A program developed by the French company NukkAI beat eight world bridge champions last week. A real challenge, due to the rules of this card game which is played with four players (two against two) and requires making decisions based on incomplete information and the behavior of other players. Tactical and collective, it is considered the most complex of card games.
“The game that most closely resembles real life”
“Bridge is the game that most resembles real life: the information is partial, there are partners and opponents, and the rules impose explainability. It is the ideal test for the AIs that we design”explains the start-up on its website.
After chess, the game of Go and poker (without forgetting “Jeopardy”), this is therefore a new advance for the AI. The bridge had so far resisted attempts to develop automated systems. “What we have seen represents a fundamental advance in the field of artificial intelligence systems”believes Stephen Muggleton, professor of machine learning at Imperial College London, interviewed by the Guardian.
The computer won 83% of the rounds
For this experiment, each champion had the NukkAI system as a partner, facing two robots who had already won tournaments between automated systems, but never against champions in the discipline. 800 tricks were played, each champion having the same hands as the NukkAI system (called Nook) in order to compare the results.
The bidding phase that normally exists at the start of the game, which consists of reaching a “contract”, in which one of the two pairs of players undertakes to perform a certain number of tricks, was not part of the experience. The AI won 83% of the rounds.
The mathematician and deputy of Essone Cédric Villani described this success as“historic feat”. “These are fascinating perspectives on the understanding of probabilistic reasoning, with hidden information and explainability of algorithms”he wrote on LinkedIn.
A collaborative AI
Founded in 2018 by Véronique Ventos, a researcher in artificial intelligence, professor at Paris Saclay and ranked 59th French bridge player, and Jean-Baptiste Fantun, agrégé in mathematics who has also had responsibilities within ministerial cabinets, consultant specializing in health digital and winner of the French Bridge Cup in 2017, NukkAI focuses its research on AI intended to collaborate with humans. They find applications in the aeronautics, education, cybersecurity and defense sectors.
Technologically, NukkAI’s AI is based on an alternative learning system to that usually used in machine learning. Marko Erman, scientific director at Thalesa partner of the start-up, explains it this way on LinkedIn: “For several years, the requirements of complex systems (such as explainability and certification, for example) have led Thales to explore alternatives to conventional stochastic AI approaches (neural networks + learning to put it simply). The hybridization of Stochastic AI with symbolic AI seems to us to be one of the most interesting options. The NukkAI approach in particular is very interesting and the recent results obtained are very promising”.
Developed using a CNRS supercomputer
NukkAI is also a partner of the CNRS, which has made its Jean Zay supercomputer available to it, and of the BPI who supported the project as part of the Deep Tech Development Aid. With approximately 100,000 licensees and 1,300 clubs, France is the world’s second bridge nation behind the United Statesand regularly wins medals at the world championships.
We want to give thanks to the author of this write-up for this outstanding content
Artificial intelligence beats world bridge champions for the first time
You can find our social media profiles here as well as other pages on related topics here.https://www.ai-magazine.com/related-pages/