In social dilemmas individuals are faced with the choice of paying a cost to contribute to a common good (cooperate) or simply abstaining from doing so and free ride on the contributions of other people (cheat). Countless models, simulations and experiments with humans, have shown that nothing good is going to come out from leaving individuals to regulate themselves. If we want to ensure socially desirable behaviour, we need to put mechanisms in place that discourage cheating and promote cooperation. In this talk, I will explore the emergence of cooperation in social dilemmas with the option of opting out, where individuals play a social dilemma in pairs but can also decide whether to continue playing with their current partner or break ties and be randomly assigned to someone else instead. We will observe that this simple mechanism is enough to promote cooperative behaviour even in populations of self-interested individuals. To this end, I will show that utility-maximising Q-learning agents are able to discover and adopt strategies for partner selection that ostracise selfish agents, completely by themselves. I will then move to experiments conducted with humans on these games, analysing the conditions under which cooperation prevails, and show how a simple Reinforcement Learning algorithm can faithfully simulate the way humans behave. What I will to present is joint work with Chin-wing Leung, Apurva Shah, Eugene Malthouse, Charlie Pilgrim and Thomas Hills

 

Short-Bio:
Paolo Turrini is an Associate Professor in the Department of Computer Science of the University of Warwick. His research is in Artificial intelligence for Social Good, designing artificial agents that are driven towards socially desirable objectives. He is particularly interested in strategic interaction in time and space, going beyond classical evolutionary game theory models, paying particular attention to the interaction that happens when self-interested agents are connected in networks. In his work, he uses theoretical analysis as well as computer-aided simulation, with an eye on real-world behaviour. He consistently publishes in top AI venues, such as IJCAI, AAAI and AAMAS. He is a member of the Board of Directors of IFAAMAS, the International Foundation for Autonomous Agents and Multiagent Systems.
Paolo Turrini Webpage: https://www.dcs.warwick.ac.uk/~pturrini/