The importance of precision to updating models of the sensory environment

Abstract

The existence and updating of “sensory beliefs” or internal models can be studied using auditory evoked potentials (AEPs) when there is some form of predictable pattern in sound. Internal models are proposed to enable predictions about the most likely next sound-activation-state leading to small AEPs to standard sounds matching model predictions, and larger AEPs to sound that deviate. Internal models are precision-weighted with the standard-deviant difference being largest when precision is high (variability is low). Here we expose how order-effects determine whether a change in variability impacts model-precision estimates. Thirty participants heard 3000 t (30 ms standard p = 0.90 and 60 ms deviant p = 0.10) that either moved from a more precise stimulus onset asymmetry (n = 15, first 1000 tones 500 ms ± 10 ms) to a more variable one (n = 15, subsequent 2000 tones 500 ms ± 200 ms) or from variable (first 1000 t) to more precise (subsequent 2000 t). AEPs were equivalent between groups for the first 1000 tones but differed dramatically in the face of timing changes. Where timing precision decreased, the standard-deviant difference was impervious to the change but where precision increased, the standard-deviant difference increased dramatically after the timing change signalling a transient increase in model precision that subsided over the final 1000 tones. The results support contemporary models proposing that updates to an active internal model will be a function of the quality of the evidence upon which it has been built and the information value of subsequent errors in improving the predictive success of the active model.