Practice teaches the brain to effectively handle multiple tasks at once

To understand how the brain copes with performing two tasks simultaneously, researchers developed a special task for mice. The animals had to continuously move a lever and, in parallel, respond to sound signals: depending on the tone, they had to either perform an action (“Go”) or refrain from it (“No‑Go”). Thus, the mice simultaneously performed a motor task and a task involving the making of a sensory decision.

Using two‑photon calcium imaging, the scientists observed the activity of the same neurons in the secondary motor cortex (M2) in the same animals for weeks. This made it possible to track how brain function changes as mice learn to handle two tasks simultaneously.

In the early stages, the brain does not try to rigidly separate the tasks. Instead, it actively coordinates competing demands: even those neurons that are primarily responsible for one task change their activity when the second task is performed. It turns out that the brain temporarily adjusts some processes to accommodate others, so that both tasks can be performed, albeit not perfectly, but simultaneously. At the same time, the most intense conflicts arise in the neurons involved in both tasks simultaneously: they become hotspots of competition and reflect the limited resources of the brain.

As training progresses, the strategy changes. The brain begins to recruit more neurons specialized for each individual task, and the representations of the tasks gradually become more distinct. As a result, the tasks begin to be performed more independently, and mutual interference decreases.

The secondary motor cortex (M2) plays a key role in this process. When experts moderately suppressed its activity during training, the mice did not improve their performance with practice. But as soon as the suppression was stopped, the ability to multitask quickly recovered and continued to grow. This proves that M2 is not just involved in task performance, but is critically important specifically for multitasking learning.

The researchers tested whether the identified biological principle could be applied to machine systems. They trained recurrent neural networks to solve a similar dual task. It turned out that a simple, rigid separation of task-related concepts is not the most effective approach. The networks learned faster if coordination between tasks was maintained in the early stages, and the separation of concepts occurred gradually, just as a living brain does.

According to Professor Ke Ya, the study made it possible to see the very mechanism of multitasking learning: at an early stage, even a decrease in the activity of neurons responsible for one task helps to cope with a competing task and ensures the first successes.

Professor Jung Win‑ho emphasizes that effective multitasking is a balance between coordination and specialization. These principles can not only help to better understand multitasking disorders in neurological diseases but also serve as the foundation for creating AI that will learn to manage multiple tasks as flexibly as the human brain does.