Reinforcement learning is used to construct counterexamples to a conjecture relating the index and matching number of a graph. The possibility of applying RL to Sperner families is also investigated. This is posted as a short online book.