Stubborn Predictions in Primary Visual Cortex

摘要

Perceivers can use past experiences to make sense of ambiguoussensory signals. However, this may be inappropriate whenthe world changes and past experiences no longer predict whatthe future holds. Optimal learning models propose thatobservers decide whether to stick with or update their predictionsby tracking the uncertainty or “precision” of their expectations.However, contrasting theories of prediction haveargued that we are prone to misestimate uncertainty-leadingto stubborn predictions that are difficult to dislodge. To comparethese possibilities, we had participants learn novel perceptualpredictions before using fMRI to record visual brain activitywhen predictive contingencies were disrupted-meaning thatpreviously “expected” events became objectively improbable.Multivariate pattern analyses revealed that expected events continuedto be decoded with greater fidelity from primary visualcortex, despite marked changes in the statistical structure of theenvironment, which rendered these expectations no longervalid. These results suggest that our perceptual systems doindeed form stubborn predictions even from short periods oflearning-and more generally suggest that top-down expectationshave the potential to help or hinder perceptual inferencein bounded minds like ours.