Transfer entropy predicts pupillary response and cognitive effort during a tracking task

1.

Cognitive cost plays a crucial role in determining action choices and task engagement. We have proposed that cognitive cost relates to the amount of Bayesian inference required, quantified using relative entropy. In continuous tasks, such as visuomotor tracking, this demand can be estimated using transfer entropy (TE), which accounts for the information transferred from input to output variables. This study aimed to test experimentally the relationship between TE and markers of cognitive cost.

We designed a continuous tracking task, manipulating target speed, predictability, and response lag. The effects of task variables on performance, subjective effort and pupil-linked arousal were assessed using Bayesian mixed models. We disentangled the effect of TE from that of other correlated variables (predictability, acceleration, spatial error), by using model comparison and mediation models. Results showed that TE was mostly affected by target predictability and impacted significantly on subjective effort and pupil response. While the variations of pupillary response induced by task conditions were accounted for by TE only, subjective effort was only partially predicted by TE, and depended also on target predictability.

These findings suggest that the cognitive demand of motor control tasks can be quantified using TE, which relates to both subjective effort and pupillary response. However, the multifaceted nature of effort in motor tasks and the limitations of subjective measures highlight the need for more nuanced tools to isolate mental and physical effort.