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

AIs can beat the world’s best players at the board game Go but humans are starting to improve as well. Analysis of millions of Go moves found that professional players have been making better and more original game choices since Go-playing AIs surpassed humans.

Before 2016, AIs couldn’t beat the best Go players in the world. But that changed with an AI called AlphaGo developed by London-based research company DeepMind. AlphaGo has defeated many Go champions, including then number one ranked human player.

Since then, other AIs that are considered “superhuman” have also been developed. Although they can only be used as opposition players, they can also help in evaluating the quality of any given move and thus also act as a Go coach.

Minkyu Shin at the City University of Hong Kong and colleagues decided to investigate whether the introduction of these superhuman Go-playing AIs led to a noticeable improvement in human playing.

The researchers gathered a dataset consisting of 5.8 million transfer decisions of professional players between 1950 and 2021. They then used a Go-playing AI to help calculate a measure called the “index of decision quality”, or DQI, which assesses the quality of a transfer. They considered a move “novel” if it had not been tried before with previous moves.

The analysis found that human players made significantly better and more novel moves in response to the arrival of superhuman AI in 2016. Between 1950 and 2015, the improvement in the quality of play was relatively small, with median annual DQI hovering between approximately -0.2 and 0.2. Whereas after superhuman AI, DQI jumped upwards, with median values ​​above 0.7 from 2018 to 2021. In 2015, 63 percent of games featured new strategies, while in 2018, that number has increased to 88 percent.

Stuart Russell at the University of California, Berkeley, says the improved human playing of Go resembles a phenomenon in the 1990s, when backgammon players began changing opening moves in response to the advent of skilled computer players . The fact that it’s an AI doing the assessment also plays a role, he says. “It’s not surprising that players who train against machines tend to make more moves that are approved by the machines.”

The paper shows cultural transmission from Go-playing AIs back to humans, he said Noah Goodman at Stanford University in California. “The thing [the paper] So hard that I think we are now seeing another sudden change in AI chatbots. What sudden changes will we see in different cultures as a result of interaction and learning from chatbots?”