They created a 100% functional human mind, it has AI and answers millions of questions like a human

In the race to develop artificial general intelligence (AGI), companies like OpenAI and Meta are investing billions of dollars in systems capable of replicating human thought.

Although there is still no single, accepted definition for AGI, the concept aims to build an artificial intelligence so versatile that it can tackle multiple tasks similarly to a person.

And while current AI has demonstrated amazing abilities—such as beating chess champions or predicting protein structures—it is still far from matching the human mind in terms of reasoning and general behavior.

However, an international team of scientists has developed a new system called Centaur, which could bring us a little closer to that goal. The project, led by cognitive scientist Marcel Binz of the Helmholtz Research Center in Munich, seeks to explore whether a language model similar to ChatGPT can behave like a human in psychological experiments.

The system, named after the mythological half-human, half-horse creature, was trained to replicate the decisions and behaviors of real volunteers in more than 160 psychological studies.

These experiments, compiled by Binz and his team, included tasks as diverse as memorizing words, playing slot machines, or piloting a virtual spaceship to find treasure.

Unlike traditional models, which typically focus on a single skill, Centaur was designed to generalize human behaviors across multiple contexts. “We wanted to see if it was possible to mimic different mental functions in a single model, without having to design a theory from scratch for each task,” Binz explained.

To achieve this, the researchers used the open-source LLaMA model created by Meta, which allows fine-tuning of its internal workings. They fed the system more than 10 million human responses and rewarded it whenever its choices aligned with those of real participants.

The result was a model that not only mimicked response patterns but also displayed human-like reasoning in new situations.

One of the most notable aspects of the experiment was Centaur's knowledge transfer capabilities. When presented with new versions of the games—such as replacing the spaceship with a flying carpet—the model retained the original search strategy, just as a person would.

Furthermore, when asked to solve logic problems not included in its training, Centaur replicated not only human successes but also their most common errors.

This performance sparked the interest of several experts. Russ Poldrack, a cognitive scientist at Stanford University, called the model "impressive" and asserted that it is the first to replicate so many tasks with such fidelity to human behavior. Another specialist, Ilia Sucholutsky of New York University, emphasized that Centaur significantly outperforms traditional cognitive models.

However, criticism also arose. Olivia Guest of Radboud University in the Netherlands questioned the fact that Centaur is not based on a concrete theory of mind. “Making predictions is not the same as understanding how the mind works,” she noted.

A similar opinion was expressed by Gary Lupyan of the University of Wisconsin-Madison, who asserted that the goal of cognitive science is not to replicate behaviors, but rather to understand the mechanisms that generate them.

Binz acknowledges these limitations. According to him, Centaur is not intended to be a definitive theory of human thought, but it can serve as a platform for creating new hypotheses and exploring how certain cognitive patterns emerge. His team is currently working on increasing the experimental database fivefold to further expand the model's capabilities.

As artificial intelligence advances by leaps and bounds, systems like Centaur open a fascinating door: using AI not just as a tool, but as a mirror to better understand the human mind. Although there is still a long way to go toward true AI, each step like this allows us to see more clearly what makes us human.