Proceedings of the Annual Meeting of the Cognitive Science Society

Rational updating of beliefs requires the processing of new information, which, in turn, requires computational resources, which are limited. Limited computational resources may give rise to an information-computation gap: the quality of an agent's beliefs may not reach the information-theoretic limit due to limited computational resources. Here, we quantify the computational resources required for belief updating in an objective, ex-ante, and task-independent way and document, using a laboratory experiment, that indeed the quality of human beliefs declines in proportion to the computational demands of updating. Further, we validated this metric by showing that it makes unique predictions regarding belief quality. These findings advance a prescriptive theory of beliefs updating that accounts for systematic deviations from rational behaviour. Our results highlight the importance of incorporating computational constraints when evaluating decision-making performance.