Fig. 5: Computational modeling results.
From: Humans monitor learning progress in curiosity-driven exploration
a The bivariate models had better AIC scores both across and within groups (NEG = 188; NIG = 177), compared to random-choice and univariate baselines univariate models. Box boundaries represent the 1st and the 3rd quartiles, and the lines inside show median scores; whiskers represent the full sample range. The dotted red line shows the AIC of the random-choice model. b Fitted coefficients reproduce choice patterns across instruction and NAM groups. The panels show the average time allocation patterns obtained by simulating activity choices over 250 trials using N = 500 randomly sampled coefficients from the pool of all fitted bivariate models. c Models of two distinct activity-selection strategies. The top row shows the joint distributions of normalized bivariate-utility coefficients. Subsets of these distributions whose data is presented below are highlighted with solid colors. These subsets were formed by first grouping all fitted models into three segments along \(\hat{w}{{{{{{{\rm{PC}}}}}}}}\) and \(\hat{w}{{{{{{{\rm{LP}}}}}}}}\), and then selecting groups corresponding to PC-driven and LP-driven profiles. Sample sizes of each subset are shown their respective subpanels. The bottom row shows mean relative frequencies of selecting each activity in the corresponding subset of participants depicted immediately above. LP-driven participants sampled the unlearnable activity (A4) in relative moderation compared the PC-driven group. d LP-driven participants selected allocated time more efficiently for learning and had better learning outcomes. The top row shows fractions of participants in the two groups that reached an objective criterion of 13/15 trials on the hardest learnable activity (A3) at least once in the experiment. The middle row shows the relative preference for activity A4 over A3, defined as the difference between fractions of participants (that still have not mastered A3) who selected A4 minus the fraction selecting A3. The bottom row shows average SC scores in the two groups (shaded regions indicate the standard error). Source data are provided as a Source Data file.
