Scientists have created an artificial intelligence (AI) system called Centaur that can predict human behavior across various psychological experiments with unprecedented accuracy. The AI outperformed decades-old specialized models in every experiment, including those with new and modified scenarios.
Centaur was trained on data from over 60,000 people making more than 10 million decisions. This training allowed the AI to capture underlying patterns of human cognition, enabling it to predict behavior in complex challenges such as learning new skills or exploring uncharted territory.
The researchers’ goal was to create an AI that could truly understand human cognition. They assembled a dataset called Psych-101 containing 160 experiments covering memory tests, learning games, risk-taking scenarios, and moral dilemmas. The team used Meta’s Llama 3.1 language model as the foundation and fine-tuned it on the Psych-101 dataset using a parameter-efficient training technique.
Centaur’s success lies in its ability to generalize to new scenarios, structure changes, and entirely new domains such as logical reasoning. In open-loop simulations, the AI generated realistic human-like behavior patterns, achieving comparable performance to actual humans in exploration tasks.
A surprising discovery was that Centaur’s internal workings became more aligned with human brain activity, even though it was never explicitly trained to match neural data. This suggests that learning to predict human behavior forced the AI to develop internal representations that mirror how our brains process information.
The research team plans to expand their dataset to include more diverse domains and populations, aiming to create a comprehensive model that could serve as a unified theory of human cognition. They have made both their dataset and AI model publicly available for other researchers to build upon.
This breakthrough has significant implications for various fields such as marketing, education, mental health treatment, and product design. However, it also raises concerns about privacy and manipulation when our digital footprints reveal more about us than ever before.
Source: https://studyfinds.org/ai-thinks-like-humans-unprecedented-accuracy