Origins of noise in both improving and degrading decision making
Unlike the predictions of normative choice theories, the real-world accuracy of human and animal decision-making depends on context. Observed impairments of choice performance by context can be explained by computations like divisive normalization. However, recent behavioral evidence suggests that context can sometimes enhance choice performance, a finding that challenges existing explanations. Here, we propose and test an extension of existing frameworks for context-dependent choice that incorporates a richer notion of noise. Model simulations show that noise arising before (early noise) and after (late noise) normalization predict opposing effects: early noise can cause contextual information to enhance choice performance while late noise can cause context to degrade choice performance. In experiments with human subjects, we confirm key elements of this model. Manipulating early and late noise - by inducing uncertainty in option values and controlling time pressure - produces dissociable positive and negative effects of context. Together, these findings suggest a new unifying mechanism for contextual modulation, highlighting the significant role of noise source in neural computation and behavior.