People have gotten better at Go since AI became the best in the world.

AI can beat the best players in the world at the Go board game, but humans are starting to improve too. An analysis of millions of moves in Go has shown that professional players have been making better and more original game decisions ever since AIs playing Go have overtaken humans.

Before 2016, AI couldn’t beat the best Go players in the world. But that all changed with the advent of artificial intelligence called AlphaGo, developed by London-based research firm DeepMind. AlphaGo defeated several Go champions, including the then-best human player.

Since then, other AIs have been developed that are considered “superhuman”. While they can be used simply as opponent players, they can also help analyze the quality of any given move and thus act as a Go coach.

Minkyu Shin of the City University of Hong Kong and his colleagues set out to investigate whether the advent of these superhuman AIs for playing Go resulted in a noticeable improvement in the human game.

The researchers collected a dataset consisting of 5.8 million moves taken by professional players between 1950 and 2021. They then used Go AI to calculate a metric called the Decision Quality Index, or DQI, that measures the quality of a move. . They considered a move “new” unless it had previously been taken in conjunction with previous moves.

The analysis showed that human players made significantly better and newer moves in response to the emergence of superhuman AI in 2016. Between 1950 and 2015, the improvement in game quality was relatively small, with the average annual DQI hovering around -0.2 to 0.2. Whereas after superhuman AI, DQI jumped up, with an average above 0.7 from 2018 to 2021. In 2015, 63% of games featured new strategies, while by 2018 this figure had risen to 88%.

Stuart Russell of the University of California at Berkeley says the improved human game of Go is reminiscent of a phenomenon in the 1990s when backgammon players began to change opening moves in response to the emergence of highly skilled computer players. The fact that artificial intelligence is doing the evaluation also plays a role, he says. “Not surprisingly, players who train with machines tend to make more moves that the machines approve of.”

The document shows cultural transmission from AI playing Go back to humans, says Noah Goodman of Stanford University in California. “Thing [the paper] makes me seriously think that right now we are seeing another dramatic change in AI, namely chatbots. What dramatic changes will we see across cultures as a result of interacting with and learning from chatbots?”