It doesn't miss a trick! Artificial intelligence beats EIGHT world champion bridge players at their own game


SOURCE: DAILYMAIL.CO.UK
MAR 29, 2022

Bill Gates famously described bridge as 'one of the last games in which the computer is not better'.

But the Microsoft co-founder will be eating his words this week, following the news that an artificial intelligence bot has managed to beat not just one, but eight world champion bridge players at the game.

French startup NukkAI spent four years developing the AI bot, called NooK, which took home the crown at the two-day Nukkai Challenge in Paris last week.

An artificial intelligence bot has managed to beat not just one, but eight world champion bridge players

An artificial intelligence bot has managed to beat not just one, but eight world champion bridge players

In bridge, each of the four players, split into two teams, receives 13 cards in a hand. The object of bridge is to win tricks

In bridge, each of the four players, split into two teams, receives 13 cards in a hand. The object of bridge is to win tricks

NooK was trained using a 'hybrid' approach

While other AI systems are typically trained by playing billions of rounds of a game, NooK was trained using a hybrid approach.

The bot first learned the rules of the game, before improving its play through practice.

'The NooK approach learns in a way that is much closer to human beings,' Professor Muggleton said.

While AI bots have previously beaten humans at various games including chess and the ancient game Go, this is the first time that AI has won at bridge – a game in which players must react to the behaviour of several others to win.

'What we've seen represents a fundamentally important advance in the state of artificial intelligence systems,' said Professor Stephen Muggleton, an expert in machine learning at Imperial College London, speaking to The Guardian.

NukkAi announced the news of the victory on Twitter at the end of the tournament last week, writing: '[Robot emoji]: 6136 [Human emoji]: 5238. NooK won The Nukkai Challenge!'

While other AI systems are typically trained by playing billions of rounds of a game, NooK was trained using a hybrid approach.

The bot first learned the rules of the game, before improving its play through practice.

'The NooK approach learns in a way that is much closer to human beings,' Professor Muggleton said.

NukkAi announced the news of the victory on Twitter at the end of the tournament last week, writing: '[Robot emoji]: 6136 [Human emoji]: 5238. NooK won The Nukkai Challenge!'

NukkAi announced the news of the victory on Twitter at the end of the tournament last week, writing: '[Robot emoji]: 6136 [Human emoji]: 5238. NooK won The Nukkai Challenge!'

Bridge: The basics and scoring

Each of the four players, split into two teams, receives 13 cards in a hand. The object of bridge is to win tricks.

Each person will 'bid' for tricks, based on how many they believe their cards can help them win.

They will make this 'bid' based on the number of scoring cards they have.

These are as follows:

Ace: Four points

King: Three points

Queen: Two points

Jack: One point

There are 13 tricks to be won and a trick consists of four cards - one from each player.

The highest card wins the trick, unless a trump card is laid.

In this instance, experts have suggested that the winning hand contained a jack, four kings and four aces.

During the two-day tournament, the eight world champions played 80 sets of 10 – a total of 800 consecutive deals.

The game did not involve the initial bidding component, in which players arrive at a contract that they then must meet by playing their cards.

The human players played their own and a dummy partner's cards against NooK, with the final score being the difference between those of the human and the bot, averaged over each set.

At the end of the tournament, the results showed that NooK won 83 per cent (67) of the 80 sets.

Nevena Senior, one of the human champions beaten by NooK, took the defeat well, and said the AI had done a 'magnificent' job.

She even went so far as to say that the AI read its opponents better than humans did, allowing it to better exploit their mistakes.

'This is something that humans do after enough experience and I was pleasantly surprised that a robot mimics typical human skills,' she said.

While winning the bridge tournament is all fun and games, the hybrid-training approach could be applied in a range of important domains in the future, including self-driving cars.

'As AI progresses, a central challenge is to move beyond simple games with complete information and towards functioning a more complex real world filled with human agents and incomplete information,' said Gary Marcus, founder of Geometric Intelligence.

'Bridge offers an important microcosm that should be interesting to watch.'

Similar articles you can read