Ai that helps to explain what we think like we do and explain what we think

Researchers at Helmholtz Munich have developed an artificial intelligence model that can simulate human behavior with remarkable accuracy. A linguistic model called Centaur is trained on over 10 million decisions from psychological experiments and makes decisions in a way that resembles those of real people. This opens up new pathways for understanding human cognition and improving psychological theories.

For decades, psychology has aimed to explain the complete complexity of human thought. However, traditional models can provide a transparent explanation of how people think, or can certainly predict how they will behave. Achieving both seemed out of reach for a long time.

The team led by Dr. Marcel Binz, PhD and Dr. Marcel Binz and Eric Schulz, researchers at Helmholtz Munich’s Human-Centered AI Institute, developed a model that combines both. Centaur was trained using a specially curated dataset called Psych-101, which contained over 10 million personal decisions from 160 behavioral experiments. This study has been published in the journal Nature.

What makes Centaur unique is its ability to predict human behavior not only in familiar tasks but also in completely new situations that have never been encountered before. Identify common decision-making strategies, adapt flexibly to context changes, and predict reaction times with incredible accuracy.

“We created tools that can predict human behavior in any situation explained in natural language, like a virtual laboratory,” says Binz, the research’s lead author.

Potential applications range from analyzing classical psychological experiments to simulating individual decision-making processes, such as depression and anxiety disorders, such as clinical contexts. This model opens up new perspectives, particularly in health research. For example, it helps you understand how people with different psychological states make decisions. The dataset is set up to be extended to include demographic and psychological characteristics.

Centaur: The Bridge between Theory and Prediction

Centaur previously bridges two separate domains. Interpretable theory and predictive power. It uncovers where classical models are lacking and provides insight into how they can be improved. This opens up new possibilities for research and real-world applications, from medicine to environmental science and social sciences.

“We’re just starting out and we’re already seeing great potential,” says Schultz, director of the institute. It is important to ensure that such systems remain transparent and controllable, for example, by using an open localhost model that protects full data sovereignty.

Second, researchers aim to take a closer look inside Centaur. Which calculation patterns correspond to a particular decision-making process? Can they be used to infer how people process information or how decision strategies differ between healthy individuals and people with mental health?

The researchers stated, “These models have the potential to fundamentally deepen our understanding of human cognition. We use them responsibly.” It is no coincidence that this research is being conducted in Helmholtz Munich, not in the development department of major high-tech companies.

“We combine AI research with psychological theory and with clear ethical commitment,” says Binz. “In a public research environment, there is freedom to pursue basic cognitive questions that are often not the focus of industry.”