Learn which shape earns a reward, then adapt when the rule silently reverses. Rewards are probabilistic - not every correct choice is rewarded.
Scientific Background
Based on: Probabilistic Reversal Learning (Cools et al., 2002) (Cools et al., 2002)