How does the brain overcome the limits of multitasking? Neuroscientists at CityUHK and CUHK reveals a dynamic neural strategy and hints for artificial intelligence training

Why is it so difficult to perform two tasks at the same time? For decades, scientists have debated whether multitasking is limited by a “central bottleneck” that restricts simultaneous processing, or by competition for limited neural resources shared across tasks. But neither view fully explains how the brain manages these competing demands or how practice transforms an interference-prone system into more efficient multitasking.

A study co-led by Professor Yung Wing-ho, Chair Professor in the Department of Neuroscience at the College of Biomedicine of City University of Hong Kong (CityUHK), and Professor Ke Ya, from the School of Biomedical Sciences at the Faculty of Medicine (CU Medicine) of The Chinese University of Hong Kong (CUHK), has uncovered how the brain dynamically manages and reorganises its neural resources when learning to multitask. By tracking the same individual neurons throughout dual-task training via mice models, the researchers found that multitasking is not governed simply by a fixed processing bottleneck. Instead, the brain initially combines resource competition with neural coordination, then progressively recruits more task-specific neurons and separates task representations through learning, enabling more efficient parallel processing. They said the biological findings can be applied to the artificial intelligence training in multitasking. These findings have been published in the leading neuroscience journal Neuron, under the title “Dynamic coordination and segregation mechanisms in higher cortex for parallel task processing”.

To investigate how the brain handles two tasks simultaneously, the researchers developed an original paradigm in which mice had to maintain a continuous lever movement task while listening to different auditory cues and deciding whether to respond, a sensory decision-making task known as a “Go/No-Go” test. Using longitudinal two-photon calcium imaging, the researchers tracked the activity of the same individual neurons within a large neuronal population in secondary motor cortex (M2) over several weeks of training, allowing them to observe how neural activity was reorganised as the animals learned to multitask.

The study showed that multitasking interference can be traced at the level of individual neurons. Neurons involved in both tasks became hotspots of competition, providing a cellular basis for the brain’s limited capacity to process competing demands. But this was only part of the story. Surprisingly, the researcher found that the competition was not confined to neurons shared by both tasks. Even neurons mainly responsible for one task adjusted their activity when the other task was being processed. This adjustment helped the brain coordinate the two competing demands and achieve early multitasking success.

With continued training, the brain adopted a different strategy. More task specific neurons were recruited, while the neural representations of two tasks became progressively separated, allowing the tasks to be progressed more independently with less interference. The team further showed that M2 plays a causal role in this learning process. When M2 activity was moderately suppressed during training, the animals failed to improve with practice. Once the suppression was removed, their multitasking performance rapidly improved.

Professor Ke said, “We discovered the neural mechanism of learning multitask in this study. During early learning, reduced activity in neurons mainly supporting one task actually contributed to processing the competing task and helped both tasks succeed before a more efficient solution has been learned.”

Beyond explaining how the brain learns multitask, the researcher asked whether the strategy might represent a more general computational principle for managing competing demands. Using Recurrent Neural Networks trained on a similar dual-task problem, the team found that simply separating the representation of the two tasks was not the most effective solution. Networks learned faster when early coordination was preserved while other tasks representations became progressively separate, closely mirroring the strategy observed in the biological brain.

Professor Yung said, “Our findings suggest that efficient multitasking requires a balance between coordination and specialisation.These principles may provide a framework for understanding multitasking deficits in neurological disorders and, importantly, offer a biologically inspired strategy for designing artificial intelligence systems that can learn and manage multiple competing tasks more efficiently.”

Professor Yung and Professor Ke are the corresponding authors of the paper. Other study members include first author Dr Wang Shuting, and contributing authors Dr Zhu Yun and Dr Li Chunyue.