A team of scientists has created an artificial intelligence system called Centaur that mimics human psychology by answering questions and behaving like a human in psychological experiments. The system was trained on 10 million responses from over 60,000 volunteers and is capable of predicting human behavior with surprising accuracy.
Cognitive scientists have long sought to understand the workings of the human mind, but existing theories only explain certain aspects of human cognition. To address this gap, researchers gathered a range of psychological experiments, including games, memory tests, and decision-making challenges. They then trained an open-source language model on these data to mimic human-like behavior.
The resulting Centaur system was tested with remarkable success, performing well in tasks that were designed to test its limits. It was able to generalize across different scenarios, transferring strategies from one game to another, and even demonstrating a level of logical reasoning comparable to humans.
While some experts hailed Centaur as “impressive,” others noted that the system’s limitations highlight the need for more comprehensive theories of human cognition. The researchers themselves acknowledge that their creation does not yet reveal new insights into the mind, but hopes it can serve as a benchmark for future research and help advance our understanding of human behavior.
As the team expands Centaur’s database by a factor of five and trains the system further, they predict it will be able to tackle even more complex tasks. With this potential breakthrough in artificial intelligence, researchers may finally have a tool that mirrors the complexities of the human mind.
Source: https://www.nytimes.com/2025/07/02/science/ai-psychology-mind.html