The human brain remains superior to computers in its ability to transfer skills and learn across tasks, despite the advancements in artificial intelligence. A recent study conducted by researchers from Princeton University sheds light on how this process likely occurs.
Instead of testing humans, the study utilized rhesus macaques, animals closely related to humans in terms of biology and brain function. The monkeys were tasked with identifying shapes and colors on a screen and directing their gaze in specific directions to provide answers. Brain scans were used to identify overlapping patterns and shared areas of brain activity during these tasks.
The researchers discovered that the monkey brains utilized different blocks of neurons, which they referred to as ‘cognitive Legos’, across tasks. This neural flexibility allows for the repurposing and recombination of existing blocks for new tasks, a capability that surpasses even the most advanced AI models.
According to neuroscientist Tim Buschman, the brain’s ability to reuse components of cognition in various tasks sets it apart from AI models, which struggle with learning and performing multiple tasks. The study showcased how the brain can adapt to new challenges by utilizing existing knowledge, a feat that current AI technology struggles to replicate.
The cognitive Lego blocks identified by the researchers were concentrated in the brain’s prefrontal cortex, a region associated with higher cognitive functions such as problem-solving and decision-making. The researchers also observed that when certain cognitive blocks were not needed, activity in those areas decreased, suggesting that the brain can prioritize relevant neural processes for the task at hand.
The findings of this study could have implications for training AI models to be more adaptable and for developing treatments for neurological and psychiatric disorders. By understanding how the brain reuses representations and computations across tasks, researchers hope to improve AI adaptability and address issues like catastrophic forgetting, where neural networks struggle to learn consecutive tasks without forgetting previous ones.
While task-switching may not be optimal for the brain, the ability to apply knowledge from one task to another can be a valuable shortcut. The researchers emphasize the brain’s capacity to rapidly adapt to changes in the environment by learning task representations through feedback or recalling them from long-term memory.
The study, published in Nature, highlights the remarkable flexibility and adaptability of the human brain compared to AI models. By uncovering the mechanisms behind cognitive flexibility, researchers aim to enhance AI capabilities and improve our understanding of brain function.
