A group of scientists from prestigious universities around the world have announced the creation of an innovative computational model that can accurately predict and simulate human behavior. The researchers assert that this groundbreaking model, known as Centaur, has the capability to simulate and predict human behavior across various domains. This advancement has the potential to accelerate scientific discovery, model development, and experimental piloting.
The researchers developed Centaur using a vast database called Psych101, which contains an extensive range of information regarding human behavior. Psych101 consists of 160 psychological experiments involving 60,092 participants, who collectively made over 10 million choices. The researchers emphasize that while previous attempts have been made to create computational models based on theories of human behavior, there has not yet been a model that fully captures the complexities of the human mind.
In a recent Xpost, lead researcher Marcel Binz explained that Centaur was fine-tuned using Meta’s AI-based language model, Llama 3.1 70B. The results were remarkable, as the model exhibited superior predictive capabilities compared to existing cognitive models in nearly every experiment. Cognitive models aim to simulate various mental processes, including perception, reasoning, and memory, in both humans and animals.
Binz further highlighted that Centaur’s “internal representations” became more aligned with human neural activity after fine-tuning, even though it was not explicitly trained for that purpose. The researchers revealed that Centaur can operate in real time, display rational adaptive behavior, and learn from its environment. Binz expressed the belief that Centaur is a promising candidate for a unified model of human cognition, a concept envisioned by renowned cognitive scientist Alan Newell.
Alan Newell, known for his expertise in computer science and cognitive psychology, was affiliated with the RAND Corporation, a prominent American think tank and research institute.