i) using game theory to design and control artificial decentralized systems,

ii) controlling systems that contain interacting humans with conflicting goals (i.e., control of human game players).

I will start with a brief historical survey of topic (i).

I will then discuss Quantal Response Equilibria (QRE) models of how game players behave, and how to control systems assumed to obey that model. In particular I will discuss a project with people at Oxford using the QRE to design multiple interacting surveillance robots (topic (i)), and another project with people at Cambridge using the QRE to model human behavior behind the financial crisis (topic (ii)).

Finally, I will briefly mention "level-K" and "best-of-M/M" models of player behavior. These seem to more accurately model real human beings than the QRE, and at the same time provide many computational advantages over the QRE for design and control of decentralized systems. I will discuss how they're being used in a current project on next generation aircraft collision avoidance (topic (ii)).

Host: Misha Chertkov