Even though artificial intelligence has made significant progress lately, the human brain still outperforms computers when it comes to transferring skills and adapting across different tasks. A recent study sheds some light on how we manage this.
Conducted by a research team from Princeton University, the study didn’t involve human subjects but instead looked at rhesus macaques, which are biologically close to us.
The monkeys participated in identifying various shapes and colors on a screen and turned their heads in specific directions to respond. During this process, the researchers monitored brain activity to find overlapping patterns in the monkeys’ neural responses.
Scanning results revealed that the monkeys were using varying clusters of neurons – referred to as ‘cognitive Legos’ by the researchers – to tackle different tasks. It turns out that these neural blocks can be reused and reconfigured for new challenges, showcasing a level of neural flexibility that, surprisingly, outpaces even the best AI models.
“While top AI systems can achieve human or even superior performance on single tasks,” notes neuroscientist Tim Buschman, “they face challenges when it comes to learning and executing multiple tasks.”
“Our findings show that the brain’s flexibility comes from its ability to recycle cognitive components for various tasks. By assembling these ‘cognitive Legos,’ the brain can create new task solutions,” Buschman elaborates.
As demonstrated in an accompanying video, the monkeys were required to distinguish shapes and colors across three related trials, allowing them to continuously learn and apply their knowledge from one task to the next.
The cognitive blocks identified by the researchers were particularly concentrated in the prefrontal cortex, a region associated with advanced cognitive functions like problem-solving and decision-making. This area appears vital for cognitive flexibility.
Additionally, the study found that when certain cognitive blocks were unnecessary, their activity decreased, suggesting that the brain can effectively prioritize and set aside certain neural resources to focus better on immediate tasks.
“I liken a cognitive block to a function in software,” shares Buschman.
“One group of neurons might be tasked with distinguishing colors, which can then be linked to another function that triggers an action. This structure allows the brain to complete a task by systematically executing each of its components.”
This perspective helps explain how both monkeys and potentially humans can adapt to new challenges and utilize preexisting knowledge to address them—something that current artificial intelligence falls short in.
The researchers speculate that their discoveries could not only enhance AI’s adaptability to new tasks but might also influence treatment approaches for neurological and psychiatric disorders where individuals struggle to transfer skills to different contexts.
These cognitive Legos, at their core, underscore how our brains possess a level of flexibility and adaptability that most AI models cannot match, particularly as AI often suffers from issues like catastrophic forgetting, meaning that they can’t easily retain skills across tasks.
And while switching tasks isn’t the easiest on our brains, leveraging knowledge from one task to help with another can serve as a beneficial shortcut.
“If our findings are accurate, then the brain’s ability to reuse representations and computations across tasks could enable rapid adaptation to environmental changes, either through learning new task representations via reward feedback or retracing them from long-term memory,” the researchers conclude.
This research has been published in Nature.
