|
Briefing
|
- DeepRole – Researchers at MIT developed gaming bot DeepRole, artificial intelligence algorithm equipped with deductive reasoning that can identify which player in game of Avalon is an ally or opponent and what actions it should take to win
- Avalon – Multiplayer game where players are randomly assigned as resistance or spy, with players having to deduce which players are from which team, and must vote whether a mission succeeds or fails to win
- What It Does – Algorithm creates decision tree of potential actions for each player, comparing players’ reactions to its expectations to deduce probability of player being a resistance or spy, and estimating how other players may react in voting to make more intelligent decisions
- Training – Bot played against itself multiple times in both roles as resistance and spy, learned which actions taken increased or decreased its chances of winning, and refined its strategy
- Testing – Bot outperformed human players consistently in over 4,000 rounds
- Next Steps – Include potentially adding capability for bot to communicate with other players through simple text and expanding to more language-heavy social deduction games like Werewolf
|
|
Accelerator
|
|
|
Sector
|
Information Technology, Media and Entertainment
|
|
Function
|
Research and Development
|
|
Organization
|
Massachusetts Institute of Technology
|
|
Source
|
|
|
Original Publication Date
|
November 19, 2019
|
