Artificial intelligence (AI) has reached another level as a new program managed to beat eight world champions simultaneously in the card game bridge. The win is a landmark in AI development as a game of bridge involves the computer playing against several players at once with an incomplete set of information.

Most importantly, the computer must also play reactively to the behaviour of the human players at the table. It is for that reason that until now no AI had defeated human players in the game before.

Also read: How far Artificial Intelligence has come, and what the future looks like

“What we’ve seen represents a fundamentally important advance in the state of artificial intelligence systems,” said Stephen Muggleton, a professor of machine learning at Imperial College London.

AI has famously beaten humans at games like chess and an ancient board game, go, before -- IBM’s Deep Blue winning against chess legend Gary Kasparov in 1997 and Google DeepMind’s AlphaGo defeating go champion Lee Sedol in 2016.

Also read: How is AI slowly eating up white-collar jobs?

The bridge AI, NooK, was created by French startup NukkAI, which emerged victorious after the two-day tournament in Paris. The AI managed to win 83 percent of the 80 sets of 10 deals each that it competed in.

Though it was not directly playing with other human players. Instead, the NooK and the human champions were playing against the previous best AI bridge players that had been created previously. The champions and NooK had the same hands and same opponents, with the score being determined by the difference between the other bridge AIs and the human players.

Also read: Explained: How AI can change future of modern warfare

While the tournament did not include the initial bidding phase of the game, NooK’s success comes from a new type of learning that companies are using to train their AIs. Instead of relying on traditional black-box systems, where a machine learns how to play a game by brute-forcing through thousands of games to learn the rules, NooK first learns the rules of the game and then improves upon the play through practice.

This white hat system of learning is much closer to how actual humans learn new skills and presents a new possibility for AI to start learning skills in critical domains like health and engineering.

(Edited by : Shoma Bhattacharjee)

First Published:Apr 5, 2022 8:37 PM IST