Artificial intelligence not only plays chess better than humans, but also handles complexity differently.
For decades, chess has served as a laboratory for studying intelligence and decision-making. Current AI- powered chess engines are known to be capable of outperforming even the strongest top players, but what is less understood is how their gameplay actually differs. By analyzing the interactions step by step in more than 2,000 matches between human players and matches between artificial intelligence systems, a team of researchers at the Center for Complexity Science and the University of Parma in Italy found a fundamental difference in how humans and artificial intelligence deal with what they called "strategic tension". “Humans and artificial intelligence differ not only in their level of skill at play, but also in how they deal with complexity. Human players reduce complexity quickly, while artificial intelligence can maintain it,” said Vito D. B. Cervedio, lead researcher on the study and senior scientist at the Center for Complexity Science, according to a report by the technology news site Tech Xplore, which was reviewed by Al Arabiya Business. The researchers measured this difference by determining the number of possibilities that remained open after each move. The wider the range of possibilities available at any given moment, the greater what the researchers described as "strategic tension." At the beginning of a chess game, strategic tension begins to rise as the game progresses, since each move reshapes the network of possible moves, further complicating the situation.
In the middle of the match... humans and artificial intelligence part ways
Researchers found that both human players and AI systems raise their "strategic tension" levels during the opening and early middle of a game. Even the best human players briefly reach a level of tension higher than that of chess engines, which are AI programs specialized in playing chess and analyzing moves. But then, the paths diverge. The top players usually start swapping pieces and simplifying the situation once they get out of the opening lines that were prepared, studied, and memorized beforehand. In contrast, advanced chess engines, such as Stockfish and Leela Chess Zero, maintain highly interconnected modes for a much larger number of moves.
In other words, humans give up this tension, while chess engines maintain a higher level of complexity on the board for a much longer period. The study, published in the journal "APS Open Science", showed that the ability to maintain strategic tension increases with the level of play.
Stronger human players maintain complex positions for longer than weaker players, and classical games, in which players have more time to think, have a much higher cumulative level of tension than rapid chess games. A similar pattern is evident in artificial intelligence. Increasing the search depth of the Stockfish engine also leads to an increased level of strategic tension that it maintains throughout the match, indicating a close relationship between maintaining complexity and computational resources.
The outcome of the match is decided before the complexity reaches its peak.
Interestingly, the researchers found that the outcome of the match was often determined before the "strategic tension" reached its peak. Once one side has the upper hand, both humans and artificial intelligence tend to simplify the situation, but AI systems maintain a higher level of complexity compared to human players.
0 Comments