Defining and Mitigating Collusion in Multi-Agent Systems
Paper accepted to MASEC Workshop at NeurIPS 2023
Collusion between learning agents is increasingly becoming a topic of concern with the advent of more powerful, complex multi-agent systems. In contrast to existing work in narrow settings, we present a general formalisation of collusion between learning agents in partially-observable stochastic games. We discuss methods for intervening on a game to mitigate collusion and provide theoretical as well as empirical results demonstrating the effectiveness of three such interventions.
Programme
LASR
LASR Labs is a technical AI safety research programme focused on reducing the risk of loss of control to advanced AI.
Participants work in teams of three to four, supervised by an experienced AI safety researcher, to write an academic-style paper and accompanying blog post. Participation is full-time and in-person from the London Initiative for Safe AI alongside other AI safety researchers. The programme is designed to “learn by doing”; taking a research project from proposal all the way to publication.