Oscillatory Entrainment Guides Temporal Prediction in Non-Deterministic Continuous Environments, Independent of Bayesian Sequential Learning

Abstract

Predicting future event timing is essential for human interaction with our environment, and has been established in stimulus streams containing deterministic regularities. However, real-world streams often lack such regularities, leaving it unclear whether and how temporal predictions operate. Here we show that oscillatory entrainment, a mechanism identified for deterministic streams, operates in non-deterministic streams containing substantial temporal jitter. Model simulations revealed divergent dynamics and predicted timepoints between the oscillatory mechanism, which tracked local input regularities, and Bayesian learning, an alternative computational mechanism, which encoded global statistics. Across two speeded detection experiments using mechanism-dissociating streams, human performance was facilitated for targets appearing at oscillator-predicted times in streams with intermediate jitter levels, relative to maximal-jitter streams. Importantly, this effect was decoupled from Bayesian-based prediction and non-specific preparation, and could flexibly adapt to stream rate variations. These findings reveal the parallel mechanisms employed by humans when faced with continuous temporal uncertainty.