The Human Mind in Software. It Is Alive!
August 11, 2025
Has this team of researchers found LLM’s holy grail? Science magazine reports, “Researchers Claim their AI Model Simulates the Human Mind. Others are Skeptical.” The team’s paper, published in Nature, claims the model can both predict and simulate human behavior. Predict is believable. Simulate? That is a much higher bar.
The team started by carefully assembling data from 160 previously published psychology experiments. Writer Cathleen O’Grady tells us:
“The researchers then trained Llama, an LLM produced by Meta, by feeding it the information about the decisions participants faced in each experiment, and the choices they made. They called the resulting model ‘Centaur’—the closest mythical beast they could find to something half-llama, half-human, [researcher Marcel] Binz says.”
Cute. The data collection represents a total of over 60,000 participants who made over 10 million choices. That sounds like a lot. But, as computational cognitive scientist Federico Adolfi notes, 160 experiments is but “a grain of sand in the infinite pool of cognition.” See the write-up for the study’s methodology. The paper claims Centaur’s choices closely aligned with those of human subjects. This means, researchers assert, Centaur could be used to develop experiments before involving human subjects. Hmm, this sounds vaguely familiar.
Other cognitive scientists remain unconvinced. For example:
“Jeffrey Bowers, a cognitive scientist at the University of Bristol, thinks the model is ‘absurd.’ He and his colleagues tested Centaur … and found decidedly un-humanlike behavior. In tests of short-term memory, it could recall up to 256 digits, whereas humans can commonly remember approximately seven. In a test of reaction time, the model could be prompted to respond in ‘superhuman’ times of 1 millisecond, Bowers says. This means the model can’t be trusted to generalize beyond its training data, he concludes.
More important, Bowers says, is that Centaur can’t explain anything about human cognition. Much like an analog and digital clock can agree on the time but have vastly different internal processes, Centaur can give humanlike outputs but relies on mechanisms that are nothing like those of a human mind, he says.”
Indeed. Still, even if the central assertion turns out to be malarky, there may be value in this research. Both vision scientist Rachel Heaton and computational visual neuroscientist Katherine Storrs are enthusiastic about the dataset itself. Heaton is also eager to learn how, exactly, Centaur derives its answers. Storrs emphasizes a lot of work has gone into the dataset and the model, and is optimistic that work will prove valuable in the end. Even if Centaur turns out to be less human and more Llama.
Cynthia Murrell, August 11, 2025
Comments
Got something to say?
-
- Subscribe to Beyond Search
Feature archive
News archive
-
