AI seminar: Dealing with risk aversion in multiagent systems: An insurance policy based approach

Speaker: Greg Hines

In much of the literature in multiagent systems, researchers assume (either implicitly or explicitly) that the agents are risk-neutral. However, this assumption is not always warranted in that agents can be risk-averse. Understanding the implications of risk-aversion in multiagent systems and ways of mitigating it are interesting open problems which may have ramifications on how we design mechanisms and algorithms. In this talk I will present an insurance mechanism that can be applied to multiagent settings where the agents are self-interested and risk-averse. I will discuss experimental results which illustrate the benefit of the insurance mechanism, as well as describe what types of multiagent interactions most benefit from it.