1963, Harvard University Neurophysiology Laboratory; The regular sound of the apparatus recording neuronal activity broke the .silence of the laboratory “…Tick… Tick… Tick“

David Hubel and Torsten Wiesel had been recording the activity of cells in the visual cortex of a kitten for hours [1, 2]. A very fine electrode was positioned in the primary visual cortex (V1); the delicate electrode was used to record the activity of individual neuronal units. The cat’s eyes were open, and lines with different orientations were displayed on a screen in front of it.

Each time a neuron became active, the loudspeaker produced a short sound. “Tick… tick… tick…”

But this time, the experiment was different. Several weeks earlier, the researchers had gently sutured one of the kitten’s eyelids shut just a few days after birth. Neither the cornea, nor the retina, nor the optic nerve had been damaged; only light was prevented from entering that eye. Now it was time to open the eyelid.

The eye looked completely healthy. The cornea was clear. The retina was healthy. The optic nerve was also undamaged.

If you looked only at the eye’s appearance, you would think that both eyes could see equally well. But Hubel and Wiesel wanted to ask the brain itself a question.

First, they left the healthy eye open and covered the other eye. The loudspeaker immediately came alive. “Tick… tick… tick… tick…”

The neurons fired one after another.

Now they switched the eyes. The eye that had been closed for several weeks was opened. Everyone expected to hear the same sounds.

But…

Silence. Only an occasional weak discharge. As if that eye did not exist at all.

At first, Hubel and Wiesel thought that perhaps the eye had been damaged. But examination showed that the eye was completely healthy.

So where was the problem?

The answer lay in the brain itself. During the first weeks of life, neurons in the visual cortex are still forming their connections. During this critical period, each neuron is like a child who must decide which eye to “listen to.” When both eyes receive natural visual input, these connections are distributed almost equally between the two eyes. But in this kitten, only one eye was active.

For weeks, the brain’s neurons had received signals only from that eye. As a result, the neural connections associated with the open eye became stronger day by day, while the connections associated with the closed eye, the closed eye, which received no visual inputs, gradually weakened and were eliminated.

When the eyelid was opened several weeks later, it was already too late.

The eye was healthy…

But the brain had no longer learned to use it.

To make sure, Hubel and Wiesel repeated the same experiment in adult cats as well. This time, opening the eyelid produced a completely different result. After several weeks of closing one eye, the cats still responded to both eyes almost as before. The adult brain was no longer nearly as capable of modifying the organization of its visual cortical neurons. At that moment, one of the most fundamental principles of neuroscience became clear: “For normal development, the brain depends, occur, the brain dependsnot only on genes, but also on experience. An eye can be completely healthy, but if the brain loses the opportunity to learn to see at the appropriate time (the critical period), normal vision will never fully develop.”

This experiment showed that the brain, contrary to the prevailing view at the time, is not a fixed and immutable structure. Hubel and Wiesel demonstrated that experience can shape neural circuits; neurons that receive more input strengthen their connections, while neurons deprived of experience gradually lose some or all of their capabilities.

These studies provided important experimental evidence for the role of experience in the formation of sensory circuits [1, 2].

From Harvard to MIT; 60 Years of Searching to Understand How the Brain Learns

But alongside their major answer, Hubel and Wiesel also left a larger mystery for future generations of scientists.

If experience changes the brain, exactly where does this change occur?

When we remember the face of a new colleague, or learn to recognize a species of bird, a new car model, or even a medical image faster than before, exactly which part of the brain is changing?

Every time we learn something new, are the visual circuits being rewired from scratch? Or has the brain found a more subtle and intelligent way to learn?

This question remained unanswered for more than 60 years. During these years, technologies for recording neuronal activity, brain-imaging methods, and computational neuroscience advanced at a remarkable pace; yet the answer to this question remained unclear.

Until the summer of 2026, when a team of researchers from the McGovern Institute for Brain Research at MIT, together with collaborators at York University in Toronto, published a study in the prestigious journal Nature Communications that once again continued the story from where Hubel and Wiesel had stopped [3].

But this time, their question was no longer, “Does experience change the brain?” Everyone already knew the answer to that question. The new question was: “Exactly what does the brain change while learning?”

