Postersession 2
Poster #: 50
Topic: Error signals
Thursday, Sep 10, 2015
1st floor

Prediction errors – Mismatch negativity (MMN) reveals how higher-level models govern error detection at lower levels

Daniel Mullens, István Winkler, Karlye Damaso, Andrew Heathcote, Lisa Whitson, Alexander Provost, & Juanita Todd1

1Psychology, University of Newcastle, Newcastle, Australia

First-impressions are known to impact decision-making and are attributed to assumptions we make that have prolonged effects on reasoning. We have observed similar phenomena (termed “primacy bias”) affecting the amplitude of the mismatch-negativity (MMN) component in the auditory evoked potential. Here we test whether this phenomenon can be explained by the formation of a confident-weighted first impression bias. In two studies participants were asked to focus attention on a silent movie and ignore auditory stimuli while we recorded evoked responses to sound sequences. Sequences consisted of concatenated segments alternating the sequential probabilities (roles) of the same two tones (frequent/predictable vs. infrequent/unpredictable). Primacy bias refers to the finding that the event-related brain potentials expected on the basis of transition statistics are only found for the segments in which the roles are identical to how they were first encountered, but not for segments with the reversed roles. Study 1 shows that this bias is initially prevented if there is no 1:1 mapping between sound attributes and roles, but it returns once the auditory system determines which properties provide the highest predictive value. Study 2 shows that confidence in this bias drops if assumptions about the temporal stability of the pattern are violated. Together, these results provide compelling evidence that, the context (here the large-scale structure of the sequences) affects these sensory first impressions. The results are compatible with hierarchical predictive coding theories, demonstrating how higher-level models govern error detection at lower levels.