Abstract
Predictive coding theories of perception posit that unexpected events trigger prediction errors and updating of predictive models. In this study, we tested the hypothesis that these processes modulate the availability of cognitive resources serving other tasks. We asked participants to perform a simple orientation discrimination judgement on a target Gabor patch that was preceded, on some trials, by a task-irrelevant disk flashed briefly either near or far from the target. At the start of each trial, a colored cue at fixation signaled the probability (80% or 20%) that the disk would appear. This information was not relevant to the orientation discrimination task. To probe the temporal dynamics of target stimulus processing, we forced participants to respond at seven different response delays. Assessing discriminability as a function of response time allowed us to generate the full speed-accuracy tradeoff (SAT) function, which reveals the accumulation of decision evidence over time. The main question of this study is whether violations of expectation regarding the task-irrelevant disk modulated processing of the target stimulus, compared to when the disk’s presence or absence was accurately predicted by the visual system. Two observers completed 20 experimental sessions each, generating a total of 33,600 trials. When expectations were accurate, discriminability rose monotonically with processing time. But violations of expectation caused a “wrinkle” in the SAT function: there was a significant boost in accuracy ~580ms post-stimulus, followed by a subsequent dip at ~670ms, after which accuracy continued its monotonic rise. We interpret this as an effect of a prediction error elicited by the unexpected appearance or omission of the irrelevant disk. The computations triggered by that event transiently marshal and then withdraw cognitive resources available for the current task. Our data thus demonstrate a time-locked psychophysical signature of prediction errors, even for stimuli that are uninformative and task-irrelevant.