1981 Nobel Prize in Physiology or Medicine — Sperry, Hubel, and Wiesel: The Brain Was a Parallel Distributed System

What You Will Learn in This Article

This article explains how the understanding that the brain is not a single, integrated computer, but rather a parallel distributed system came about. Roger Sperry's split-brain experiments demonstrated that the left and right hemispheres essentially have different personalities. David Hubel and Torsten Wiesel showed that neurons in the cat visual cortex detect features in a hierarchical manner. We will also explore how these two discoveries directly led to today's convolutional neural networks (CNNs) and deep learning, and how our understanding of the brain exploded in the same year that the personal computer was born.

A Story Different from Common Sense: Two People Living in One Brain

We generally believe that a unified consciousness, "me," operates within a single brain. Although we know about the left and right hemispheres, we usually think that they work together to make a single decision. This is common sense.

Sperry's split-brain experiments turned this common sense on its head. When the corpus callosum — the bundle of nerve fibers connecting the left and right hemispheres — is severed, it becomes clear that two independent consciousnesses effectively exist within one person. The left hand does not know what the right hand is doing, and what is seen in the left visual field cannot be verbally described. Furthermore, the two hemispheres can even make different decisions.

Meanwhile, Hubel and Wiesel delved into the reality of the brain from a different angle. What do individual neurons in the visual cortex respond to? The common sense answer would be, "They would respond to specific objects or faces." However, the actual results were different. They responded extremely selectively to very simple features — short line segments at a specific angle, or movement in a specific direction. And the neurons in the layers above combine these simple features to detect more complex features. As you go up the hierarchy, the level of abstraction increases.

If we put the discoveries of these three people together, we get the following picture. The brain is not a single, integrated CPU, but a distributed system in which multiple specialized processors operate in parallel and process information hierarchically. In the language of computer science, it is closest to a hierarchical convolutional neural network running on a dual-core CPU. The left and right hemispheres, two cores, run in parallel, and sensory information such as vision and hearing is filtered through hierarchical layers within each core to be abstracted.

Of course, this analogy is not perfect. The brain does not execute commands sequentially like a silicon core; it expresses itself in probabilistic firing patterns. However, the structure of "modular parallel processing + hierarchical feature abstraction" was validated by the success of deep learning 40 years later.

The Zeitgeist: The Dawn of Personal Computing and the New Cold War

1981 was a year of simultaneous transformation in both information technology and international politics.

One of the most decisive events in world history occurred: the release of the IBM PC 5150 on August 12th. An Intel 8088 processor, 16KB of RAM, and a DOS operating system. This single box brought computing out of the large corporate data centers and onto desktops. The age of the personal computer began on this day. The following year, Microsoft Word was released, and the year after that, the Apple Lisa. The fact that the Nobel Prize for understanding the brain and the popularization of personal computing occurred in exactly the same year is not a coincidence — as we will see later.

In international politics, on October 6th, Egyptian President Sadat was assassinated during a military parade in Cairo by a gunman affiliated with the Muslim Brotherhood. This was the assassination of the protagonist of the 1978 Camp David Accords by his own extremist. On December 13th, the Jaruzelski regime in Poland imposed martial law on the Solidarity movement — the last stand of the Eastern European communist regimes began. And on June 5th, the U.S. CDC reported an unexplained case of pneumonia in five gay men in Los Angeles — which would later be recorded as the beginning of the HIV/AIDS pandemic.

In Korean history, on March 3rd, Chun Doo-hwan took office as president — the opening of the Fifth Republic. The upheaval of May 1980 was solidified into a political system. The operation of the Samcheong Education Corps was coming to an end, and color television broadcasting was becoming widespread, and preparations were underway for hosting the 1988 Olympics (Seoul was selected in September). A dualistic landscape in which politics were suppressed, and consumer society gained color.

In this turbulent year, the Nobel Committee recognized the parallel distributed structure of the brain. In the year that the personal computer was introduced to the world, a picture of how the ultimate computer, the brain, works was drawn. In a sense, this provided half the answer to the question of what to imitate if artificial intelligence wants to imitate the brain.

Roger Sperry: The Interdisciplinary Integrator of Caltech

Roger W. Sperry (1913-1994) was an American neurobiologist. His academic background is particularly unique. He majored in English literature, earned a master's degree in psychology, and a doctorate in zoology, receiving his Ph.D. in zoology from the University of Chicago in 1941. This is a trajectory close to the archetype of today's interdisciplinary researchers.

