The key mechanism of the human brain being superior to artificial intelligence is revealed: it has the ability to reorganize cognitive modules to "put together building blocks"

In a new study, neuroscientists at Princeton University in the United States have revealed a key mechanism by which the brain maintains an advantage in the face of artificial intelligence (AI): the brain will reuse the same cognitive "modules" in different tasks, and by combining and reorganizing these modules, the brain can quickly construct new behavior patterns like "building blocks". The study was published in the latest issue of the journal Nature.

The most advanced AI models can reach or even surpass humans on a single task, but they do not perform well when it learns and performs many different tasks. This is because the human brain still maintains an advantage in one key area: cognitive flexibility. For example, humans can adapt to new information or unfamiliar challenges relatively easily, such as learning new software, cooking according to recipes, or quickly picking up a new game, while AI struggles to achieve such "instant" learning.

Studies have found that the brain is flexible because it can reuse cognitive components, like building blocks, and splicing these "cognitive building blocks" together. For example, people who can repair bicycles will call on relevant skills when learning to repair motorcycles, and they will be more handy. But there is still a lack of consistent and sufficient evidence in the scientific community on how the brain achieves this cognitive flexibility.

The team trained two male rhesus monkeys to complete three related classification tasks and recorded their brain activity in the process. These tasks are similar to determining whether a blurry image on the screen is more like a rabbit or the letter "T", or closer to red or green. The blur of the image varies, sometimes noticeable, sometimes very subtle. Monkeys express their judgment by looking in different directions. The key to experimental design is that each task, although different, shares certain cognitive elements that allow it to test how the brain reuses the same patterns of neural activity.

After analyzing multiple brain regions, the team found that the prefrontal cortex (the brain region responsible for advanced cognitive functions) contains several reusable activity patterns that correspond to different "cognitive building blocks" that the brain flexibly splices together to construct new behaviors.

In addition, the prefrontal cortex inhibits its activity when certain cognitive modules are not needed, helping the brain focus on the task at hand. This is because cognitive resources are limited, and the brain ensures that the main target task is not disturbed.

According to the team, this combination-based learning method may be the key to humans being able to efficiently learn new skills without forgetting old ones. In contrast, current AI systems often encounter the problem of "catastrophic interference" - after learning new tasks, they often overwrite previous memories. If the modular and reconfigurable mechanism of the human brain is introduced into AI, it may be possible to develop an intelligent system that can continue to learn without forgetting. At the same time, this finding also has clinical implications for understanding certain mental illnesses and brain damage.

【Editor-in-Chief】

Many characteristics of the human brain make today's AI beyond the reach of today's AI. If you have children around you, you will find that infants and toddlers as young as two or three years old can also build toys with graphic instructions, or learn to play with mobile phones without a teacher. The study found that the prefrontal cortex of our brain flexibly combines existing standardized "cognitive modules" to quickly respond to new challenges like playing with building blocks. If this mechanism can be integrated into AI design, it may be possible to develop intelligent systems that can continuously accumulate experience, taking an important step towards general artificial intelligence. Research also suggests that some mental illnesses may also be related to impaired cognitive module reorganization mechanisms.