Fig. 5.
Parameter distributions. (A) Participants learned the structure of the environment. Distribution of participants’ priors over environment complexity, α. Each individual’s parameter is shown relative to a baseline threshold, 0.8. This threshold is the lowest value that produced multicluster inference in simulation. Most participants (76%) fall above this threshold, indicating that a majority learned the environment’s multicluster structure. (B) Environment complexity parameters were positively related to reaction time sensitivity to transition frequency. An individual must infer multiple planet types to be sensitive to the transition structure between them. In terms of the model, this would correspond to having a sufficiently high environment complexity parameter. Validating this parameter, it was positively correlated with the individual’s modulation of reaction time following a rare transition to a different planet type. (C) Participants adapted their discounting computations to their uncertainty over environment structure. Distribution of the participant’s uncertainty adaptation parameter, γcoef. Each individual’s parameter is shown relative to a baseline of 0. A majority were above this threshold (93%), indicating that most participants dynamically adjusted their discounting, increasing it when they experienced greater internal uncertainty. (D) Uncertainty adaptation parameters were positively related to overharvesting sensitivity to transition frequency. If individuals increase their discounting to their internal uncertainty over environment structure, then they should discount more heavily following rare transitions and consequently, stay longer with the current option. Consistent with this, we found that the extent to which individuals increased their overharvesting following a rare transition was related to their uncertainty adaptation parameter.