Super AI thinks and makes decisions surprisingly like humans

Researchers at Helmholtz Munich have created an advanced artificial intelligence system capable of simulating human decision-making with impressive accuracy.

The model, called Centaur, was trained using data from more than ten million decisions collected through psychological studies, allowing it to generate responses that realistically reflect human behavior.

This breakthrough opens up new possibilities for deeper understanding of how humans think and refining existing psychological frameworks.

The Centaur AI meta-language model was trained on more than ten million decisions from psychology experiments. Photo: Helmholtz Munich

For many years, the field of psychology has struggled to fully grasp the complexities of human thinking. However, previous models have often been limited to explaining how people think or predicting how they act, rarely achieving both goals.

Led by Dr. Marcel Binz and Dr. Eric Schulz from the Institute for Human-Centered AI at Helmholtz Munich, the team has now presented a model that bridges this gap. Centaur was trained on a comprehensive dataset called Psych-101, which aggregates over ten million decisions from 160 different behavioral experiments.

Centaur stands out for its ability to predict human responses not only in familiar contexts but also in completely new situations. It recognizes recurring decision-making patterns, adapts easily to new environments, and can even estimate reaction times with incredible detail.

“We have created a tool that allows us to predict human behavior in any situation described in natural language – like a virtual laboratory,” said Marcel Binz, lead author of the study.

Potential applications range from analyzing classic psychological experiments to simulating individual decision-making processes in clinical settings—for example, in cases of depression or anxiety disorders.

This model opens up new perspectives in health research in particular – for example, by helping us understand how people with different mental health conditions make decisions. The dataset will be expanded to include demographic and psychological characteristics.

The Bridge Between Theory and Prediction

Centaur bridges two previously separate fields: interpretability theory and prediction. It can reveal the weaknesses of classical models – and provide insights into how they can be improved. This opens up new possibilities for research and practical applications, from medicine to environmental and social sciences .

“We’re just getting started and we already see huge potential,” said institute director Eric Schulz. Ensuring such systems remain transparent and auditable is key, Binz added — for example, using open, locally hosted models that preserve full data sovereignty .

Next, the researchers wanted to take a closer look inside Centaur: Which computational models correspond to specific decision-making processes? Could they be used to infer how humans process information—or how decision-making strategies differ between healthy people and those with mental illness?

Researchers believe these models have the potential to deepen our understanding of human cognition — as long as we use them responsibly.

The fact that this research was conducted at Helmholtz Munich and not in the development departments of large tech companies is no coincidence.

“We combine AI research with psychological theory—and with a clear ethical commitment,” Binz says. “In a public research environment, we have the freedom to pursue fundamental cognitive questions that are not often focused on by industry.”