Humans have improved at Go since AIs have become best in the world

AI can beat the world’s best players at the board game Go, but humans are starting to improve, too. An analysis of millions of Go moves shows that professional players are making better and more creative game choices since Go-playing AI surpassed humans.

Prior to 2016, AI could not beat the world’s best Go player. But that has changed with an AI called AlphaGo, developed by London-based research firm DeepMind. AlphaGo beat multiple Go champions, including the top ranked human player at the time.

Other AIs deemed “superhuman” have also been developed since then. It can be used simply as an opponent, but it also helps analyze the quality of certain moves and acts as a Go coach.

Minkyu Shin and colleagues at the City University of Hong Kong decided to investigate whether the introduction of these superhuman Go AIs led to significant improvements in human Go.

The researchers collected a dataset consisting of 5.8 million movement decisions made by professional players between 1950 and 2021. Go AI was then used to calculate a measure called the “Decision Quality Index” or DQI, which evaluates the quality of the moves. They considered the move “novel” if it had not been previously attempted in combination with the previous move.

Our analysis found that human players performed better and novel moves in response to the advent of superhuman AI in 2016. Between 1950 and 2015, the improvement in quality of play was relatively small, with median annual DQI varying between approximately -0.2 and 0.2. Meanwhile, after superhuman AI, DQI skyrocketed, with a median of over 0.7 from 2018 to 2021. In 2015, 63% of games showed novel strategies, but by 2018 that number had risen to 88% of his.

Stuart Russell of the University of California, Berkeley, says the improvement in human Go play resembles the phenomenon of the 1990s. This phenomenon was when backgammon his players began to change their initial moves in response to the emergence of highly skilled computer his players. The fact that he’s an AI doing the evaluation also plays a role, he says. “It’s not surprising that players who train against machines tend to make more moves than the machine approves.”

The paper shows cultural transfer from Go-playing AI to humans, said Noah Goodman of Stanford University in California. “case [the paper] What really gets me thinking is that we’re seeing another sudden shift in AI, chatbots right now. What kind of radical changes do you see in different cultures as a result of interacting with and learning from chatbots?”