Inferring When to Act from Temporal Regularities

Adaptive behaviour often requires deciding when to act in the absence of an explicit sensory cue. While temporal expectations are known to optimize behaviour when anticipated events trigger responses, it remains unclear how learned temporal regularities are transformed into internally generated decisions. Here, we developed the Temporal Inference Task, a novel virtual reality paradigm designed to isolate this transformation. Participants repeatedly observed an identical visual sequence in which a virtual object approached their hand. On most trials, a brief tactile No-Go signal instructed them to withhold their response, whereas on the remaining trials its omission required them to respond. Because Go trials were never accompanied by an explicit cue, participants had to infer when the expected No-Go signal could no longer occur before initiating a response. Across blocks, the timing of the No-Go signal was systematically varied, allowing participants to learn distinct temporal regularities while Go trials remained physically identical. Participants systematically shifted their response timing according to the learned temporal regularities. This behavioural adaptation was accompanied by corresponding shifts in parietal alpha- and beta-band desynchronization, with steeper pre-response desynchronization consistently preceding faster responses. Together, these findings identify a candidate neural mechanism through which learned temporal regularities are transformed into internally generated decisions about when to act.