SARL: A Computational Reinforcement Learning Model with Selective Attention


A model relating eye-movements and decision making is proposed focused on the iterated prisoner’s dilemma game. Its main aim is to model previous experiments with eye-tracking recordings which show that participants attend to only a small part of the game payoff information. The model presented generates eye-movements based on two main mechanisms. The first takes into account the importance of the information attended with respect to the decision making process and while the second takes into account the variability of the information attended. The model is a discrete dynamical system which integrates learned selective attention with move choice. The model is found to reproduce fairly well the sensitivity to the payoff structure of the game and the attendance to payoffs found in experiments with human subjects. These results seem to be a promising first step in explaining the impact of partial and selective information acquisition in the prisoner’s dilemma.

Back to Thursday Posters