- Aristides Zenonos
Multi-agent AI system masters collaborative game
So far, Reinforcement Learning (RL) AI systems have been good at tackling self-play games such as “Poker” and “Go”. However, researchers working with Facebook have now created an AI system that is at least as good as expert human players at the game of “Diplomacy”. The game requires the players to cooperate with each other by making deals and alliances. The RL algorithm imitates how humans play. What is impressive is that, when asked, the other players picked the AI agent as the strongest player that they would like to cooperate. This result points to the direction that, contrary to what most people believe, AI systems and humans will complement and not compete with each other in performing tasks.
https://arxiv.org/pdf/2210.05492.pdf
