Musicians are often said to possess "good timing". Technically, this time-based behavior can be defined as appropriate action based on accurate prediction. Or, phrased differently: musicians know when to play because they estimate when events will occur in the music. This ubiquitous process is of importance in such diverse domains as sports, dancing, or maneuvering urban traffic. A key prerequisite for adequate timing behavior is knowledge about the temporal structure of the world – when do events happen in time?
In music this ranges from complete event predictability in the case of high regularity (e.g. a metronome), to situations where prediction fails entirely and notes arrive unexpectedly (e.g. in free improvisation). In between these extremes, temporal information can be formalized mathematically by a probability density function which represents the event probability distributed across time. The brain is commonly argued to estimate a related, but more complex, mathematical model of temporal probability, the so-called hazard rate, to predict upcoming events. In this project, we tested the prominent hazard rate hypothesis of event anticipation and investigated alternative models of how the brain estimates its environment's temporal-probabilistic structure to guide its actions. Crucially, we found that the brain does not model the hazard rate but computes a much simpler and more stable model: the reciprocal probability density function of events in time (Grabenhorst, Michalareas, et al., Nature Communications, 2019). In a series of experiments, we observed highly similar results in audition, vision, and touch indicating that the estimation of event probability density is a canonical, fundamental operation, independent of sensory input modality. Guided by these results, we perform further experiments using magnetoencephalography to investigate the neural dynamics of the processing of temporal information in the brain.