A pair of monkeys staring at colored shapes in a Princeton lab may be getting closer to understanding how their own minds work. New research shows that the brain solves difficult problems by reusing simple brain parts across many tasks, much like assembling pieces of a toy set. This research helps explain why you can go from making dinner to learning new software without having to start from scratch every time.
Scientists have long puzzled about how the brain links small actions into more complex ones. Learn when fruits ripen and apply those skills when shopping, cooking, and choosing meals. Your brain doesn’t rewire each skill every time. Reuse what you already know and combine those skills in new ways.
In this study, researchers at Princeton University trained two male rhesus macaques to play three related visual games. Each trial presented a crazy image that changed color and shape. The animals had to judge either the shape or the color and signal their answer by rapidly moving their eyes to one of four targets. I heard the voice just by glancing at the screen.
Two tasks focused on color and one on shape. The secret lies in how the answers are reported. The color game shared its gaze direction with the shape game. For the other color games, a different set of gaze directions was used. The monkeys were never told which game was being run. They had to infer the rules from feedback and adjust when the rules changed without warning.
Both monkeys did well. They achieved 80% accuracy in two games and over 90% in the third game. They didn’t fall apart when the rules were reversed. They tested their guesses, learned quickly, and settled into new patterns. The most difficult switch occurred when the rule changed from shape to color, even though the same eye direction was used. In this case, initial errors were common, but accuracy improved within about 75 trials.
Inside the learning brain
To see what the brain was doing at similar moments, the researchers recorded the activity of more than 1,000 neurons across five regions of the brain at once. These included areas that guide vision, movement, and decision making.
A pattern quickly emerged. A traffic light was displayed throughout the area indicating the current task. Color and shape signals followed the image. The eye movement plan was then displayed, followed by news about the reward.
Many neurons played a dual role. A single cell may respond to both the color and the task itself. This combination of roles packed more meaning into each cell and helped keep the brain flexible.
The biggest surprise appeared in the prefrontal cortex, an area near the forehead associated with planning and control. There, the same neural patterns were used to represent colors in both color games, even though the eye movements were different. When the team trained a computer to read colors from brain activity in one game, it was able to read colors in other games as well.
The same reuse occurred with eye movements. The code learned from the shape game worked in the linked color game. It was as if the brain kept a shared library of parts and pulled out the right parts for the job.
Working “cognitive Lego”
The results of this study were published in the journal Nature on November 26, and attracted attention for their implications regarding intelligence.
“Cutting-edge AI models can reach human, even superhuman, performance on individual tasks, but they have difficulty learning and performing many different tasks,” said Tim Bushman, Ph.D., senior author of the study and associate director of the Princeton Neuroscience Institute. “We found that the brain is flexible because it can reuse cognitive building blocks for different tasks. By combining these ‘cognitive Legos’, the brain can construct new tasks.”
Lead author Dr. Sina Tafazoli provided a clear example. “If you already know how to bake bread, you can use this ability to bake cakes without having to learn how to bake from scratch,” he said. “Reuse existing skills, such as using an oven, measuring ingredients, and kneading dough, and combine them with new skills, such as whipping dough or making frosting, to create something completely different.”
For the monkeys, that meant pairing blocks that judged colors with blocks that moved their eyes in specific ways. When the game changed, the brain swapped one block for another and continued.
The researchers also observed that the brain faded out parts that were not needed. When color became important, color signals became stronger and shapes faded. When the rules were reversed, the volume knob was rotated in the opposite direction. “The brain’s ability to control cognition is limited,” Tafazoli says. “I need to compress some of my abilities so I can focus on what’s important right now.”
Why does the brain outperform machines in terms of flexibility?
If you’ve ever seen a program learn another skill and then forget it, you’ve seen a problem called catastrophic interference. Tafazoli said: “When a machine or neural network learns something new, it forgets and overwrites its previous memories. An artificial neural network might know how to bake a cake, but the next time it learns how to bake cookies, it will forget how to bake a cake.”
The brain handles this better by keeping the parts separate and recombining them when necessary. Its design may point the way to smarter machines.
The study also suggests medical benefits. Some people with brain injuries or mental illnesses have difficulty changing rules or applying skills in new environments. If the problem lies in a disconnected connection between these mental parts, therapy may aim to restore that network.
“Imagine being able to help people regain the ability to change strategies, learn new habits, and adapt to change,” Tafazoli said. “In the long term, understanding how the brain reuses and recombines knowledge may help us design treatments that restore that process.”
So far, simple experimental work has revealed deep truths. Intelligence grows from reuse. Your brain builds new things from old things, one piece at a time.
Practical implications of the research
This effort aims for AI systems that learn faster by reusing core skills, rather than rebuilding them for each task. This could allow the software to learn on the fly without forgetting previous lessons.
In medicine, the findings could help guide treatments for people who struggle to switch rules after brain injury or in conditions such as schizophrenia and obsessive-compulsive disorder. Physicians may be able to help patients regain flexibility in their daily lives by targeting how the mental part is recycled.
The study also provides a tool to track learning by reading brain signals associated with task beliefs, which could improve training methods in schools and clinics.
The research results are available online in the journal Nature.
