The Importance of Multi-Agent Reinforcement Learning

The team created environments where the only way the agents could possibly solve the problems was if they were to send information to each other. They also put some of these bottlenecks like shumpling before the softmax and so on to force the models to use discrete communication. The thing that was very hard to fake your art for me is the next step when eventually one monkey sort of learns to manipulate that symbol or starts thinking you know I would like to get all this food but there is this other makshit that I don't like. There's just being some progress but there's still we still have a long way to go, he says.