-
Abalance presents Went.ai
-
Beta 0.6 Ingame view | begin
-
Beta 0.6 Ingame view | endgame
-
ICC 2020 Lucien Jeanprêtre & Valentin Schneeberger
-
ICC 2020 Dr. Luc Pétiard & Valentin Schneeberger
Abstract strategy games have been historically used to demonstrate to a generic audience our technological progresses in the field of AI; 1997 Chess (Kasparov vs Deepblue), 2016 Go (Lee Sedol vs Alphago). The rise of deep learning technologies seemed to allow the emergence of a new kind of game in the abstract strategy space: One that goes away from the confrontation paradigm in which human and machine evolves.
Went.ai is specifically designed in its core gameplay to partner human and AI in intriguing and unexpected way that both improve gameplay as well as build a better understanding of what AI is in general. In this game, players team-up with AI to defeat the opponent human-machine team. Went.ai will allow players to build, train and parameter their own deep learning agents, until they are ready to be brought to battle as their teammates... Or until they betray their creators.
The game takes the shape of a web app written in PHP, Python and React. It involved a team of 15 individuals from various fields: Artists, designers, engineers, mathematicians and business experts, all reunited in their passion for chess and go, AI, innovation and impact.
The game being new, we didn't have the datas to define that a move would be good or bad in this game, which complicated the construction of capable AI. We needed to attack the problem from a different perspective and we developed a method to address this issue by considering the deepness of similar games, then calculate the theoretical deepness of ours in order to deduce the threshold at which we can expect unsupervised learning shall give meaningful results.
Our prototype to test the gameplay human to human is ready and will be released this year! Our "tweakable" deep learning agent "Abazero" is already functional on simpler games.
During this journey we have come a long way, and learn how to work as a team with various skills and views, we also significantly increased our knowledges in deep learning and machine learning, which brought us to consider even more ideas and directions for the project to be more than a game, we are now researching towards practical applications of our technology for industry 4.0.
What's next for Went.ai is the release of its Open Beta to test gameplay human to human first, then its update into final release: the implementation of its test results coupled with our AI interactions system.
