Sign In

Communications of the ACM

ACM TechNews

Facebook's AI Masters the Card Game Hanabi


View as: Print Mobile App Share: Send by email Share on reddit Share on StumbleUpon Share on Hacker News Share on Tweeter Share on Facebook
The card game Hanabi.

Facebook researchers taught the company's artificial intelligence software how to play a Solitaire-like card game that requires players to work together.

Credit: R&R Games

Researchers at Facebook have taught the company's artificial intelligence (AI) software how to play the game Hanabi, a Solitaire-like card game that requires players to work together.

Hanabi is viewed as a capable testbed for AI because it requires teamwork and strategy.

The Facebook researchers had to find a way to give the Hanabi bots a method to understand the hints of their teammate bots, based on the limited information they have about their own cards.

The researchers helped the bots using a variant of the Monte Carlo "search" technology technique to help them evaluate their possible moves. The modified technique uses a combination of reinforcement learning and the so-called multi-search technique, which helped the bots learn to play Hanabi on their own and in tandem with one another.

The researchers found that digital bots playing the game learned to help each other achieve higher scores than previous AI systems.

From Fortune
View Full Article

 

Abstracts Copyright © 2019 SmithBucklin, Washington, DC, USA


 

No entries found