This workshop, sponsored by AIM and the NSF, will be devoted to the mathematical foundations for AI agents in complex environments. Current artificial intelligence and machine learning paradigms largely focus on training models in silico —- optimizing them for isolated, static benchmarks without accounting for the dynamic ecosystems they will eventually inhabit. This approach is well documented to cause misalignment and poor real-world performance, as algorithms frequently fail to generalize to novel situations beyond their training data. Addressing these critical limitations requires a new set of mathematical foundations that accurately reflect the intricate, interactive conditions under which AI systems are actually deployed.

To bridge this gap, this workshop aims to explore the mathematical underpinnings of algorithmic decision-making, with a primary focus on dynamic, multi-agent environments. Because most modern applications of "agentic" AI are fundamentally multi-agent in nature, these systems require multiple models to coordinate, compete, or interact with one another — and with humans — to achieve their goals.

Discussions will center on robustness, equilibrium analysis, and control theory to establish the rigorous groundwork necessary for AI agents to operate safely and effectively in complex systems. Over the course of the workshop, we will synthesize emerging concepts at the intersection of AI, game theory, behavioral economics, and control theory, laying the foundation for the design of robust, generalizable AI agents.