Surprise impairs perception of surprising, and incidental, events

When environmental regularities change, new observations should be weighted more highly than old observations, to allow model updating in a fluctuating world. These changes influence learning rates, but it is less clear how they influence perception. A recent theory suggests that surprising observations trigger reactive processes which increase sensory gain in the subsequent few hundred ms. Such processes would generate more reliable sensory estimates of surprising events, thereby optimising accurate model updating. To test this account, we asked whether surprising events boost detection of the surprising events themselves, at 200-300 ms delay relative to 0-50 ms delay, and of other incidental events presented simultaneously. In four online experiments, participants were required to detect a feature of visual stimuli presented on the circumference of a circle, alongside presence of a central stimulus. Circumference stimulus location and orientation followed an environmental regularity which changed once without warning during the task. We modelled a surprise-induced weighting factor, based on Bayesian Changepoint Modelling, to ask whether the distribution change inflated hitrates both of features of the circumference event, which themselves generated the surprise, and of simultaneous central events. We instead, unexpectedly, found consistently lower hitrates on surprising trials for all event types. These findings suggest that surprising observations do not automatically increase sensory gain across channels, but instead that foraging for new information operates slowly, or that neural error processing enables model updating without enhancing perception.