Abstract
<title>Abstract</title> <p>Although action outcomes are central to learning, little is known about whether predictive outcomes become part of the task representations used for cognitive control. The present study addressed this question by incorporating the Differential Outcomes Procedure into an alternating-runs task-switching paradigm. Participants performed a shape-color task under four outcome conditions: Differential outcomes, in which each task-relevant cue-choice relation predicted a unique outcome; non-differential outcomes with two possible random outcomes within each task context; non-differential outcomes with four possible random outcomes across task contexts; and a no-outcome control condition. This design enabled us to distinguish the effects of outcome specificity from those of the mere presence of positive outcomes, the number of possible outcomes, and random outcome uncertainty. Differential outcomes did not reliably modulate switch costs, suggesting that predictive outcomes primarily affect processes other than continuously switching between task sets. In contrast, differential outcomes produced larger mixing costs than both non-differential outcome conditions, suggesting that differential outcomes affect sustained task-maintenance processes. The two non-differential outcome conditions did not differ from each other or from the control condition. These findings suggest that predictive outcomes are incorporated into event files, thereby increasing the representational demands involved in maintaining multiple task sets. More generally, they demonstrate that action outcomes are not merely the consequences of behavior or sources of reinforcement; when they are predictive, they can become part of the representational structure on which cognitive control operates.</p>