Behavior of trained reinforcement learning agent.

(A) The accuracy that the agent achieves at different levels of signal strength (i.e. 100—% noise). Less signal (i.e. more noise) makes it harder for the agent to classify correctly (solid line represents fit of psychophysical function). (B) The average number of observations that the agent uses before making a decision on the class. More observations are made before deciding for trials with lower signal strength (i.e. higher noise levels) (solid line represents linear fit). (C) The distribution over trial lengths for different noise levels shows that higher noise leads to longer reaction times, but also to more variance in the reaction times.