Interpretable Learned Emergent Communication for Human-Agent Teams

Learning interpretable communication is essential for multi-agent and human-agent teams (HATs). In reinforcement learning partially-observable environments, agents may convey information to others via learned communication, allowing the team complete its task. Inspired by human languages, recent works study discrete (using only a finite set of tokens) sparse (communicating at some time-steps) communication. However, utility such in experiments has not yet been investigated. this work, we analyze efficacy sparse-discrete methods producing emergent that enables high agent-only performance. We develop communicate sparsely our scheme Enforcers sufficiently constrain any budget. Our results show no loss or minimal performance benchmark environments tasks. tested where have modeled using Enforcers, find prototype-based method produces meaningful tokens enable partners learn agent faster better than one-hot baseline. Additional HAT an appropriate sparsity level lowers cognitive load humans when communicating with leads superior