Learning to Cooperate in Multi-Agent Systems

In the rapidly evolving landscape of artificial intelligence, envisioning a future where diverse ecosystems of AI systems intricately interact among themselves and with humans has become increasingly relevant. From collaborative industrial robots operating alongside workers on factory floors to care robots supporting health professionals, and personal AI assistants managing schedules in rudimentary ways, the integration of AI into our daily lives underscores the critical need for effective and reliable coordination among agents and between agents and humans. This endeavor poses a significant challenge, particularly in environments where the limitations of feedback mechanisms in practice can lead multi-agent learning to converge on equilibria that diverge from human expectations.

Addressing this challenge necessitates the development of AI systems capable of collaborating with other AI systems and humans across a variety of contexts, giving rise to the agent alignment problem. This problem centers on the creation of agents that not only act in harmony with user intentions but also possess the flexibility to adapt to evolving circumstances. The pursuit of AI alignment extends beyond individual interactions, extending to the incorporation of broader societal considerations. It requires to consider the perspectives of individuals as well as the norms, expectations, and values of the affected groups. In this presentation, I will provide an overview of the problem, highlighting our contributions to understanding and enhancing collaboration within social contexts, fostering human-AI coordination, and achieving scalable alignment.

Post Talk Link:  Click Here 

Passcode: Jx715*P^