AI beats eight world champions at Bridge

• 

Eight world champions have been defeated by artificial intelligence in the bridge, a game in which human superiority has resisted the march of the robots until now.

Because bridge players operate with imperfect knowledge and must react to the conduct of several other players – a scenario significantly closer to human decision-making – the triumph signals a new milestone for AI.

In contrast, in chess and Go – both of which AIs have already defeated human champions – a player only has one opponent at a time, and both have complete information.

“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.

What is NukkAI?

NukkAI, a French startup, claimed its AI's triumph on Friday, at the conclusion of a two-day tournament in Paris.

The human champions had to play 800 consecutive deals divided into 80 sets of ten for the NukkAI challenge. It omitted the game's initial bidding component, in which players arrive at a contract that they must subsequently meet by playing their cards.

Each champion played their own cards as well as the cards of their "dummy" partner against a pair of opponents. These opponents were the world's best robot champions to date — robots that have won numerous robot tournaments but are unanimously recognized to be nowhere near as good as expert human players.

Read more: China Beats the US in AI Race

The AI, known as NooK, took on the same role as the human champion, using the same cards and facing the same opponents. The score was calculated as the difference between the human and AI scores, averaged over each set. NooK won 67 of the 80 sets or 83%

A co-founder Jean-Baptiste Fantun said he was convinced the machine – which the business has been building for five years – will win thousands of deals, but with only 800, it was a toss-up.

NukkAI was described as a "superb French success story" by mathematician Cédric Villani, who won the Fields award in 2010.

Its co-founder, AI researcher Véronique Ventos, refers to NooK as a "new generation AI" because it explains its decisions as it goes. "You can't play bridge unless you explain," she says.

How is the game played? 

Explainability is a hot problem in artificial intelligence. "Most of what the general public has heard in recent years about machine learning is based on black-box algorithms like AlphaGo, which are unable to explain to humans how judgments are made," Muggleton added.

NooK, on the other hand, represents a "white box" or "neurosymbolic" approach. Rather than playing billions of rounds of a game, it first learns the laws of the game and then improves its play via practice. It is a cross between rule-based and deep learning systems. "The NooK technique learns in a much more human-like manner," Muggleton said.

“The pendulum is swinging towards these kinds of methods,” says Michael Littman, a professor of computer science at Brown University in Rhode Island. “Not being able to tell people what’s going on just doesn’t work in our societies.”

Even if a person or AI cannot describe what they are doing in words, Littman believes that their behavior must be "legible" to others, enacting laws that they comprehend.

This will be crucial in fields such as medicine and engineering. Self-driving cars negotiating a junction, for example, will need to be able to read each other's behavior.

Read more: US Intel: China Could Dominate AI Technology

Littman was upset that the task did not include bidding, which is where much of the most engaging communication – and deception – in bridge takes place.

However, Nevena Senior, a multiple-time world bridge champion for England and one of NooK's adversaries, claims that the contracts the humans and NooK were given to play were so unpredictable that the card game became as essential as the bidding.

She praised NooK's creators for their "magnificent" work. She discovered that it read its opponents better than humans did and was better equipped to capitalize on their blunders.

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