Will Deep Mind become chief diplomat soon?
New York, June 11, 2020
An AI developed at DeepMind achieved high scores and improved continuously in the seven-player board game “Diplomacy”. Researchers at the Alphabet-supported AI Lab believe that the AI could lead to new research in the field of deep reinforcement learning, i.e. training machine learning agents to make decisions about rewards.
The preprint paper was published on arXiv.
The DeepMind agents surpassed earlier state-of-the-art models. Research in the field of diplomacy could lead to the creation of artificial agents that can successfully cooperate with others, which according to co-authors involves “building and maintaining trust and alliances”.
The system’s profit rates of 12.7% to 32.5% represented a remarkable improvement over supervised, adaptive agents.
DeepMind has used AI to beat human players in chess, Montezuma’s Revenge, Go, and Starcraft 2. Whether they will replace human diplomats is still open.