After receiving his Ph.D., he worked at Harvard University and the University of Chicago (professor from 1946 to 1954), and in 1954, he was appointed as a professor of psychobiology (Hixon Professor) at the California Institute of Technology (Caltech). His representative research was conducted here for 30 years, and he retired in 1984 and remained as an emeritus professor (Trustee Professor Emeritus) until his death on April 1994 in Pasadena. The fact that his title was "psychobiology" is symbolic — that research was conducted at the intersection of psychology and biology, rather than pure biology.

Sperry's early research focused on nerve regeneration and circuit formation. The experiment in which he cut and reversed the optic nerve of a salamander, and the salamander's visual field was also reversed, was a crucial observation that demonstrated that the nerve circuits were strictly hard-wired. This circuit specificity principle later became the chemoaffinity hypothesis.

The turning point came with his encounter with patients with temporal lobe epilepsy who underwent corpus callosotomy. Since the late 1940s, neurosurgeons had been attempting corpus callosotomy — a surgery that severs the corpus callosum, which connects the left and right hemispheres — to prevent seizures that were resistant to medication. This was an extreme measure to prevent seizures from spreading from side to side.

Surprisingly, the patients who underwent surgery were almost normal in their daily lives. Their intelligence test scores were normal, and they could carry on a conversation normally. The doctors who first observed this concluded that the corpus callosum was just a cable that connected the two sides, and that it didn't actually have much function. Sperry didn't believe this conclusion.

The Split-Brain Experiment: The Moment When the Two Sides Make Different Decisions

The experiment designed by Sperry and his graduate student, Michael Gazzaniga, was extremely precise.

The key was to present stimuli to only one visual field. Information from the left visual field is transmitted to the right hemisphere, and information from the right visual field is transmitted to the left hemisphere. When the corpus callosum is intact, this information is immediately shared between the two hemispheres. When the corpus callosum is severed, each hemisphere only knows the information on its own side.

The subject is seated in front of a screen and asked to fixate on the center. A brief (0.1 second) image of an apple is shown in the left visual field. This information goes only to the right hemisphere. When asked, "What did you see?" the patient, who has no language ability in the right hemisphere, will answer, "I didn't see anything." However, when the patient is asked to touch various objects with their left hand (controlled by the right hemisphere), they will accurately pick up the apple. The right hemisphere saw it and can respond with its hand, but it cannot express it verbally.

A more dramatic experiment was also conducted. An image of a snowman is shown in the left visual field, and an image of a chicken foot is shown in the right visual field simultaneously. The left hemisphere sees the chicken foot, and the right hemisphere sees the snowman. When the patient is asked to choose one object from several images that is related to what they just saw, with each hand, the right hand (left hemisphere) picks the chicken, and the left hand (right hemisphere) picks the shovel — a tool needed to clear the snowman.

The crucial moment comes here. When asked why they picked the shovel, the patient's left hemisphere (the speaking hemisphere) will answer, "Because I need a shovel to clean the chicken coop." The left hemisphere, which did not see the snowman, makes up a story to explain the reason why the left hand picked the shovel. Sperry and Gazzaniga called this the interpreter module. The left hemisphere's basic function is to create stories after the fact and deceive itself.

This result opened up a larger picture of brain function. Sperry clearly demonstrated that the left hemisphere is responsible for language, writing, logic, and reasoning, while the right hemisphere is responsible for visual and artistic perception. And his research served as the starting point for subsequent research by other scholars, which completed the picture of the cerebral cortex being divided into sensory cortex (processing information from sensory organs), motor cortex (issuing commands to response organs), and association cortex (integrating and judging sensory information, thinking, and memory).

In the language of computer science, this is similar to the post-log cleanup of a multi-process system. Multiple processes operate in parallel to produce results, and a reporting process creates a story about "why this happened" afterward. What we feel as "I decided to do this" is not the execution, but the narrativization of the post-log. However, this analogy breaks down in that the brain does not deceive itself — the left-brain interpreter deceives itself.

In addition to this, Sperry also devoted himself to studying the precise mechanisms of how nerves form accurate synapses at the correct time in the central nervous system during embryonic development. This research was a continuation of his early research on nerve regeneration.

Hubel and Wiesel: From Johns Hopkins to Harvard, a 22-Year Journey

David H. Hubel (1926-2013) was an American neuroscientist. Interestingly, he was born in Canada but had parents who were U.S. citizens, making him a dual citizen. When he won the Nobel Prize, there was even debate about which country should claim the honor.

Hubel excelled in math, physics, and chemistry in college, and his grades in these subjects were excellent. However, he hadn't studied biology in high school. Nevertheless, in 1957, he was accepted into McGill University's medical school – something he later said he found remarkable. He also recalled that he struggled for a while in medical school due to his lack of biological knowledge. He received his Ph.D. in Medicine from McGill University in 1951 and then moved to Johns Hopkins University as a professor from 1955 to 1958. It was during this time that he met Wiesel, and their long-term collaborative research began.

