AI multi-agent game breakthroughs and advances have almost become the norm in recent years. However, these games hadn't yet established methods of addressing real-life challenges of team cooperation while playing with or against uncertain or unknown team members.
This is crucial for hidden-role multiplayer games.
Now, MIT researchers have created a bot that can play and beat human players at interactive multiplayer hidden-role online games.
Named DeepRole, the bot is a multi-agent reinforcement learning agent that works with Artificial Intelligence (AI).
The bot and the game
This is an exciting advancement as DeepRole is the first bot that can beat humans at online games where the players' allegiances aren't clear at the start of the game.
Structured with innovative "deductive reasoning" that's added into an AI algorithm typically used when playing poker, the bot can reason with only partially observable actions. The bot then figures out whether or not a player is a friend or a foe.
Happy to share what we’ve been working on this year - an agent that can play Avalon at Human-level performance. It can find cooperators in adversarial environments to win the game in a variety of team mixtures. https://t.co/ehPMBu3FnF— Jack Serrino (@Detry322) June 7, 2019
Jack Serrino, the first author of the paper and MIT graduate in electrical engineering and computer science, said "If you replace a human teammate with a bot, you can expect a higher win rate for your team. Bots are better partners."
Co-author, Max Kleiman-Weiner, MIT post-doctoral student at the Center for Brains, Minds, and Machines, and the Department of Brain and Cognitive Science added that "Humans learn from and cooperate with others, and that enables us to achieve together things that none of us can achieve alone. Games like ‘Avalon’ better mimic the dynamic social settings humans experience in everyday life. You have to figure out who’s on your team and will work with you, whether it’s your first day of kindergarten or another day in your office."
DeepRole's AI algorithm
The MIT researchers used an AI algorithm on the bot called 'counterfactual regret minimization' (CFR). This algorithm worked out how to play a game by repeatedly playing against itself.
At each point in the game, CFR uses a 'game tree' of lines and nodes that describe the potential future actions of all players.
'Game trees' represent every possible action a player in the game can take at every decision point.
The next steps the researchers are looking at are developing ways to teach the bot to communicate with other players during a game by using simple text.