A State-Transition-Free Delayed-Feedback Task Elicits Heterogeneous Human Responses

TL;DR

This study explores how humans learn to connect actions with delayed outcomes, revealing significant individual differences in understanding and behavior.

Contribution

The study introduces a novel delayed-feedback task to investigate how humans cognitively link actions with outcomes that occur later.

Findings

Nine participants learned to choose an action leading to a delayed reward.

Participants showed varied explicit and implicit understanding of the task structure.

Individual differences included irrational biases and superstitious behaviors in action selection.

Abstract

Humans and nonhuman animals learn to perform actions by associating actions with outcomes. In everyday life, outcomes sometimes occur only after a delay, and at an unexpected moment. The ability to connect actions and delayed outcomes has received less attention than performance in tasks where rewards follow the most recent action. Here, following a previous study (Sato et al. 2023), we designed a learning task to investigate humans’ ability to link actions and outcomes which occurred after intervening choices. We prepared a total of six visual stimuli for use in three types of trials: A vs B, where choosing A immediately led to reward and choosing B was never rewarded, C vs D, where neither choice was immediately rewarded but choice of C led to reward in a later E vs F trial, and E vs F, where neither stimulus was associated with reward but a reward was given based on choice of C in the past. Results showed that nine individuals learned to choose C, thereby receiving a delayed reward. Among them, one participant subsequently correctly described the task structure in words, while the remaining eight did so with misunderstandings. We also observed large individual differences in participants’ action selection (e.g., an irrational bias for D, a possible superstitious bias for either E or F) and explicit/implicit understanding of the link between action and delayed outcome expressed in words. Our results offer new insights into the ability to cognitively link actions and outcomes following a time lag.

Figures