Probing reaction channels via reinforcement learning