Evidence for a hierarchy of predictions and prediction errors in human cortex

Abstract

According to hierarchical predictive coding models, the cortex constantly generates predictions of incoming stimuli at multiple levels of processing. Responses to auditory mismatches and omissions are interpreted as reflecting the prediction error when these predictions are violated. An alternative interpretation, however, is that neurons passively adapt to repeated stimuli. We separated these alternative interpretations by designing a hierarchical auditory novelty paradigm and recording human EEG and magnetoencephalographic (MEG) responses to mismatching or omitted stimuli. In the crucial condition, participants listened to frequent series of four identical tones followed by a fifth different tone, which generates a mismatch response. Because this response itself is frequent and expected, the hierarchical predictive coding hypothesis suggests that it should be cancelled out by a higher-order prediction. Three consequences ensue. First, the mismatch response should be larger when it is unexpected than when it is expected. Second, a perfectly monotonic sequence of five identical tones should now elicit a higher-order novelty response. Third, omitting the fifth tone should reveal the brain's hierarchical predictions. The rationale here is that, when a deviant tone is expected, its omission represents a violation of two expectations: a local prediction of a tone plus a hierarchically higher expectation of its deviancy. Thus, such an omission should induce a greater prediction error than when a standard tone is expected. Simultaneous EEE- magnetoencephalographic recordings verify those predictions and thus strongly support the predictive coding hypothesis. Higher-order predictions appear to be generated in multiple areas of frontal and associative cortices.

Fig. 1.

Experimental design. Three auditory stimuli could be presented: local standards (a series of five identical tones, denoted xxxxx), local deviants (four identical tones followed by a different tone; denoted xxxxY), and omissions (four identical tones; denoted xxxx).These stimuli were presented in three types of blocks in which one series was presented with a high frequency (initially 100%, then 75%) and the other series were rare. This design thus separated the local deviancy of the fifth sound from the global deviance of the entire sequence and also allowed us to probe whether the omission effect differed when a standard or a deviant tone was expected.

Fig. 2.

Sensor-level topography and time course of the brain responses to distinct forms of novelty. In each panel, the three topographies show the spatial distribution, on a top view of the scalp, of EEG signals (Left), MEG gradiometers (norm; Middle), and MEG magnetometers (Right) at the time indicated by the vertical dotted line in the graphs. Graphs show the time course of these signals as recorded from an individual sensor (marked by a dot on the corresponding topographical map). (A and B) Effects of local mismatch: Bilateral auditory areas show a rapid response to the fifth deviant tone, whether it is rare (xxxxx blocks, in A) or frequent (xxxxY blocks, in B). (C) Effect of global deviance: A temporally and spatially extended response, corresponding to the P3b in ERPs, is seen in response to rare sequences. (D–F) Responses to omission of the fifth tone. The brain responds to omission by emitting a sharp response whose amplitude is smaller when a standard is expected (D) than when a deviant is expected (E), resulting in a significant difference (F). omiXX, rare omissions in the xxxxx block; omi XY, rare omissions in the xxxxy block; omi exp, expected omissions in the control omission block.

Fig. 3.

Source modeling of the effects. The reconstructed signals from the right auditory cortex (A) and right precentral cortex (B) are shown. Upper panels show the signals in each of the seven experimental conditions, and the lower panels show subtractions of each rare sequence (identified by the same color as in the upper plot) minus the frequent sequence of the same block, and of omissions in each block type (same color as in the upper plot) minus the expected omission from the control block. (C and D) The z-score–corrected source reconstruction of the local effect (C) and effect of deviance from global rule (D) on inflated cortex. Local effect (C) is the average effect of deviance from local regularity over both types of blocks at time = 110 ms. Global effect (D) is the average of contrasts between rare and frequent trials, at constant stimulus (rare violation − frequent sequences and rare omissions − expected omissions), averaged over the late period of the trials (300–600 ms). The source in auditory cortex shows a sharp and rapid response to local deviance, followed by a late and sustained response to global deviance. The source in precental cortex shows only the late sustained response. Both effects are present on omission trials (red/pink curves), with greater initial response to omissions in xxxxY blocks (thin red curve) than in xxxxx blocks (thick pink curve).