Notwithstanding the swift advancements in artificial intelligence witnessed in recent years, the intricate human brain maintains an advantage over computational systems in its capacity for skill transference and cross-task learning. A recent scientific investigation sheds light on the probable mechanisms underlying this capability.
The research, spearheaded by a consortium from Princeton University, did not involve direct experimentation with human subjects. Instead, the scientists utilized animal models biologically and neurologically proximate to humans: rhesus macaques, scientifically designated as Macaca mulatta.
These primates were tasked with discerning visual patterns and hues presented on a display, employing specific directional movements to indicate their responses. Concurrently, neurological imaging techniques were employed to detect congruent activation patterns and shared neuronal territories within the animals’ brains.
The resultant scans demonstrated the primate brains employing distinct neuronal clusters—termed ‘cognitive Legos’ by the investigators—across varied assignments. These pre-existing modules, it appears, can be recontextualized and integrated into novel undertakings, revealing a degree of neural plasticity that surpasses even the most sophisticated AI architectures.
“Currently, cutting-edge AI models can achieve human-level, or even superior, proficiency within singular tasks,” states neuroscientist Tim Buschman of Princeton University. “However, their aptitude for learning and executing a broad spectrum of distinct activities remains constrained.”
“Our findings indicate that the brain’s adaptability stems from its ability to repurpose cognitive components across a multitude of assignments. By assembling these ‘cognitive Legos,’ the brain constructs new functional pathways.”
As illustrated in the accompanying video, the subjects were required to differentiate between shapes and colors in three discrete yet interconnected scenarios. These tasks necessitated continuous learning and the application of acquired knowledge from one context to the next.
The identified ‘cognitive Lego’ modules were predominantly localized within the brain’s prefrontal cortex. This region is intrinsically linked to advanced cognitive functions, including problem-solving, strategic planning, and decision-making, and appears to play a pivotal role in cognitive flexibility.
Further observations by the research team revealed a reduction in neural activity within specific cognitive modules that were not actively engaged. This suggests a mechanism whereby the brain conserves neuronal resources, effectively shelving ‘neural Legos’ not immediately required, thereby optimizing focus on the current objective.
“I conceptualize a cognitive module as analogous to a function within a computer program,” remarks Buschman. “A particular set of neurons might be responsible for color differentiation, with its output seamlessly routed to another function governing motor action. This organizational principle empowers the brain to execute a task by sequentially performing each constituent element.”
This mechanism provides an explanation for how primates, and by extension humans, can adeptly navigate unforeseen challenges and unfamiliar tasks by leveraging existing knowledge—a feat that current artificial intelligence capabilities struggle to emulate.
Looking ahead, the researchers propose that their discoveries could inform the development of AIs exhibiting enhanced adaptability to novel situations. Moreover, this research may prove instrumental in devising therapeutic interventions for neurological and psychiatric conditions characterized by difficulties in transferring learned skills to new environments.
For the present, these ‘cognitive Legos’ provide a fundamental insight into the superior flexibility and adaptability of our brains compared to AI models, which are susceptible to a phenomenon known as catastrophic forgetting. This vulnerability means that neural networks often fail at learning sequential tasks without losing proficiency in previously acquired ones.
While the mental exertion of task-switching is not entirely inconsequential, the ability to apply knowledge from one domain to another can serve as an invaluable cognitive shortcut.
“If, as our findings suggest, the brain can effectively reuse representations and computational processes across tasks, this capability would facilitate rapid adaptation to environmental changes. This adaptation can occur either by acquiring appropriate task representations through reward-based feedback or by retrieving them from long-term memory,” the researchers conclude.
