Our Brains Can Still Outsmart AI Using One Clever Trick

Despite the impressive progress made in artificial intelligence, our brains retain a distinct advantage: the ability to flexibly learn and transfer knowledge between different tasks. New research sheds light on how we achieve this cognitive feat.

Researchers at Princeton University explored this phenomenon by studying rhesus macaques, whose brains share key similarities with our own. The monkeys were trained to identify shapes and colors on a screen and indicate directions. Throughout these tasks, brain scans monitored neural activity, revealing how the brain tackles diverse challenges.

The scans revealed that the monkeys' brains employed distinct groups of neurons – described by the team as 'cognitive Legos' – that could be repurposed across various tasks. This remarkable adaptability allows the brain to combine and reuse existing neural circuits for new challenges, a capability that current AI models struggle to replicate.

According to neuroscientist Tim Buschman from Princeton University, while AI excels at individual tasks, even surpassing human performance in some cases, it falters when faced with learning and performing multiple, varied tasks. The brain's flexibility, stemming from its ability to reuse cognitive components, enables it to efficiently adapt to new demands.

The cognitive Lego blocks identified in the study were primarily located in the prefrontal cortex, a brain region associated with higher-level cognitive functions like problem-solving, planning, and decision-making. This area appears crucial for cognitive flexibility.

Furthermore, the researchers observed that when specific cognitive blocks were not required for a task, their activity decreased. This suggests the brain can effectively store away unused neural circuits, optimizing focus on the immediate task.

Buschman likens these cognitive blocks to functions in computer programs. One set of neurons might discern color, with its output directing another function to trigger an action. This organization enables the brain to execute a task by sequentially performing each of its components.

This explains how humans and monkeys can adapt to unfamiliar challenges by leveraging existing knowledge, a feat that remains difficult for current AI systems. The findings suggest potential avenues for training AI to become more adaptable. Moreover, this research could contribute to developing treatments for neurological and psychiatric conditions where individuals struggle to apply learned skills in new contexts.

Ultimately, the discovery of these cognitive Legos highlights the fundamental difference in adaptability between our brains and AI models. AI often suffers from 'catastrophic forgetting,' where learning new tasks erases previously acquired knowledge. While multitasking might not be ideal, the brain's ability to transfer knowledge between tasks provides a valuable shortcut.

The researchers concluded that the brain's capacity to reuse representations and computations across tasks allows for rapid adaptation to environmental changes, whether through learning via reward feedback or recalling information from long-term memory.