A major new study on the structure of human intelligence suggests that current approaches to artificial intelligence may be hitting a fundamental ceiling, and that true “general intelligence” in machines will require a radical architectural rethink.
Researchers at the University of Notre Dame have found that human intelligence does not reside in specific brain regions or “processors,” but rather emerges from how the entire brain coordinates diverse networks to function as a single, unified system.
The findings, published in Nature Communications, challenge the “bigger is better” ethos often seen in AI development, where specialised capabilities are simply scaled up. Instead, the study argues that flexibility — the hallmark of human thought — arises from a global organisational structure that current AI models largely lack.
“Many AI systems can perform specific tasks very well, but they still struggle to apply what they know across different situations,” says Aron Barbey, Professor of Psychology and director of the Notre Dame Human Neuroimaging Center. “Human intelligence is defined by this flexibility — and it reflects the unique organisation of the human brain.”
The limits of specialisation
Modern neuroscience, like modern AI, has often viewed the mind as a set of specialised modules — distinct areas for memory, language, and perception. However, this modular view fails to explain “general intelligence,” or the ability to adapt to novel problems.
By analysing brain scans from nearly 1,000 adults, Barbey and his team confirmed the “Network Neuroscience Theory.” They found that high intelligence isn’t about having a stronger processor for a specific task; it’s about having a brain architecture that allows for “system-wide coordination.”
This involves “regulatory hubs” that act as traffic controllers, dynamically reconfiguring neural networks to switch between careful deliberation and rapid intuition depending on the context.
Biologically inspired AI
The implications for Silicon Valley are significant. If human intelligence is a product of network organisation rather than raw processing power, achieving Artificial General Intelligence (AGI) may require engineers to stop building better modules and start building better “brains.”
Barbey suggests that the path forward lies in “human-centred, biologically inspired artificial intelligence” that mimics the brain’s balance of local specialisation and global integration.
“If general intelligence in humans arises from system-level organisation rather than from a dedicated general-purpose mechanism, then achieving general intelligence in artificial systems may require more than the accumulation or scaling of specialised capabilities,” Barbey notes.
The architecture of thought
The study identified specific structural features that enable this high-level coordination, including long-range connections that serve as “shortcuts” linking distant brain regions.
Lead author Ramsey Wilcox explains that this global architecture sets the boundaries of what a mind can do.
“This coordination does not carry out cognition itself, but determines the range of cognitive operations the system can support,” Wilcox says. “Once the question shifts from where intelligence is to how the system is organized, the empirical targets change.”
