Bridge is a card game of incomplete information that demands cooperation and sly communication. Over the years, human body language has evolved to read subtle facial queues and body language. One might believe these are matters beyond the reach of a machine. But it is not. 

A few days ago, a French startup NukkAI took a step forward when its bridge-playing AI outplayed eight bridge world champions in a competition held in Paris. The game was simplified, and the model did not go head-to-head with the players. The algorithm’s performance was otherwise spectacular. 

The Guardian reported that the victory is a milestone for AI because, in bridge, players must act to the behavior of several other players- a scenario far closer to human decision-making. In contrast, chess and Go- in both of which AIs have already defeated human champions- a player has a single opponent at a time, and both are in possession of all the information.  

According to a professor of ML at Imperial College London, the game proved an essential advance in the state of the AI systems. 

AI’s game victories 

It was in 1996 that IBM’s Deep Blue chess machine won a game against the world’s chess champion Garry Kasparov for the first time. Later in 2007, researchers at the University of Alberta in Canada built a checkers-playing computer that could not be defeated.  

In 2011, IBM’s Watson computer defeated the TV game show Jeopardy! champions Brad Rutter and Ken Jennings, claiming the $1m first prize. And in 2016, Google DeepMind’s AlphaGo defeated the Korean Go champion Lee Sedol and won the highest Go grandmaster rank.  

NukkAI’s win 

NukkAI announced the news of its victory on Friday at the end of a two-day tournament in Paris. The challenge required human champions to play 800 consecutive deals divided into 80 sets of 10. It did not involve the initial bidding component of the game, during which players arrive at contact that they must meet by playing their cards. 

A bridge game begins with players on how many tricks, or rounds of play, they think they can win. The highest bid is called the contract and whoever sets the contract is the declarer. The declarer’s partner, or the dummy, lays their hand down on the table face up and exists the game. The declarer then plays both hands and tries to win enough tricks. 

Each champion played their own and their “dummy” partners cards against a pair of opponents. These opponents were the best robot champions in the world to date- robots that have won many robot competitions but that are universally acknowledged to be nowhere near as good as expert human players.  

The model played the same role as the human champion, with the same cards and opponents. The score was the difference between humans and AI, averaged over each set.  

Neurosymbolic approach 

The general public might have heard in recent years that ML is based on black box systems such as Alpha Go, which cannot explain how decisions are made to human beings. In this case, the model represents a “neurosymbolic” or a “white box” approach. Rather than playing billions of rounds, it first learns the game’s rules and then improves its play through practice. It is a hybrid of rules-based and deep learning systems. 

AI outscoring top human players in part of the game is a crucial milestone. While bigger AI algorithms like OpenAI’s GPT-3 continue to impress, the model’s performance in bridge may add weight to the argument for a hybrid approach. 

Source: The Guardian

Image Source: Unsplash