Two machines are better than one: Using multi-agent reinforcement in machine learning to simulate a clinical second