Torsten N. Wiesel (1924- ) is a Swedish neuroscientist. He received his Ph.D. in Medicine from the Karolinska Institute in 1954 and moved to the United States in 1955, where he met Hubel at Johns Hopkins University. Wiesel was also talented in various sports, serving as the head of the sports club in high school, and he claimed that this was the period in his life where he felt the greatest sense of accomplishment.

The two men had complementary personalities. Hubel was strong in theory and was more spontaneous, while Wiesel was more precise and cautious in his experiments. They complemented each other's weaknesses, and in 1959, the two of them moved together to Harvard University's medical school to run a laboratory. Hubel was a professor at Harvard University's medical school from 1959 to 1982 and became an emeritus professor in 1983. Wiesel was a professor at Harvard University from 1959 to 1983, and then served as a professor and president at Rockefeller University from 1983 to 1998, becoming an emeritus president. This collaborative research, which lasted for more than 22 years, is recorded as one of the most productive duos in the history of neuroscience.

The two men had one goal: to directly measure how individual neurons process visual information. They opened the skull of an anesthetized cat and inserted a microelectrode into the visual cortex – until they reached a single neuron. They showed various stimuli on a screen while listening to the firing (electrical signals) of a single cell through a speaker. If a cell reacted, then that stimulus was of interest to that cell.

Their first discovery was almost accidental. One day in 1959, the two were stimulating the visual cortex cells of a cat by showing various circular images using a slide projector. The cells were silent. Then, as they changed the slide, the edge of the slide – that is, a line that ran across the screen at a specific angle – passed by, and the cell reacted explosively. They had accidentally discovered that this cell responded to a line segment at a specific angle.

The Hierarchical Structure of the Visual Cortex: The Prototype of CNN

Systematic experiments followed. Hubel and Wiesel classified the neurons in the primary visual cortex (V-1) into two types.

Simple cells respond only to specific angle line segments at a specific location in the visual field. Each cell has a different preferred angle – some cells respond to vertical lines, some to 45-degree diagonal lines, and some to horizontal lines. To cover the entire visual field, multiple angle-specialized cells are arranged at each location.

Complex cells respond to a line segment at a specific angle moving across the visual field. The location is somewhat flexible. It appears to be created by combining the output of simple cells.

Hypercomplex cells are more complex. They respond to line segments of a specific length, corners in a specific direction, or movements in a specific direction. As you go up the hierarchy, the level of abstraction of the features increases.

This hierarchical structure can be summarized in the following diagram:

Pixels → Specific angle line segments (simple) → Location-flexible line segments (complex) → Edges/length (hypercomplex) → Shapes → Objects → Faces

Furthermore, they thoroughly investigated the function of the V-2 area and which types of visual information (the shape, color, movement, and perspective of objects) each area of the visual cortex specifically responds to. The process by which the numerous neural impulses originating in the eye are decoded into a single image in the brain's visual cortex – a process that Wiesel particularly elucidated – is summarized as being "decoded" in the brain like a code.

This is the exact structure of today's convolutional neural networks (CNNs). In 1980, Kunihiko Fukushima, inspired by Hubel and Wiesel's discoveries, designed the Neocognitron. In 1998, Yann LeCun created LeNet, and in 2012, Alex Krizhevsky achieved a resounding victory in the ImageNet competition with AlexNet. The idea of hierarchical feature detection has gone directly from the brain to deep learning.

In the language of computer science, the visual cortex is a hierarchical convolutional filter bank. Lower layers detect local features (edges), and as you go up the layers, the receptive field becomes larger, and the features become more abstract. The principle behind how smartphones recognize faces today – the roots of that principle lie in the cat experiments in the basement of Harvard in 1959.

The analogy is that while computer CNNs learn filters through learning (backpropagation), the visual cortex in the brain acquires filters through development and early experience. However, the final structure – hierarchical receptive field expansion – is remarkably consistent.

Decisive Experiment: Developmental Period and Brain Plasticity

Hubel and Wiesel also conducted experiments on the critical period of development. If one eye of a kitten is covered for a few weeks early on, the vision in that eye will be impaired for life. The corresponding area of the visual cortex in the brain is taken over by the other eye. This critical period is irreversible.

This discovery had an immediate impact on clinical practice. The principles of treatment for childhood strabismus and amblyopia – early detection and early intervention – come from this research. Previously, it was believed that vision problems in children would resolve with age, but now, the principle is established that treatment must be done within the critical period.

