Predictive eye movements are adjusted in a Bayes-optimal fashion in response to unexpectedly changing environmental probabilities

Abstract

This study examined the application of active inference to dynamic visuomotor control. Active inference proposes that actions are dynamically adjusted according to uncertainty about sensory information, prior expectations, or the environment and serve to minimise future prediction errors.We investigated whether predictive gaze behaviours are indeed adjusted in this Bayes-optimal fashion during a virtual racquetball task. In this task, participants intercepted bouncing balls with varying levels of elasticity, under conditions of high and low environmental volatility. Participants' gaze patterns differed between stable and volatile conditions in a manner consistent with generative models of Bayes-optimal behaviour. Partially observable Markov models also revealed an increased rate of associative learning in response to unpredictable shifts in environmental probabilities, although there was no overall effect of volatility on this parameter. Findings extend active inference frameworks into complex and unconstrained visuomotor tasks and present important implications for a neurocomputational understanding of the visual guidance of action.

Task-evoked pupillary responses track precision-weighted prediction errors and learning rate during interceptive visuomotor actions

“…It is important to note, however, that the better fit of the HGF does not in itself mean eye movements are the result of a Bayesian inference processes in the brain, only that this model better accounted for the data than the alternative learning model. Nonetheless, this result is consistent with a growing body of evidence from this task 35 , and other simpler eye movement tasks that indicate eye movements may follow Bayesian principles 58 – 60 .…”

“…Previous work has demonstrated that predictive eye movements can be used to model active inference during interception of a bouncing ball 35 . When intercepting a ball in this task, individuals have been shown to direct a single fixation to a location a few degrees above the bounce point of the oncoming ball 36 , 37 (see Fig.…”

“…Crucially, the spatial position of this fixation (the gaze pitch angle) is sensitive to beliefs about likely ball trajectories, with fixations directed to a higher location when higher bounces are expected 36 . As the fixation occurs before the bounce is observed, the fixation location is driven by an agent's prior expectations about ball elasticity and therefore provides an indicator of the evolution of beliefs over time 35 .…”

“…Data values that were > 3.29 SD away from the mean were classed as outliers ( p < 0.001), and participants with > 15% of data identified as missing and/or outliers were excluded (in line with 33 ). As in Arthur and Harris 35 the pitch angle variable was then converted to a discrete variable for modelling purposes; when the gaze angle shifted to a lower spatial location than on the previous trial (> 1 SD change) this was taken as a shift towards higher expectation of p(normal) and vice versa.…”

A Bayesian computational model to investigate expert anticipation of a seemingly unpredictable ball bounce

“…Active inference extends predictive coding models of perception to the use of action to minimize future prediction errors (Friston, 2005 ; Parr & Friston, 2018 , 2019 ). In addition to constantly revising their generative model, a Bayesian agent can minimize prediction errors through movements (e.g., see the use of vision to minimize surprisal; Arthur & Harris, 2021 ; Parr et al, 2021 ) or can actively change the world into the predicted state (Adams et al, 2013 ; Sarpeshkar et al, 2017 ). Harris et al ( 2021 ) have recently suggested that active inference can enhance our understanding of skilled anticipation by providing a principled account of how actions are used to optimize predictions, as well as accounting for decision-making via Bayesian inference.…”