Proceedings of the Annual Meeting of the Cognitive Science Society

Real-word decision making often involves selecting a singlechoice from an arbitrarily large set of possible options. Giventhat it is typically not feasible to evaluate every possible op-tion in real world decision making, how are human decisionmakers able to efficiently make good decisions? We proposeand evaluate a two-step architecture according to which peoplefirst sample a small subset of options weighted by their previ-ously learned value, and then evaluate those options within thecurrent decision-making context. We demonstrate that a ver-sion of this model captures human decision making in prob-lems where time and resource constraints prevent the evalua-tion of every option, and connect this research to the growingliterature on the representation of non-actual possibilities.