They investigate how the brain learns when there is no task, reward or instructions

Aimless browsing can also be a form of learning, even if we don't realize it. Looking past shop windows without looking for anything, letting ourselves be drawn to a new park, or walking down a supermarket aisle without a list in hand. Moments that seem trivial, almost devoid of purpose. But, according to new scientific research, in those moments the brain is working silently and efficiently: organizing the environment, capturing visual patterns, memorizing without being asked.

A series of experiments conducted at the Janelia Institute of the Howard Hughes Medical Institute (HHMI) in the United States suggests that this form of "mental wandering" not only makes sense, but may be one of the pillars of learning.

By simultaneously recording the activity of thousands of neurons in mice, researchers observed that the brain can learn without instructions, without rewards, and without a clear goal. The study was published in the journal Nature.

What a new study reveals about learning without rewards

A team of HHMI neuroscientists analyzed whether the brain can be reconfigured through spontaneous exploration of the environment, without the need for rewards or training. The finding postulates that even when nothing "useful" is being done, the nervous system continues to form internal maps and encode visual information that will be valuable later, explained Marius Pachitariu, the group's leader at the Janelia campus.

To study this form of unsupervised learning, the team designed a virtual reality system with digital corridors that mice could navigate. These corridors were covered with different visual textures, some of which were associated with rewards (such as food). The key: there were also mice that didn't receive any rewards; they simply walked through the environment without any prompts.

During the experiment, the scientists used a device called a mesoscope that allowed them to record the activity of up to 90,000 neurons at the same time. Thanks to this volume of data, they were able to observe which brain regions were activated during simple exposure to stimuli, without the mice knowing what to do.

Lin Zhong, the study's lead researcher, noted: "I was very surprised. I've been conducting behavioral experiments since my PhD, and I never imagined that without training mice to perform a task, the same neuroplasticity would be found."

The Value of Simply Looking

When comparing two groups—one trained with rewards and another that simply explored the corridors without any prompts—the scientists found that both exhibited similar changes in the visual cortex. In other words, learning occurred even without reinforcement. In Zhong's words, "it means you don't always need a teacher: you can still learn about your environment unconsciously, and this type of learning can prepare you for the future."

Furthermore, when these exploratory mice were assigned a specific task, they solved the challenge much faster than those starting from scratch. Unsupervised learning acted as a kind of brain warm-up.

Learning Before There's Anything to Learn

The study distinguishes two mechanisms that can occur simultaneously in the brain: unsupervised learning, which occurs during passive observation of the environment, and supervised learning, which is activated when there is a clear goal or reward.

In the neural recording, it was observed that different regions of the visual cortex responded depending on the type of stimulus: some passively encoded visual patterns, while others were activated when a command was incorporated. The authors explain that the brain “could simultaneously use both algorithms: an unsupervised component to extract features and a supervised component to assign them meaning.”

“The results suggest that the neural representations that emerge from unsupervised learning can accelerate learning in subsequent tasks,” Zhong and Pachitariu wrote in Nature.

The study was conducted in mice, and it is still unclear how the brain chooses which stimuli to retain or discard. However, according to Pachitariu, this work “is a gateway to studying these unsupervised learning algorithms in the brain.” He added: “If that is the main way the brain learns, as opposed to a more instructed and goal-directed way, then we need to study that part as well.”