Fig. 1: Rule reversal task and behavioral performance in humans.

a Task schematic of the human rule reversal task. b Schematic of a learning block. In each block, 70% of trials presenting one tactile pattern were assigned to “Go,” while 70% of trials presenting the alternative tactile pattern were assigned to “NoGo.” Within each block, the stimulus-response association was reversed randomly between the 20th and 25th trials. c Average probability of selecting Rule#2 across all blocks, aligned to the reversal point. Shaded error bars represent SEM. A logistic regression model with three parameters—switch offset (s), slope (α), and lapse rate (ε)—was fitted to individual choice data. d Box plots of the fitted parameters (n = 32 participants): switch offset (s), slope (α), and lapse rate (ε). The red dashed lines represent the parameter fits for a “win-stay-lose-shift” (WSLS) agent that updates decisions based on feedback. Box plots indicate the median (middle line), 25th, and 75th percentile (box), and the maximum and minimum (whiskers), as well as the outlier (red cross). e Group-averaged proportion of correct strategies plotted across trials. The vertical dashed line represents the rule reversal point, while the horizontal dashed line indicates chance-level performance. Shaded error bars represent the standard error of the mean (SEM). The blue-shaded area denotes the Steady State (SS) period, where participants exploited a learned decision strategy, while the orange-shaded area indicates the Switch (SW) period, where participants adjusted their strategy. f The proportion of correct strategies in the Steady State was significantly higher than in the Switch period (p = 2.85 × 10⁻¹⁴; paired two-sample t-tests with two tails, n = 32 participants). ***p < 0.001. The definition of the median, minima, and maxima for the box plot is the same as (d). Source data are provided as a Source Data file.