Maximizing Human Agents' Cooperative Behavior in Collective Action Games Through a Deep Q-Learning AI Agent