This AI 'thinks' like a human - after training on 160 psychology studies
Briefly

Centaur is an AI system developed to predict human decisions in various situations, relying on a large language model fine-tuned with data from 160 psychology experiments involving 60,000 participants and over 10 million choices. Unlike traditional models that are task-specific, Centaur simulates behavior across multiple activities, including gambling and problem-solving. Its potential in cognitive science is significant, allowing researchers to conduct virtual experiments instead of involving actual human participants. This innovation addresses the challenges of building reliable theories in cognitive science, offering promising avenues for exploration.
The researchers fine-tuned a large language model using data from 160 psychology experiments, where 60,000 people made over 10 million choices across multiple tasks.
Centaur can simulate human behaviour across a range of tasks, including gambling, memory games, and problem-solving, outperforming classical theories used in psychology.
The system could become a valuable tool in cognitive science, allowing experimental sessions to be run in silico instead of on human participants.
Building theories in cognitive science is very difficult, and the aid of machines offers exciting possibilities for new discoveries.
Read at Nature
[
|
]