Multi-Agent Common Sense - IPAM at UCLA

Explore a 33-minute conference talk on multi-agent common sense presented by Max Kleiman-Weiner at IPAM's Mathematics of Collective Intelligence Workshop. Delve into the unique power of human collective intelligence, examining how collaboration and fair benefit-sharing set us apart from other species and current AI systems. Discover a mathematical framework combining hierarchical Bayesian models of learning with game-theoretic models of social interaction, providing insights into distinctly human aspects of multi-agent common sense. Investigate topics such as inferring and forming joint intentions, reasoning about cooperation with common knowledge, and learning structured cultural knowledge from sparse examples. Gain a deeper understanding of the computational basis for human collaboration through behavioral experiments and theoretical models.

Intro
Cooperation is Ubiquitous
Human Cooperation is Distinct
Cooperation is Challenging
Reverse-Engineering Collective Intelligence
In The Moment: Naturalistic Games
Abstract Reciprocity via Theory of Mind
Overcooked: Coordinating Joint Intentions
Bayesian Delegation of Joint Plans
Towards an Abstract Theory of Reciprocity
Multi-agent Common Sense