Brain rewires itself to bypass multitasking bottleneck
A New Window Into the Brain’s Learning Architecture
For decades, scientists have studied how the brain learns by observing snapshots - measuring activity before training and after expertise has formed. They knew that experts in bird identification, car recognition, or even Pokemon characters showed distinct neural activity in the temporal cortex. But no one had watched the transformation happen in real time. The Georgetown University team led by Maximilian Riesenhuber, PhD, a professor of neuroscience at Georgetown University School of Medicine and co-director of the Center for Neuroengineering, designed a longitudinal study that tracked the same brains before and after intensive training. [2] Volunteers sorted morphed images of cars into two categories by identifying subtle visual differences. They completed more than 30,000 sorting trials over 5 to 10 weeks using a smartphone app designed as a game. The researchers scanned participants’ brains with fMRI and EEG before training began and again after the practice period ended. This before-and-after measurement method allowed them to directly observe what had only been calculated or inferred from cross-sectional studies.
The initial scans revealed that early in learning, the sorting task primarily activated the prefrontal cortex. This brain region handles executive functions such as planning, reasoning, and conscious decision-making. Because the prefrontal cortex generally manages one demanding task at a time, it has long been viewed as the major bottleneck limiting human multitasking. The data showed exactly what the standard model predicted: new skills required conscious attention and deliberate effort. The prefrontal cortex lit up with activity as participants strained to categorize the morphed car images. This finding matched decades of research showing that the frontal lobes are essential for learning novel tasks.
But the final scans told a different story. After weeks of practice, brain activity had shifted dramatically. The same categorization task was now being handled mainly by the temporal cortex, a region involved in memory and recognizing complex objects. First author Patrick Cox, PhD, who began the study as a graduate student in Riesenhuber’s lab and is now an assistant professor of psychology at Lehigh University, noted that previous studies had shown parts of the temporal cortex activated by particular object categories in experienced observers - birds, cars, even Pokemon. [3] But those studies only looked after people became experts. The strength of this longitudinal study is that it measured before and after training, revealing that extensive training essentially created a category-selective area in the temporal lobe that was not there before. This direct observation of neural reorganization challenged the existing model that learning primarily strengthens existing connections within the same brain regions.
The Data That Defied the Frontal Bottleneck Model
The standard model of multitasking held that humans cannot truly perform two tasks simultaneously. Instead, scientists argued that the brain simply alternates attention between tasks so quickly that it creates the illusion of doing both at once. This rapid-switching theory explained why people make more errors when trying to talk on the phone while driving or walk while texting. The prefrontal cortex, with its limited capacity for conscious control, was thought to be the immutable gatekeeper that prevented true parallel processing. But the Georgetown data did not fit this model. The researchers found that information from the newly developed car-selective area in the temporal cortex could bypass the prefrontal cortex entirely. Instead of routing through the frontal bottleneck, neural signals traveled directly to brain regions responsible for producing responses. This bypass meant the prefrontal cortex was no longer required for the automatic task.
The researchers then tested whether this neural reorganization actually enabled true multitasking. They asked participants to perform the now-automatic car sorting task while simultaneously doing a second task. The results showed that the more the car sorting task was “offloaded” from the prefrontal cortex, the better participants performed the second task at the same time. This finding directly contradicted the rapid-switching theory. If the brain were simply alternating attention, performance on both tasks should have declined equally regardless of how well-practiced the first task was. Instead, the data showed that automaticity freed cognitive resources for parallel processing. Riesenhuber explained that experience remodels the brain to bypass that frontal bottleneck. The prefrontal cortex then stays free for whatever else you want to do, increasing your capacity.
The researchers measured this capacity increase precisely. Participants who showed the greatest shift of neural activity from prefrontal to temporal cortex also showed the best dual-task performance. The correlation was strong and consistent. This data point anchored the finding that the brain can physically reorganize to support true multitasking. Riesenhuber stated clearly: “What we show is that the circuitry actually changes so the brain can do two things at once. This really is true multitasking.” The study, published June 4 in the Journal of Cognitive Neuroscience under the title “Extensive Experience Remodels Neural Task Circuitry to Escape the Frontal Bottleneck and Increase Automaticity of Categorization,” provided the first direct evidence that the brain builds specialized circuits that operate independently of conscious control.
Implications for Habit Formation and AI Design
The implications extend beyond understanding how people learn to drive or sort car images. The research team identified a specific mechanism that explains why habits are so difficult to break. Because well-learned behaviors move into brain circuits that are less dependent on conscious control, simply trying to think about something else may not be enough to stop an unwanted habit. Riesenhuber explained that the first step to unlearning something is understanding where it is actually happening in the brain. This shows why strategies like telling someone to think of something else don’t really help, because they don’t really have the behavior under conscious control. The temporal cortex circuits operate automatically, without requiring the prefrontal cortex’s oversight.
Current AI systems struggle to learn continuously without disrupting previously acquired knowledge, a problem known as catastrophic forgetting. The Georgetown team suggests that the human brain solves this by transferring well-learned skills into the temporal cortex, freeing the prefrontal cortex to focus on new challenges. This flexible architecture allows existing knowledge to serve as the foundation for future learning. Today’s AI systems generally lack that kind of flexible architecture, according to Riesenhuber. The brain’s ability to create new category-selective areas in the temporal lobe may offer a biological blueprint for building AI systems that can accumulate skills without overwriting old ones.
The team now plans to investigate exactly what signals move learning from one brain region to another. Cox raised another critical question: what kinds of tasks can be learned well enough to do in parallel? He noted that people can walk and chew gum at the same time, but looking at phones to text while driving will never be safe because drivers take their eyes away from the road. It comes down to being able to train fully separate neural circuits for two tasks to become compatible. The research team included Clara A. Scholl, Marissa L. Laws, Nelson E. Jaimes, and Xiong Jiang of Georgetown University. The work was supported by the National Science Foundation (BCS-1232530), the ARCS Foundation, and the Army Research Laboratory (W911NF-24-1-0097). The authors reported no personal financial interests related to the study. The brain does not just strengthen existing pathways with practice. It builds entirely new neural circuits that operate independently, allowing true parallel processing of well-learned tasks.
Sources
2. Georgetown University School of Medicine
4. Journal of Cognitive Neuroscience