Today, artificial intelligence is capable of producing award-winning essays and assisting doctors in diagnosing diseases with remarkable accuracy. However, when it comes to true mental flexibility, the human brain still holds a clear advantage.

Mental Flexibility: The Human Brain’s Edge

Humans adapt to new situations and information with impressive ease. Whether it’s learning unfamiliar software, trying out a new recipe, or figuring out the rules of a new game, people often pick things up quickly, while AI typically struggles to adapt in real time and to learn “on the fly.”

A recent study by neuroscientists at Princeton University identified one of the key reasons for this difference. The human brain repeatedly uses the same “cognitive building blocks” in various situations, combining them to create new behavioral models.

“Modern AI models can reach human or even superhuman levels in specific tasks. But they find it difficult to learn and perform a wide range of different tasks,” notes Dr. Tim Buschman, senior author of the study and deputy director of the Princeton Neuroscience Institute. “We found that the brain is flexible because it can use components of thought in different tasks. By combining these ‘cognitive legos,’ the brain can create new solutions.”

The study was published on November 26 in the journal Nature.

Compositionality: Reusing Skills

If someone already knows how to tune a bicycle, learning to repair a motorcycle may seem easier. This ability to build new skills from simpler, familiar ones acquired in similar situations is called compositionality.

“If you already know how to bake bread, you can use those skills to bake a cake without having to start from scratch,” explains Sina Tafazoli, PhD, a postdoctoral researcher in Buschman’s lab at Princeton and lead author of the study. “You use existing skills—working with the oven, measuring ingredients, kneading dough—and combine them with new ones, like whipping batter or making icing, to create something entirely different.”

Until now, evidence of exactly how the brain supports such flexible, compositional thinking has been limited and sometimes contradictory.

The Experiment: Visual Categorization Tasks

To gain a clearer picture, Tafazoli trained two male rhesus macaques to perform three related tasks while simultaneously recording their brain activity.

Instead of real-world tasks like baking or bike repair, the animals were given three visual categorization challenges. On a screen, they saw a series of colored blobs resembling balloons. Their task was to decide whether a blob looked more like a rabbit or the letter “T” (shape categorization), or whether it appeared more red or more green (color categorization).

The tasks were more complex than they seemed: the blobs varied in how strongly they resembled certain features. Some images clearly looked like rabbits or were bright red, while others were ambiguous and required careful judgment to categorize.

To indicate their decision about shape or color, each monkey would look in one of four directions on the screen. In one version of the task, for example, looking left meant the animal thought the blob resembled a rabbit, while looking right meant it looked more like a “T.”

A key part of the experiment was that each task had its own specific rules but also shared important components with the other tasks. One of the color tasks and the shape task required the animals to look in the same directions to indicate their choice, while both color tasks asked the macaques to categorize color in the same way but look in different directions to signal their decision.

This design allowed the researchers to determine whether the brain uses the same neural patterns—or cognitive building blocks—when tasks share common features.

Prefrontal Cortex: The Hub of Reusable Cognitive Blocks

By analyzing patterns of brain activity, Tafazoli and Buschman discovered that the prefrontal cortex—a brain region responsible for higher-order thinking and decision-making—contains several recurring activity patterns. These patterns emerged when groups of neurons worked together to achieve a common goal, such as distinguishing colors.

Buschman called these patterns the brain’s “cognitive legos”—a set of building blocks that can be flexibly combined to form different behavioral models.

“I think of a cognitive block like a function in a computer program,” says Buschman. “One group of neurons might distinguish colors, and its output can be linked to another function that controls an action. This organization allows the brain to perform a task by sequentially executing each component.”

For example, in one of the color tasks, the brain combines a block that determines the color of an image with a block that controls eye movement in a specific direction. When the animal switches to a different task—say, identifying shape instead of color but using the same eye movements—the brain simply activates the shape-processing block along with the same eye-movement block.

This joint use of blocks was observed mainly in the prefrontal cortex and was less pronounced in other brain regions. This suggests that compositionality may be a distinctive feature of this area.

Attention Control: Turning Blocks On and Off

Tafazoli and Buschman also noticed that the prefrontal cortex can “mute” certain cognitive blocks when they are not needed. This likely helps the brain focus on the most relevant task at any given moment.

“The brain has limited resources for cognitive control,” Tafazoli notes. “It’s necessary to suppress some abilities to concentrate on those that are important right now. For example, focusing on shape categorization temporarily reduces the ability to encode color, because the goal is to distinguish shape, not color.”

By selectively activating and suppressing different blocks, the brain avoids overload and maintains concentration on the current task.

Cognitive Legos, Artificial Intelligence, and Mental Health

These cognitive legos may explain why people often master new tasks so quickly. The brain doesn’t always have to start from scratch—it can use existing mental components, combine them, and avoid duplicating work, something modern AI systems typically cannot do.

“The main problem in machine learning is catastrophic forgetting,” says Tafazoli. “When a machine or neural network learns something new, it forgets and overwrites previous memories. If an artificial neural network learns to bake a cake and then learns to bake cookies, it will forget how to bake the cake.”

Incorporating compositionality into AI could eventually make artificial systems more human-like in their learning, allowing them to acquire new skills without erasing old ones.

The same principles could also impact medicine. Many neurological and psychiatric conditions—including schizophrenia, obsessive-compulsive disorder, and certain types of brain injury—make it difficult to apply existing skills in new situations. Such problems can arise when the brain can no longer smoothly recombine its cognitive building blocks.

“Imagine being able to help people regain the ability to switch strategies, develop new habits, or adapt to change,” says Tafazoli. “In the long run, understanding how the brain reuses and recombines knowledge could help us develop therapies that restore this process.”