AI keeps chess complexity alive longer than humans

Each move on a chessboard closes some paths and opens others. To see how humans and machines handle that shifting web, researchers from the Complexity Science Hub and the University of Parma analyzed more than 2,000 human-versus-human and AI-versus-AI games. Their finding: chess engines—software that calculates candidate moves—do not simply play more strongly; they keep more strategic tension alive.

The team defined strategic tension as the number of possibilities still open after each move. It is low at the start, when the pieces are largely disconnected. As play develops, every move reshapes the network of possible responses. Humans and AI both build tension through the opening and early middlegame. Then grandmasters generally begin exchanging pieces and simplifying once they move beyond the deeply prepared opening lines that top players study before a game.

Stockfish and Leela Chess Zero take a different route. The engines maintain highly interconnected positions for many more moves, keeping complexity spread across the board rather than releasing it through exchanges. The strongest human players briefly reach a higher peak of tension than engines, but they do not sustain it as long. The study, published in APS Open Science, found that stronger humans preserve complexity longer than weaker ones, while classical games show higher cumulative tension than rapid or blitz games.

The same relationship appears inside the machine. Increasing Stockfish's search depth increases the strategic tension it maintains, linking sustained complexity to computational resources. The researchers also found that a game's eventual outcome is often settled before tension reaches its maximum; once one side has a lasting advantage, both humans and AI tend to simplify, although AI still carries considerably more complexity into that phase.

So what does this change in practice? It gives researchers a measurable way to study how an artificial system handles a large set of interacting choices, rather than judging it only by whether it wins. That could inform work on diplomacy, military planning and financial markets, where decisions also alter future options. But chess is controlled and predictable; the authors caution that direct comparisons with real-world decisions require care, and the consequences of introducing a complexity-embracing player remain unknown.