New Delhi: A recent study from Columbia University’s School of Engineering reveals that the brain’s visual system plays a far more dynamic role in decision-making than previously understood offering insights that could help shape the future of adaptive artificial intelligence.
Published in Nature Communications, the study was led by biomedical engineer and neuroscientist Dr. Nuttida Rungratsameetaweemana. Her team found that even early-stage visual areas of the brain — those processing raw input from the eyes — can adjust how they interpret information based on what the rest of the brain is focusing on. This challenges the long-held belief that early visual regions are passive receivers of sensory data.
The researchers used functional magnetic resonance imaging (fMRI) to monitor participants’ brain activity as they sorted various shapes into categories. Uniquely, the rules for categorizing these shapes kept changing mid-task, allowing the researchers to observe how flexibly the brain responded to the shifting demands.
Machine learning tools, including multivariate classifiers, were used to analyse the fMRI data. The results showed that brain activity in the primary and secondary visual cortices reorganised in real-time depending on the categorization rules. Notably, when participants faced difficult decisions — such as shapes near the boundary between categories — the visual cortex’s activity became more distinct. This enhanced neural clarity corresponded with better performance on the task.
“We could actually see clearer neural patterns in the fMRI data in cases when people did a better job on the tasks,” said Dr. Nuttida. “That suggests the visual cortex may directly help us solve flexible categorisation tasks.”
The study offers fresh evidence that the human brain uses early sensory processing regions not just to see, but to think and decide — adapting on the fly as tasks evolve.
Researchers now hope to apply these findings to the development of more adaptable AI systems, using the brain’s flexible, context-driven approach to build machines that can navigate changing environments more effectively