An examination of active inference in autistic adults using immersive virtual reality.

Abstract

The integration of prior expectations, sensory information, and environmental volatility is proposed to be atypical in Autism Spectrum Disorder, yet few studies have tested these predictive processes in active movement tasks. We used an immersive virtual-reality racquetball paradigm to explore how visual sampling behaviours and movement kinematics are adjusted in relation to unexpected, uncertain, and volatile changes in environmental statistics. We found that prior expectations concerning ball ‘bounciness’ affected sensorimotor control in both autistic and neurotypical participants, with all individuals using prediction-driven gaze strategies to track the virtual ball. However, autistic participants showed substantial differences in visuomotor behaviour when environmental conditions were more volatile. Specifically, uncertainty-related performance difficulties in these conditions were accompanied by atypical movement kinematics and visual sampling behaviours. Results support proposals that autistic people overestimate the volatility of sensory environments, and suggest that context-sensitive differences in active inference could explain a range of movement-related difficulties in autism.

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

“…Behavioural data were primarily collected in the context of understanding how environmental uncertainty and volatility are processed in Autism Spectrum Disorder and a detailed description of all experimental procedures are thus provided in an accompanying manuscript 33 . In short, the interception task took the form of a VR racquetball (squash) game, in which participants had to return a bouncing ball back towards a target on the wall (videos available online: https://osf.io/qjbf2/).…”

“…In expected trials, ball elasticity was set at 65%, corresponding with the normal behaviour of a real tennis ball. For unexpected trials, elasticity was increased to 85%; an easily detectable change in 'bounciness' and post-bounce trajectory 6,33 .…”

“…Data extraction and cleaning procedures are described in an accompanying paper 33 . Here, participant's cyclopean gaze vector and head position (x,y,z) were recorded from the virtual environment and plotted with respect to 2D direction in space, to provide relative 'in-world' angular orientations (head-ball, gaze-head, and gaze-ball angles).…”

“…Bland and Schaefer 30 further distinguish between these latter constructs, in the sense that unexpected uncertainty is characterised by rare unforeseen changes in probabilistic relationships, while volatility typifies frequent variations that can, in effect, become expected. Although visually guided movements should theoretically account for such uncertainty and volatility statistics [33][34][35] , it remains unclear how gaze behaviours are adjusted during complex visuomotor skills. Specifically, do agents minimise prediction error in a progressive, Bayes-optimal manner over time?…”

“…Further, we sought to test whether such changes approximate Bayes-optimal behaviour, as predicted by active inference accounts of perception and action 19,20 . To do this, we studied a virtual racquetball task (see Fig 1), in which participants typically display strong prediction-driven gaze behaviours 10,11,33 . In line with active inference approaches, it was hypothesised that: i) performers will adjust predictive gaze behaviours between stable and volatile trials in a Bayes-optimal fashion, such that they will place less weight on top-down predictions under volatile conditions; ii) performers will show an adjusted learning rate, such that gaze behaviours will be more strongly influenced by recent context under volatile trial conditions; and iii) environmental shifts that are more unexpected will create a further increase in learning rate (i.e., for unexpected uncertainty compared to volatility; see ref 30 ).…”