As you all know we’ve been working with some third parties on developing Artificial Intelligence which can play the game Diplomacy. AI researchers have been interested in Diplomacy as a benchmark for AI research because of the many challenges inherent to the game: multiplayer cooperation and competition, acting under partial information due to the simultaneous turns, planning across multiple game phases, and more. If you haven’t already, you can check out the various bots on the Create Bot Game page.
I’m excited to announce that we’ve attracted interest from one of these organisations in supporting the website. I’m sure you all know that our team hasn’t had as much time for hobbyist development as we would like, and this has left our mid-2000s site in a mid-2000s state (despite some great improvements that have come in over time). Our research partners want to invest in contributing to the site under its existing open source license with a new design that will refresh the visuals and improve the experience playing, creating, and searching for games.
We have worked with MILA in the past, the Montreal Institute of Learning Algorithms, and the upcoming open source contributions are being done by Codazen with funding from FAIR, Meta’s fundamental AI research group. FAIR would like to help support the Diplomacy player community, to further the game as a go-to setting for research in Artificial Intelligence, and to make playing (including against bots, in particular) a better experience. All of these changes will be open-sourced on the WebDiplomacy github page, so please feel free to directly provide feedback on the changes, open issues if you find bugs, or even submit further code improvements for those inclined.
Some of the key things to be worked on include:
- refreshing the in-game interface to make it more beautiful and add
- point-and-click options for entering orders
- making the in-game interface more friendly to mobile devices
- more control over notifications to make it easy to keep up on active games
- easier game browsing and creation
- more responsive chat interface
- better onboarding for new players, including a better experience for using the AI to practice / learn how to play
- MILA’s No Press Diplomacy: Modeling Multi-Agent Gameplay
- Google DeepMind’s Learning to Play No-Press Diplomacy with Best Response Policy Iteration
- FAIR’s Human-Level Performance in No-Press Diplomacy via Equilibrium Search
- FAIR’s No-Press Diplomacy from Scratch (you can play against this bot by launching a "France vs Austria" bot game)
Hope you’re all as excited as I am to see these changes come through. Big thanks to everyone that has been involved in making this happen behind the scenes.
You should start to see some new functionality introduced as opt-in features as things get moving, we'll post more info as it comes.
BR,
Kestas