Guennouni, Ismail; (2023) Strategic Inference in Social Interaction: An Experimental and Computational Account. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text thesis_final.pdf - Accepted Version Download (3MB) | Preview

Abstract

At the core of social interaction is determining the goals, intentions, and decision-making process of the interaction partner. For example, if we deem a person trustworthy, we might decide to take the risk of investing in the relationship hoping for a cooperative outcome. If we misplace our trust, this can come at a high cost. Likewise, in competitive settings, inferring which actions opponents are likely to take allows us to outsmart them and win the competition. Irrespective of the setting, being able to predict the actions of the confederate (be it a potential partner or opponent) is key to adaptive behavior. There are multiple ways in which we can build predictions of other people’s behavior. I propose that people can achieve this from very sparse so- cial interaction data by making rich inferences from observing others’ actions. These inferences are then used to build and update structured representations of others’ minds, including their beliefs, preferences and evaluation function. Each chapter in this thesis is an exploration of such representations and aims to understand these cognitive abilities through recreating them using infor- mative computational models. In Chapter 2, we show in two experiments that people rely on explicit models of the opponent’s strategy in a previous competitive game to exploit the opponent and transfer that learning to novel but structurally similar games. In Chapter 3, we show that we can harness the power of hidden Markov models to characterise participants’ actions in a social dilemma game, and use these models to build computer agents whose strategy is a good proxy to human behaviour. In Chapters 4 and 5, we ex- plore the behavioral impact of intervening in the same social dilemma game, either by facilitating inference about the opponent or through a cognitive in- tervention aiming to repair an accidental breakdown of cooperative behavior. Chapter 6 uses POMDPs as a computational framework to build agents that leverage cognitive representations of the opponent’s decision process, explore the action space and find optimal solutions to the social interaction problem.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item