Lin Sörensen (a postdoctoral researcher at McGovern), James DiCarlo (a professor and senior researcher at MIT), and Kohitij Kar (an assistant professor at York University) [3] examined the activity of neurons in a brain region called the inferior temporal cortex (IT), a region that can be considered the final station in the brain’s object-recognition pathway [3].

It is in this region that, after integrating information about lines, colors, edges, and shapes, colors, edges, and shapes, the brain finally recognizes: “This is an elephant,” “This is a chair,” or “This is a human face.”

The scientists recorded activity in the IT cortex of two groups of monkeys: one group that had never been trained and for which the images they saw were almost meaningless, and another group that had practiced for weeks to correctly recognize new objects, even when they appeared at different sizes, angles, and against different backgrounds.

Nevertheless, when neuronal activity was analyzed using much more precise methods, subtle but persistent differences emerged. IT neurons in trained monkeys encoded the objects the animals had learned somewhat differently; as if the brain did not need to destroy and rebuild its architecture in order to learn, but only to make subtle adjustments to what was already there.

When They Built Several “Digital Brains”…

The researchers did not search for the answer to their question only in monkey brains. They went a step further—a step that Hubel and Wiesel in the 1960s could not even have imagined.

“If we could build a brain from scratch, could we watch, moment by moment, exactly which neurons and which circuits change during learning?”

For this purpose, Sörensen designed and trained not one but a collection of artificial neural networks whose architecture was inspired by the visual pathway of the mammalian brain, with their internal components mapped onto the monkey’s IT cortex. Like the real brain, these networks processed images step by step, from extracting simple features such as edges and lines to reaching a layer that played a role similar to the IT cortex—the same region that in the real brain is responsible for representing object identity.

Then the same categories of objects that the monkeys had seen were presented to these networks as well. The models were trained using gradient descent; that is, with every error, their internal parameters were adjusted slightly. They saw the image again, tried again, and learned from their errors. This cycle was repeated thousands of times until the training was complete.

Here there is an important point: gradient descent, in the conventional form used in artificial neural networks, is not considered a direct model of a known biological learning mechanism in the brain. The brain learns in a different way. However, the researchers were not trying to prove that the brain works exactly this way; they wanted to see whether such abstract models could predict the pattern of changes in the real brain.

And the result was interesting: among the set of networks that were trained, only some of those whose architecture resembled that of mammals showed a learning pattern similar to that of real monkeys. In those successful models, the layer that played the role of the IT cortex showed exactly the same kind of subtle changes that the researchers had previously observed in the brains of trained monkeys.

This finding suggested that learning-related changes occur not in the visual system itself, but in circuits downstream of it—circuits that interpret visual information and make decisions based on it. According to Kohitij Kar, one of the authors of the study, this finding suggests that the gap between the region from which the recordings were made and the final behavior is filled with changes that should be investigated in future research [4].

Hubel and Wiesel showed the world that experience can change the brain; but today, this research provides a more precise picture of how neural representations in the IT cortex change after learning objects and also highlights the possible role of downstream circuits.

Sixty years ago, the answer to this question emerged from the “tick… tick… tick…” of a kitten’s neurons. Today, the same story continues at the heart of artificial neural networks.

Perhaps this is the most beautiful feature of science: every discovery is not the end of a story, but the beginning of a question that the next generation of scientists will continue to pursue.

References

  1. Wiesel, T. N., & Hubel, D. H. (1963). Single-cell responses in striate cortex of kittens deprived of vision in one eye. Journal of Neurophysiology, 26(6), 1003–1017.
  2. Hubel, D. H., & Wiesel, T. N. (1963). Receptive fields of cells in striate cortex of very young, visually inexperienced kittens. Journal of Neurophysiology, 26(6), 994–1002.
  3. Sörensen, L. K. A., Kar, K., & DiCarlo, J. J. (2026). Hierarchical optimization predicts plasticity in the macaque inferior temporal cortex following object training. Nature Communications.
  4. Michalowski, J. (2026, July 8). How visual learning happens in the brain. MIT McGovern Institute for Brain Research.