Researchers at Helmholtz Munich have developed a powerful artificial intelligence system that can predict human decisions with amazing precision. The AI model, called Centaur, has been trained on data from more than ten million decisions gathered through psychological studies.
This allows it to respond in ways that closely reflect how real people think and act. The breakthrough offers new hope for understanding human behavior and improving psychological theories that have struggled to explain and predict decisions in the past.
For many years, psychologists have tried to fully understand how people make choices. But most previous models could only do one thing well. They could either explain why people think the way they do or predict how they will act, but rarely both. Now, Centaur AI is changing that.
The research was led by Dr. Marcel Binz and Dr. Eric Schulz from the Institute for Human-Centered AI at Helmholtz Munich. They trained Centaur using a large dataset called Psych-101. This dataset includes over ten million decisions from 160 behavioral experiments. It is one of the most detailed collections of human behavior ever created.
Dr. Binz described the AI as a tool that creates a “virtual laboratory” where researchers can study human behavior in any situation written in natural language. He said, “We’ve created a tool that allows us to predict human behavior in any situation described in natural language – like a virtual laboratory.”
Centaur AI has an impressive ability to adapt. It can recognize patterns in decision-making and apply them to both familiar and completely new situations. It even predicts how quickly someone might make a choice, offering detailed insights into how people think and act.
This capability could have big benefits in areas like healthcare. For example, it could help researchers simulate how people with depression or anxiety disorders make decisions. This might lead to better treatments and support for patients.
Dr. Schulz, director of the institute, said they are only at the beginning of exploring Centaur’s possibilities. “We’re just getting started and already seeing enormous potential,” he said.
The team is ensuring that Centaur remains ethical and transparent. Notably, researchers built Centaur in a public research setting. As a result, the team can keep the system open and locally hosted, protecting user privacy and allowing independent checks.
Dr. Binz added, “Ensuring that such systems remain transparent and controllable is key.” “For example, by using open, locally hosted models that safeguard full data sovereignty.”
The next step for the researchers is to look deeper inside Centaur to see how it works. Specifically, they want to find out which computational patterns link to specific decision-making processes. Ultimately, this could help them understand how people process information and how decision-making differs between healthy individuals and those with mental health challenges.
Dr. Binz said, “These models have the potential to fundamentally deepen our understanding of human cognition – provided we use them responsibly.”
He also pointed out that it is no accident that Helmholtz Munich, rather than a private company, is doing the research. “We combine AI research with psychological theory – and with a clear ethical commitment,” said Dr. Binz. “In a public research environment, we have the freedom to pursue fundamental cognitive questions that are often not the focus in industry.”
The Psych-101 dataset is a unique collection of data. It includes decisions from more than 60,000 participants in experiments covering moral reasoning, risk-taking, and reward learning.
The researchers spent time carefully processing and standardizing the data so an AI language model could understand it. They plan to expand the dataset further by adding details about participants’ demographics and psychological traits. This research has been published in the journal Nature.
