After their failures against artificial intelligence, will humans regain the upper hand?

A few years ago, AI beat many human champions in the game of Go. However, this board game is considered one of the most difficult for a machine to grasp. Only, here it seems that humans have improved since then.

Artificial intelligence is at the top of the ladder

In 1996, the IBM Deep Blue computer beat Gary Kasparov, World famous chess champion. This was the first time that an AI had reached the end of a world champion during a game. Then the AI ​​attacked other games such as backgammon, poker or even go game, the latter for many observers, The hardest board game Humans invented it.

However, DeepMind’s AlphaGo AI beat European champion Fan Hui in 2015, one of the best players in the world, Lee Sedol in 2016 and finally world champion Ke Jie in 2017. In 2019, Lee Sedol announced his retirement and announced on the occasion The advances made by the AI ​​in the go game made him realize he no longer had the weapons to compete.

In addition, you should know that an improved version of AlphaGo appeared in 2017: AlphaGo Zero. The latter soon found themselves at the top of the ladder, yet Beat the old version a hundred times in a row.

Humans still have room for improvement

Can humans ever recover from their failures? The answer is yes, according to a study by a team from the University of Hong Kong (China) published in the journal PNAS on March 13, 2023. Researchers conducted an analysis of approx 5.8 million movements Carried out between 1950 and 2021 by professional players. However, they did find an improvement in what the scientists called a “decision quality index.” For example, this indicator rose in the case of original moves that had not been previously tried with other moves.

According to the results, the annual evolution of the decision quality index was between a slight decrease of 0.2 and a slight increase of 0.2. However, the indicator contains by 0.7 between 2018 and 2021. This translates, for example, to an average of 88% of games played in 2018 showing original strategies, compared to an average of 63% prior to 2015.

Remember, however, that Go, which aims to surround the opponent, presents several possible scenarios for each move. In fact, the board is wider and the games are longer. For Minkyu Shin, lead author of the study, the characteristics of this game make humans There is still room for improvement. Thus, encounters with AI allowed them to better study the game of Go, and thus develop new strategies and new moves. By gradually moving away from traditional strategies, players have been able to improve their decision-making process.