Cognitive cartography of mammalian brains using meta-analysis of AI experts

The human brain is a vast, intricate landscape where billions of cells work together to create thought, feeling, and action. For decades, scientists have tried to map this terrain, linking specific brain regions to specific tasks like remembering a name or feeling fear. They have done this by reading thousands of research papers, each describing a small piece of the puzzle. However, the sheer volume of these studies has grown so large that no single human mind can read them all. The literature is scattered across different species, from humans to mice, and uses different methods, from brain scans to electrical stimulation. This fragmentation makes it difficult to see the full picture of how the brain works across the animal kingdom.

To solve this, a team of researchers has turned to artificial intelligence, not to replace human scientists, but to act as a tireless, super-reading assistant. They developed a system called NeuroNaut, which uses large language models—advanced computer programs trained on vast libraries of text—to read and summarize the entire body of neuroscience literature. Instead of manually combing through papers, the researchers asked these AI "experts" to estimate how likely it is that a specific brain region is involved in a specific function, based on everything the AI has learned from scientific texts. This approach allows them to build a comprehensive map of the brain's cognitive functions, covering humans, monkeys, and mice, all at once.

The researchers began by testing whether these AI experts could accurately describe the brain's functional landscape in humans. They asked the models to evaluate the role of 68 different brain regions in eight major areas of thought, such as emotion, memory, language, and vision. The AI generated detailed maps showing which parts of the brain are most associated with each task. When the researchers compared these AI-generated maps to the best existing maps created by traditional human-led reviews, they found a strong agreement. The AI correctly identified that language is primarily a left-brain function and that vision is centered in the back of the brain. This confirmed that the AI had successfully absorbed the core knowledge of the field.

However, the true power of this method emerged when the researchers looked beyond what traditional tools could see. Standard brain maps are often limited to data from brain scans, which only show activity while a person is performing a task. The AI, having read the entire history of neuroscience, could also draw on evidence from brain lesions and direct electrical stimulation. When the researchers compared the AI maps to circuits derived from patients who had brain stimulation or damage, the AI maps matched these causal evidence much better than the traditional scan-based maps did. This suggests the AI can synthesize a deeper, more causal understanding of how brain regions drive behavior, going beyond simple correlations found in imaging studies.

The most significant breakthrough came when the team applied this method to other species. Traditional brain mapping tools are almost exclusively designed for humans because they rely on specific coordinate systems that do not exist for other animals. By using the AI's ability to read text rather than coordinates, the researchers successfully generated cognitive maps for macaques and mice. They found that the brain organization in these animals mirrored the human brain in striking ways. For instance, the AI correctly identified that the visual cortex in a mouse is linked to sight, and the motor areas are linked to movement, just as in humans. Crucially, the AI also recognized that "language" is not a function associated with the brains of mice or monkeys, assigning it near-zero importance, while still identifying the auditory regions that handle their own vocalizations.

To ensure these maps were not just random guesses, the researchers tested them against real biological data. They compared the AI's cognitive maps for mice against data recorded from thousands of individual neurons during decision-making tasks. The AI's predictions aligned closely with the actual electrical activity of the brain. Similarly, for monkeys, the AI's map of working memory matched a detailed, human-curated summary of monkey studies better than a human brain map projected onto a monkey brain did. This proved that the AI was not simply copying human data but was actually synthesizing species-specific knowledge from the literature.

The researchers also explored how the brain's structure supports its function. They asked the AI to describe the top functions of every brain region and then used those descriptions to calculate how similar different regions were to one another. They found that regions with similar functional profiles were also connected by similar patterns of brain waves and blood flow. This relationship held true across humans, monkeys, and mice, suggesting that the way the brain is wired to support cognition is an evolutionary constant. The AI even uncovered a hidden hierarchy in the brain, organizing regions from simple sensory areas to complex association areas, a pattern that matched physical measurements of the brain's layers.

Finally, the team used this approach to study brain disorders. They asked the AI to map the involvement of brain regions in diseases like depression, schizophrenia, and autism. These maps were then compared to genetic data and brain scans of patients. The AI maps showed a strong connection to the genetic risks associated with these diseases, more so than traditional brain scans did. Furthermore, the AI correctly grouped diseases that share symptoms, such as linking sleep disorders with psychiatric conditions, even though it was never explicitly told about these connections. This ability to find hidden patterns across different types of data suggests that the AI can help scientists understand the complex, multifaceted nature of mental illness.

This work demonstrates that artificial intelligence can serve as a powerful tool for scientific discovery, capable of digesting the entire corpus of human knowledge to reveal patterns that are too complex for any single researcher to see. By bridging the gap between species and methods, the AI has provided a unified view of the mammalian brain, showing that the molecular and functional architecture of cognition is deeply conserved across evolution. The researchers have made their tool, NeuroNaut, freely available to the public, allowing anyone to ask questions about the brain and receive answers synthesized from the collective wisdom of the scientific community. This approach does not replace human expertise but amplifies it, turning a scattered library of facts into a coherent, navigable map of the mind.