Using Reinforcement Learning to Manage Communications Between Humans and Artificial Agents in an Evacuation Scenario