In the brain, actions are assumed to be paired with predictions of the sensory input that they produce. For example, when actions are paired with specific sounds, deviations from these predictions elicit an additional neural response that presumably reflects prediction error. However, in experimental designs applied to date predictions remain relatively fixed, leaving it unclear whether the brain can flexibly adjust predictions on a case-by-case basis. Furthermore, these designs often fail to account for neural adaptation effects, casting doubt whether additional neural activity can be interpreted as prediction error.

This study addressed these issues using a paradigm where participants performed random sequences of left and right button presses. In the 80/20 condition, one hand was coupled to a high likelihood of producing a sound (80%) and the other hand to a low likelihood (20%). In the 50/50 condition, both hands had a 50% likelihood of producing a sound. This design ensured identical adaptation effects across conditions while manipulating trial-by-trial predictions coupled to the button press.

ERP results confirm different predictions between button presses in two main ways. First, unexpected sounds (a sound when the 20%-sound button was pressed) resulted in a mismatch negativity and a P3 response, similar to sounds in the 50/50-condition. Second, unexpected omissions (an omission when the 80%-sound button was pressed) elicited an omission response when compared to expected omissions (an omission when the 20%-sound button was pressed). These findings convincingly demonstrate the brain’s ability to flexibly adjust top-down predictions in anticipation of the upcoming action.