Prediction Errors Bias Time Perception

Post by Cody Walters

What’s the science?

Dopaminergic neurons in the striatum (nuclei deep in the brain) have been implicated in both time perception and in prediction errors. Prediction errors are learning signals in the brain that convey information about the difference between expected and experienced outcomes. Despite their shared striatal circuitry, it is unclear whether time perception and prediction errors influence one another. Recently in Nature Neuroscience, Toren et al. provided evidence in support of the view that prediction errors distort time perception. 

How did they do it?

The authors used a two-alternative forced-choice task in which participants were presented with two white squares displayed sequentially on a screen. One was always presented for 500 milliseconds (the ‘reference’) while the other was displayed anywhere from 367 to 633 milliseconds. The participants’ task was to determine which square was displayed for a longer duration. Additionally, there was a number on the center of each screen: the first screen always displayed ‘0’ while the second screen ranged between -5 and +5 (in 0.5 increments). The difference between the first number (0) and the second number reflected monetary gains (if the difference was positive (e.g. 0 and +3) and losses if the difference was negative (e.g. 0 and -2) that the subject would receive at the end of the session. The second number therefore generated a prediction error. The prediction errors could be either positive (if the second number was greater than zero), neutral (if the second number was also zero), or negative (if the second number was less than zero). Participants participated in either a behavior-only group or a brain imaging group in which they performed the above task while undergoing functional magnetic resonance imaging (fMRI).

What did they find?

The authors found that participants made more time-discrimination errors on short-long trials (i.e., trials in which the first square was displayed for a shorter duration than the second square) with negative prediction errors (PE-). Similarly, they found that participants made more time-discrimination errors on long-short trials with positive prediction errors (PE+). These data suggest that negative prediction errors decrease the perceived stimulus duration while positive prediction errors increase the perceived stimulus duration.

Next, the authors designed a reinforcement learning model to predict trial-by-trial outcomes for individual subjects. The model had three key parameters: the objective time difference, the bias resulting from the prediction error, and the time-order error (a well-known phenomenon in which the ordering of sequential stimuli affects their perceived durations). The model accurately captured each subject’s time-discrimination performance. The authors then showed that fMRI activity in the ventral striatum, midbrain, and dorsal anterior cingulate cortex (structures that are known to be involved in prediction error encoding) correlated with trial-by-trial prediction errors.

The authors conducted a whole-brain analysis to identify regions whose activity correlated with the time perception bias. They found that there was increased activity in the putamen on positive prediction error  trials during incorrect discriminations (relative to correct discriminations). The opposite pattern was observed in the dorsal anterior cingulate cortex, where there was decreased activity on positive prediction error trials during incorrect discriminations. Lastly, the authors demonstrated that putaman activity was significantly correlated with subjects’ prediction error-induced time perception bias.

What’s the impact?

Much of learning and memory formation is driven by prediction error signaling, and time perception is critical in nearly every facet of daily life. The data presented in this study suggest that these two fundamental processes, traditionally considered to be independent of one another, are in fact deeply intertwined, with signed prediction errors bidirectionally biasing time perception. These results provide a novel insight into how the brain learns, forms memories, and perceives the passage of time.

Toern, et al. Prediction errors bidirectionally bias time perception. Nature Neuroscience (2020). Access the publication here.