Interaction of Conflict-Minimizing and Goal-Seeking Motor Imperatives Under Active Inference in Autism

According to a classical control-theoretic perspective on self-movement, our brains pilot our bodies to achieve goals, sometimes using open-loop predictive models and sometimes with closed-loop feedback. From this perspective, synchronization of self-motion with a moving target is a simple control problem, where error between body and target is minimized by some combination of responsive and predictive control. However, self-movement can also result from reconciliation of multisensory prediction errors. When subjected to a visual/proprioceptive conflict, our bodies respond spontaneously and unconsciously to reduce or resolve it. These two drivers of movement are unified within the theoretical framework of active inference. To study this dynamic interaction of goal-oriented and conflict resolving motor imperatives, we designed a virtual reality task where participants synchronize their hand movement with an opening and closing target hand. We introduce two perturbations: a transient delay of the target hand, and a transient delay of their own virtual hand. We model responses to these perturbations using an active inference model. Preliminary results in non-autistic individuals indicate that delays in visual self-motion cause participants to delay their own motion, demonstrating the multisensory-conflict-reducing behavior. Certain sensorimotor differences reported in autism, including differences in the weighting of proprioceptive relative to visual input and weaker error correction during entrained finger tapping, could follow from reduced sensitivity to sensory/proprioceptive prediction errors in the context of active inference. Our model suggests that reduced sensitivity to sensory/proprioceptive prediction errors should make autistic individuals less responsive to both types of perturbation compared to non-autistic peers (after controlling for visual attention). This result would establish a link among several sensorimotor differences in autism that can be explained in terms of active inference, help study interindividual differences in active inference processes and would help to identify targets for future interventions.