Our brains still beat AI at learning new tasks, and one clever trick explains why.
Despite the rapid strides in artificial intelligence in recent years, the human brain retains a notable edge when it comes to transferring skills and learning across different tasks. A new study offers insight into how this happens.
Conducted by researchers at Princeton University, the study didn’t rely on human participants. Instead, they studied rhesus macaques (Macaca mulatta), animals whose brain structure and function closely resemble ours. The monkeys performed tasks that involved identifying shapes and colors on a screen and directing their gaze in specific ways to indicate their choices. While the tasks were underway, researchers used brain imaging to look for overlapping patterns and shared regions of activity.
These scans revealed that the monkeys’ brains organized related tasks using distinct blocks of neurons—think of them as cognitive building blocks. Importantly, existing blocks could be repurposed and recombined for new tasks, demonstrating a neural flexibility that current AI models struggle to match.
“State-of-the-art AI models can achieve human, or even super-human, performance on single tasks,” notes Tim Buschman, a neuroscientist at Princeton. “But they have a hard time learning and performing across many different tasks.” The team’s interpretation is that the brain’s flexibility comes from its ability to reuse cognitive components across tasks, allowing it to assemble new task representations by snapping these blocks together.
The study’s video (embedded below) shows the monkeys learning three related tasks in sequence, continually applying what they learned from one task to the next. The researchers pinpointed these cognitive blocks primarily in the prefrontal cortex, a region tied to higher-order thinking like problem-solving, planning, and decision-making, and crucial for cognitive flexibility.
Another key finding is that when certain cognitive blocks aren’t needed, their activity diminishes. This suggests the brain can “store away” unused neural blocks to sharpen focus on the current task.
Buschman offers an analogy: a cognitive block resembles a function in a computer program. One neuron cluster might identify color, and its output can feed into another function that drives action. This modular organization enables the brain to complete a task by sequentially executing each component.
These insights help explain how monkeys—and perhaps humans—adapt to challenges they haven’t seen before by leveraging existing knowledge to address new problems. In contrast, current AI often struggles with such generalization.
Looking ahead, the researchers suggest that understanding neural block reuse could guide the development of more adaptable AI systems capable of handling a broader range of tasks. The findings might also inform treatments for neurological and psychiatric conditions where people have difficulty applying learned skills to new settings.
For now, the concept of cognitive building blocks reveals a fundamental reason our brains remain more flexible and adaptable than today’s AI, which is prone to a phenomenon known as catastrophic forgetting—where neural networks lose earlier skills as they learn new ones.
While switching between tasks isn’t always ideal for human brains, applying prior knowledge across tasks can serve as a powerful shortcut.
If these results hold up, we may see AI designed to reuse representations and computations across tasks, enabling faster adaptation to changing environments—either by learning the right task representation through feedback or by recalling it from long-term memory. The full study is published in Nature.
