extended.
MIT
University of Pennsylvania
Edge Case Research, Inc.
Washington University in St. Louis