Researchers claim computational model can mimic human cognition

A team of researchers from several of the world’s leading universities say they have created a new computational model that can predict and simulate human behavior with a high degree of accuracy. 

The researchers claim that the new model can “simulate and predict human behavior in any domain,” which could help speed up experimentation for scientific discovery, model building, and experimental piloting.

According to an Oct. 28 research paper, the researchers’ model, dubbed Centaur, was trained on a data set called Psych101 — a database that captures an “unprecedented scale” of information concerning human behavior. 

Psych101 includes 160 psychological experiments with 60,092 participants, in which the human participants made over 10 million choices in total.

“While there have been previous attempts to instantiate such theories by building computational models, we currently do not have one model that captures the human mind in its entirety,” the researchers wrote.

“The human mind is remarkably general. Not only do we routinely make mundane decisions,  like choosing a breakfast cereal or selecting an outfit, but we also tackle complex challenges, such as figuring out how to cure cancer or explore outer space,” the researchers wrote.

“In contrast to this, most contemporary computational models – whether in machine learning or the cognitive sciences – are domain-specific. They are designed to excel at one particular problem.” 

In an X post, lead researcher Marcel Binz said they fine-tuned Centaur on Meta’s AI-based language model, Llama 3.1 70B, and found the resulting model “predicts the behavior of unseen participants better than existing cognitive models in almost every single experiment.”

Cognitive models seek to simulate the mental processes of humans and other animals, such as perception, reasoning, and memory.

Related: Meta is reportedly building its own AI-powered search engine

Binz said Centaur’s “internal representations” became more aligned with human neural activity after finetuning, even though it was not explicitly trained to “capture human neural activity.”

According to the paper, Centaur can operate in real time, exhibit rational adaptive behavior, and learn from its environment.

“We believe that Centaur is the first real candidate for a unified model of human cognition as envisioned by the great cognitive scientist Alan Newell,” Binz added.

Newell was a researcher specializing in computer science and cognitive psychology who worked for the RAND Corporation, an American global policy think tank and research institute.

Magazine: Most DePIN projects barely even use blockchain: True or false?