Order-dependent effects of sensorimotor context on decision policy in probabilistic learning
Author(s)
Huang, Elliot
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Reinforcement learning is the primary framework used to understand how humans learn and make decisions. Conventional experimental paradigms tend to minimize physical execution demands, ignoring that real-world decisions are made in significantly varied sensorimotor contexts. Prior work suggests that variables modulated by sensorimotor context influence reinforcement learning; however, an integrative understanding of how sensorimotor context alters decision processes remains lacking. Ninety participants completed a two-choice probabilistic reversal learning task under two within-subject conditions: active (full-body locomotion between options) and stationary (arm movements only), with condition order counterbalanced across participants. Stay rate, accuracy, logistic regression history kernels, reversal learning curves, and computational models were compared across conditions and order groups. Sensorimotor context affected stay rate only among participants who completed the active condition first; these participants showed significantly elevated win-stay rates and choice predictability in the subsequent stationary condition. The effect was absent in stationary-first participants and was not present in first-condition data from either group, ruling out a direct effect of sensorimotor context. Computational modeling localized the effect to decision policy: win-stay, lose-stay, and inverse temperature parameters were elevated in the stationary condition for active-first participants, while the learning rate parameter and reversal performance were unchanged across all groups and conditions. Individual differences in stay tendency were stable across conditions and order groups, with the context transition producing a uniform additive shift across this distribution. These findings demonstrate that prior sensorimotor context alters subsequent decision policy through a contrast mechanism, without affecting value learning, establishing sensorimotor context as a functionally relevant variable in learning and decision-making.
Sponsor
Date
2026-05
Extent
Resource Type
Text
Resource Subtype
Undergraduate Thesis