From a computer science perspective, this discovery is close to a biological demonstration of the exploration-exploitation trade-off in reinforcement learning. In early development, the brain forms circuits in a flexible exploration mode, and after the critical period, it solidifies into an exploitation mode. Once a filter is formed, it is not easily relearned – this is also the neurological reason why adults have difficulty learning new languages.

Academic Impact: A Milestone in Brain Biology

The research of the three men is a decisive milestone in brain biology regarding the function of each area of the brain. No matter how sophisticated the research on neuron excitation conduction and synapses becomes, if the recognition function in the brain is not elucidated, the principles of thought and behavior will remain a mystery. The research of Sperry, Hubel, and Wiesel opened the door to this mystery.

After this award, brain research did not remain solely within pure biology but expanded into psychology, animal behavior, and medicine. This research laid the foundation for the diagnosis and treatment of various brain diseases, especially mental illnesses. The moment when the brain was seen not as a behavioral black box of stimulus and response, but as a structure with specialized functions in each area that cooperate in parallel, was a turning point.

It was also around this time that countries around the world began to invest massive human and research resources in brain biology research. Today's U.S. BRAIN Initiative (2013-), Europe's Human Brain Project (2013-), Japan's Brain/MINDS (2014-), and South Korea's Brain Research Promotion Act (1998, revised in 2007) are all extensions of this lineage.

The Legacy That Continues Today

The impact of the three men's discoveries continues today.

Sperry's split-brain research has established the standard for today's research on brain lateralization, stroke rehabilitation protocols, and neuropsychological assessments. Concepts such as "language is in the left brain's Broca's and Wernicke's areas" and "spatial perception is in the right brain's parieto-occipital cortex" have become clinical guidelines. Gazzaniga later created the field of cognitive neuroscience.

Hubel and Wiesel's visual cortex research is the theoretical root of deep learning. The fact that CNNs have surpassed human levels in image recognition, facial recognition, self-driving cars, and medical image diagnosis is due to the fact that they mimic the structure of the visual cortex. The credit for the 2018 Turing Award for Hinton, LeCun, and Bengio essentially began with the cat experiments in the basement of Harvard in 1959.

In clinical neurology, the concept of the critical period has become the principle of treatment for pediatric ophthalmology, speech therapy, and cognitive rehabilitation. Brain plasticity research has exploded since this discovery and is now the theoretical basis for stroke rehabilitation, neural interfaces, and brain stimulation therapies.

In South Korea's clinical practice, the impact is immediately observable. Since the late 1980s, early treatment for amblyopia has been standardized at Seoul National University and Yonsei University's pediatric ophthalmology departments, and research related to corpus callosotomy has been conducted in neurosurgery. Today, when KAIST and POSTECH researchers publish deep learning papers, these three men are at the roots of that lineage.

Why is it important?

What the three scientists established is that "the brain is not an integrated computer, but a parallel distributed system."

Sperry demonstrated that the left and right hemispheres are essentially independent processing units – dual-core. Hubel and Wiesel demonstrated that visual information is abstracted through hierarchical filters – hierarchical feature extraction. These two principles combined make today's deep learning possible.

It is not a coincidence that the parallel distributed structure of the brain received the Nobel Prize in the year that personal computers emerged. From the perspective of information processing, when humans began to understand themselves, they also began to try to implant that structure into the machines they created. Forty years have passed since 1981, and today we are using computers that mimic the brain to understand the brain itself.

After this award, the flow of understanding the brain continued as follows:

• 1986, Papanellou, Salzman, and Newsome: Visual cortex MT area and motion perception – detailed understanding of higher-level visual processing.
• 2000, Carlsson, Green, and Kandel: Signal transmission in the nervous system – synaptic chemical mechanisms.
• 2014, May-Britt Moser and Edvard Moser: Hippocampal spatial cognitive grid cells – the brain's GPS.
• 2024, Hinton and Hopfield (Physics Prize): Statistical mechanical foundation of artificial neural networks – the final convergence of the Sperry, Hubel, and Wiesel lineage.

Decisive moments when these discoveries moved to CS:

• 1980, Fukushima's Neocognitron: Directly cites the Hubel and Wiesel paper.
• 1998, Yann LeCun's LeNet-5: Practical application of CNN, MNIST handwritten digit recognition.
• 2012, AlexNet: Dominates ImageNet, the beginning of the deep learning era.
• 2017, Transformer: Attention mechanism that integrates vision and language.

→ Previous: 1980 — Benacerraf, Snell, and Dausset → Next: [1982 — Batch 8 in progress]