Noticing that this could also be how a model-based reinforcement learning algorithm would behave, using the patterns it already knew to try to elicit the results expected.