Earlier this year, an amateur Go player cleanly beat a top-ranked AI system with the help of a strategy devised by researchers. Its chosen strategy targeted finding weaknesses in AI players such as KataGo, and it turned out that this victory was only part of a human counterattack. Since AlphaGo’s landmark victory in 2016, human chess players have become increasingly creative in recent years, according to a new paper published in the journal PNAS by a team from City University of Hong Kong and Yale University.
Researchers have developed a “superhuman” Go AI solution that can play against people and score the effect of each move. It can count the so-called “decision quality index.” After analyzing more than 5.8 million moves in all professional games from 1950 to 2021 with this goal in mind, the team concluded that the way humans play chess is becoming increasingly unpredictable. According to the results they obtained, the annual index increase of professional chess players before 2016 was relatively small. The median change over this period was 0.2, and in some years the overall quality of the game even declined.
However, since the large-scale rise of AI chess players in 2018, the median change of human professional chess players has also exceeded 0.7. During this period, humans employed more novel tactics. In 88% of the games in 2018, humans came out of a combination of moves that had not appeared before. For comparison, this proportion was only 63% in 2015. “Our findings suggest that the development of superintelligent programming may prompt human chess players to break with tradition and lead them to explore novel moves, which may in turn improve their decision-making performance,” the research team wrote. .
This change is very interesting, but it is actually expected. “It’s not surprising that chess players trained with machines are more likely to make more machine-approved operations,” said Professor Stuart Russel of the University of California, Berkeley.