Researchers at Princeton University, Boston University, and other institutions have utilized machine learning to predict the strategic decisions of humans in various games, according to a report shared by alternative_right. Their paper, published in Nature Human Behavior, demonstrates that a deep neural network trained on human decisions can accurately predict the strategic choices of players. The study suggests that individuals behave more rationally and predictably in games they perceive as easier, while their decisions in complex games are influenced by numerous factors, increasing the unpredictability of their behavior.
As part of their future research, the team aims to explore what defines a game as ‘complex’ or ‘easy.’ This could be achieved by employing the context-dependent noise parameter integrated into their model as a measure of ‘perceived difficulty.’ Jian-Qiao Zhu, the first author of the study, noted that their analysis provides a robust model comparison across various decision-making models. By incorporating context-dependence into the quantal response model, they have significantly improved its ability to capture human strategic behavior. Zhu highlighted that key factors in the game matrix shape complexity, including efficiency considerations, computational difficulty of payoff differences, and the depth of reasoning required to reach a rational solution.
The findings also highlight the ‘lightness’ with which many people approach strategic decisions, potentially making them susceptible to manipulation by those seeking to influence their choices towards irrational decisions. Zhu and his colleagues hope to gain further insights into the factors that make games and decision-making scenarios more challenging for individuals. This understanding could lead to the development of new behavioral science interventions aimed at encouraging more rational decision-making processes.