Fig. 5.

(a). This figure reports the effect of perturbing prior beliefs on the sorts of actions selected. This is shown for a range of alternative initial hidden states. This action selection distribution is constructed from 32 trials of four simulated variants (I-IV) for Scenario 1 (Initial outcomes: empty horizon visible, soft sound audible – see Fig. 3a). Initial hidden state: (a) No creatures in the vicinity; (b) Distal cat; (c) Distal dog. Each bar depicts all policies in fixed ascending order (left to right), colour-coded (see key below bars), to aid comparison across bars of the relative increase or decrease in policy selection frequency. The variants serve to reveal the key influences on policy selection. I and II (unimpaired) are defined by a moderate prior preference; III and IV (impaired) are defined by an over-strong prior preference; I and III (safe) are defined by flat priors; and II and IV (unsafe) are defined by strong priors for a distal cat. (b). Policy selection distribution over 32 trials of four simulated variants (I-IV) for Scenario 2 (Initial outcome: dot on horizon visible – see Fig. 3b). Initial hidden state: (d) Distal cat; (e) Distal dog. (For further details, please see Fig. 5a caption and main text.).