Task relevance modulates the behavioural and neural effects of sensory predictions

The brain is thought to generate internal predictions to optimize behaviour. However, it is unclear whether predictions signalling is an automatic brain function or depends on task demands. Here, we manipulated the spatial/temporal predictability of visual targets, and the relevance of spatial/temporal information provided by auditory cues. We used magnetoencephalography (MEG) to measure participants’ brain activity during task performance. Task relevance modulated the influence of predictions on behaviour: spatial/temporal predictability improved spatial/temporal discrimination accuracy, but not vice versa. To explain these effects, we used behavioural responses to estimate subjective predictions under an ideal-observer model. Model-based time-series of predictions and prediction errors (PEs) were associated with dissociable neural responses: predictions correlated with cue-induced beta-band activity in auditory regions and alpha-band activity in visual regions, while stimulus-bound PEs correlated with gamma-band activity in posterior regions. Crucially, task relevance modulated these spectral correlates, suggesting that current goals influence PE and prediction signalling.Author summary: As natural environments change, animals need to continuously learn and update predictions about their current context to optimize behaviour. According to predictive coding, a general principle of brain function is the propagation of both neural predictions from hierarchically higher to lower brain regions and of the ensuing prediction-errors back up the cortical hierarchy. We show that the neural activity that signals internal predictions and prediction-errors depends on the current task or goals. We applied magnetoencephalography and computational modelling of behavioural data to a study in which human participants could generate spatial and temporal predictions about upcoming stimuli, while performing spatial or temporal tasks. We found that current context (task relevance) modulated the influence of predictions on behavioural and neural responses. At the level of behavioural responses, only the task-relevant predictions led to improvement in task performance. At the level of neural responses, we found that predictions and prediction-errors correlated with activity in different brain regions and in dissociable frequency bands—reflecting synchronized neural activity. Crucially, these specific neural signatures of prediction and prediction-error signalling were strongly modulated by their contextual relevance. Thus, our results show that current goals influence prediction and prediction-error signalling in the brain.