Fig. 4.

Modeling results. (A) The adaptive discounting model predicts overharvesting. Averaging across all planets, only the adaptive discounting model predicts overharvesting, while the temporal-difference learning model predicts MVT optimal behavior, and the MVT learning model predicts underharvesting. This demonstrates that overharvesting, a seemingly suboptimal behavior, can emerge from principled statistical inference and adaptation. (B) Model predictions diverge most on rich planets. Similar to participants, the greatest differences in behavior between the models occurred on rich planets. (C) The adaptive discounting model provides the best account for participant choices. The adaptive discounting model had the lowest mean cross-validation score, indicating that it provided the best account of participant choice at the group level.