Researchers at Helmholtz Munich have developed an artificial intelligence model called Centaur, which simulates human behavior with remarkable accuracy. The language model was trained on over ten million decisions from psychological experiments and makes decisions in ways that closely resemble those of real people.
For decades, psychology has struggled to explain the complexity of human thought, with traditional models falling short in both transparency and predictability. However, Centaur combines both by predicting human behavior in familiar tasks and new situations it’s never encountered before. It identifies common decision-making strategies and adapts flexibly to changing contexts.
The model has numerous potential applications, including analyzing classic psychological experiments and simulating individual decision-making processes in clinical contexts. Its development opens up new perspectives in health research, particularly by helping understand how people with different psychological conditions make decisions. The dataset will be expanded to include demographic and psychological characteristics.
Centaur bridges the gap between theoretical models and predictive power, revealing where classical models fall short and providing insights into how they might be improved. To ensure responsible use, the researchers are committed to transparency and controllability, using open models that safeguard data sovereignty.
The team aims to investigate Centaur’s computational patterns and infer decision-making processes to deepen our understanding of human cognition. By combining AI research with psychological theory in a public research environment, they can pursue fundamental cognitive questions without industry pressures.
Source: https://techxplore.com/news/2025-07-centaur-ai.html