Jakob Foerster
TalkRL: The Reinforcement Learning Podcast
00:00
Deep RL: A Zero-Sum Game
In the last few years, Hanabi has been trying to develop novel methods in which computers can support and help humans using large-scale compute. The challenge is that these policies are incompatible with settings whereby the equilibrium can't be jointly chosen for everyone in the team. And understanding this is important because when machines meet humans, often the problem setting will be known but the ability to specify a policy isn't there. But it would be quite hard to explain to a human that this is the policy you're playing.
Transcript
Play full episode
