AI Brain Model Learns Like Animals, Unlocks New Discovery

A team of scientists just proved that understanding the brain doesn't always require more animal testing.

Researchers at Dartmouth College, MIT, and Stony Brook University built a computer model that mimics how real brains work, from individual neurons connecting to chemicals flowing between brain regions. When they asked it to perform the same visual learning task they'd given to lab animals, it learned the exact same way, with the same stumbles and breakthroughs.

The exciting part? The model was never trained on animal data. It was built entirely from scratch using biological principles.

"It's just producing new simulated plots of brain activity that then only afterward are being compared to the lab animals," says Richard Granger, a professor at Dartmouth and senior author of the study published in Nature Communications. "The fact that they match up as strikingly as they do is kind of shocking."

The model includes four brain regions working together: a cortex, brainstem, striatum, and a structure that adds helpful randomness through chemical bursts. As the model learned to categorize dot patterns, these regions communicated just like they do in living brains, even producing the same brain wave synchronization that MIT professor Earl K. Miller has observed in his animal research.

Then the model surprised everyone. It revealed a group of neurons whose activity predicted mistakes. When these "incongruent" neurons fired, the model made errors. The researchers went back to their animal data and found the same pattern hiding there all along.

The Ripple Effect

The discovery goes beyond just matching animal behavior. The team has founded Neuroblox.ai to turn these models into tools for developing brain treatments and therapies.

"The idea is to make a platform for biomimetic modeling of the brain so you can have a more efficient way of discovering, developing, and improving neurotherapeutics," says Miller, who is also with MIT's Picower Institute for Learning and Memory. Drug testing could happen earlier in the development process, before expensive clinical trials.

The model balances tiny details with big picture architecture, something researcher Anand Pathak describes as keeping both "the tree and the forest." Small circuits of neurons perform basic functions, while larger systems regulate learning and memory across regions.

As the model practiced its task, it gradually suppressed its own randomness and acted more consistently, just like animals gaining confidence in a new skill. Brain regions synchronized in rhythms that correlated with correct answers.

This breakthrough shows that computer models based on biological reality can both replicate natural learning and reveal hidden patterns that advance our understanding of how brains actually work.