Proceedings of the Annual Meeting of the Cognitive Science Society

In this paper, we highlight the shortfall of conventionally de-scribed heuristics in multi-attribute decision theory, and pro-pose recasting these heuristics within a novel probabilisticframework. This framework is based on defining a psycho-logical feature space, with rule-based heuristics represented asprototypical representations within this space. We provide var-ious examples of meaningful heuristics that can be constructedunder this representation, including recasting probabilistic ver-sions of popular heuristics such as take-the-best. Next, wepropose an evaluation framework to measure the effectivenessof a consideration set of heuristics. This framework measureswhether the set of heuristics are sufficient to describe, predictand infer strategy selection and learning behavior. We proposethat this is a step towards a robust framework within whichmodels of strategy selection and learning should be evaluated.The framework aspires to develop a consideration set of heuris-tics that can be represented as a mathematically well-posed in-ference problem. We show that the heuristics redefined underour probabilistic framework generally perform better than con-ventional heuristics under this evaluation. We conclude with adiscussion on the possible applications of this framework.