May 29, 2024, was a very special day for Haim Sompolinsky, a professor at Harvard University in the USA and Hebrew University in Israel. On that day, he received the world’s largest brain research prize, the Lundbeck Foundation’s ‘The Brain Prize 2024,’ in Copenhagen. The day was special not only because of the prestigious award presented by King Frederik but also because the ceremony marked a return to Copenhagen.
Haim Sompolinsky was born here 74 years ago. In October 1943, his father David Sompolinsky, then a young veterinary student, was among the resistance fighters who organized the rescue of several hundred Danish Jews from the Nazis before he himself had to flee. After the war, David Sompolinsky started a family and worked at the State Serum Institute in Copenhagen. The family emigrated to Israel in 1951 when Haim was two years old.
In honor of the award ceremony, Haim Sompolinsky wrote a scientific essay for Danish Medical Journal, ‘Deciphering the Mysteries of the Neural Code,’ dedicated to his father. ‘For me, it is very emotional not only to receive the prize in Denmark but also, given my family’s history, to publish and give an interview to Danish Medical Journal,’ says Haim Sompolinsky.
‘For me, it is very emotional not only to receive the prize in Denmark but also, given my family’s history, to publish and give an interview to Danish Medical Journal’Haim Sompolinsky
He received the prize along with two other researchers from American universities: Professor Larry Abbott from Columbia University and Professor Terrence Sejnowski from the Salk Institute. The prize includes 1.3 million euros to be shared among the three. The prize’s justification states that the researchers ‘have been pioneering in computational and theoretical neuroscience and have made crucial contributions to our understanding of the brain. They have also paved the way for the development of brain-inspired artificial intelligence.’
In other words, they have developed theories and computer models that uncover key mechanisms by which neural circuits generate cognition and behavior. Their work has also inspired the development of artificial neural networks.
In Haim Sompolinsky's childhood home in Israel, his father's microscope stood on a desk in the middle of the living room. His father, who had earned a Ph.D. in microbiology and helped establish the field in Israel, was a professor of microbiology at Bar-Ilan University and director of a clinical microbiology laboratory. His research included zoonoses -diseases transmitted from animals to humans.
‘He could sit for hours looking through his microscope and taking notes,’ Haim Sompolinsky recalls. He himself would occasionally be allowed to look through the eyepiece to glimpse his father’s fascinating world of microbes. Haim also enjoyed reading books from his father's large library, which contained a mix of Jewish texts and scientific books on physics, mathematics, and microbiology. His father also had a home laboratory complete with glass flasks, gas burners, and cages with guinea pigs as experimental animals. Haim and his siblings were tasked with taking care of the animals.
Haim Sompolinsky’s choice of a research career was therefore not entirely coincidental. ‘I was very inspired by my father. He was curious and dedicated to his research. A real scientist.’ His father maintained lifelong contact with family and friends in Copenhagen and collaborated with Danish researchers. Haim Sompolinsky still has his father’s examination papers from the Veterinary School and his scientific articles.
As a young man, Haim Sompolinsky considered studying medicine but chose physics because his mind easily grasped mathematical formulas and the laws of physics, and he was interested in fundamental principles. He was particularly interested in the physics domain of statistical mechanics, which aims to explain how new forms emerge, such as cloud formations in the sky and vortices in water, without external design but under the influence of microscopic interactions and statistical regularities.
Later on, he specialized in the physics of disordered systems, and his research explained how the structural disorder in materials gives rise to intricate and rich macroscopic thermal and magnetic properties. For Haim Sompolinsky, it was natural to apply the theoretical physical approach to studying the “mess” that the brain might appear to be. ‘Living tissues and the brain in particular also have disordered structures. So, I could use our statistical mechanical models to develop theories about information processing in neural circuits,’ Haim Sompolinsky explains.
Neural circuits are groups of interconnected neurons that share a common information processing task. These circuits are part of larger neural networks spanning entire brain regions and support higher-level cognitive functions. Haim Sompolinsky has helped model fundamental mechanisms that maintain an internal balance in neural circuits by the feedback of excitation and inhibition, such that excess excitation is rapidly countered by enhanced inhibition and vice versa. These balancing mechanisms make the brain both flexible and robust.
‘If I or other employed physicists wanted to work on and teach brain science, that was our own business, we were told. We should do it in our spare time’Haim Sompolinsky
The interplay between excitation and inhibition leads to another type of balance in neural circuits, according to Sompolinsky’s research. This one is the balance between intrinsic brain dynamic, which represents the brain’s hypothesis about the world, and external sensory inputs, which tell the brain about the actual state of the world at any given moment. He has also opened the door to exploring what happens when this balance is disturbed, as seen in some psychiatric disorders and neurological diseases.
‘If we look at schizophrenia, the neural circuits generate their own hypotheses about an imaginary world through their internal dynamics, and that view of the world is not sufficiently balanced and corrected by signals from the real world,’ explains Haim Sompolinsky. ‘Conversely, in other diseases, the brain might have too much focus on signals from the external world and do not balance them against prior expectations embedded in the neural circuit. This can lead to hypersensitivity to external stimuli, where every sound is noise, and every touch is an irritation because the internal dynamics of the circuit that normally balance sensory inputs against prior experience and put these inputs in context are disturbed.’
‘This is an example of how models of neural circuits and their internal dynamics in the brain can complement the biological, molecular, and cellular levels of understanding. So, both in basic research and in the clinic, it is important to have a complementary approach so we not only target treatment at the molecular level but also develop treatments aimed at the system level.’
About 40 years ago, when Haim Sompolinsky started his research career in neuroscience, it was unusual for a physicist to move into a biological field. ‘It was controversial. I, for instance, was hired by the university to teach physics and conduct research on physical systems. If I or other employed physicists wanted to work on and teach brain science, that was our own business, we were told. We should do it in our spare time.’
Resistance also came from other professional groups. ‘Neurologists and biologists didn't understand what physics had to do with biology. To them, studying the brain was like studying any other organ: kidneys, lungs, or the heart. And they certainly didn’t like a bunch of physicists coming in to explain to them how the brain works.’ He admits that, at that time, he and other physicists might have been a bit too confident in their views of the brain. ‘We were extremely naive when we started, and it took a long time for us to become less arrogant and understand that the brain is far more complex than we thought.’
Despite the resistance, the group of physicists interested in brain science persisted, and long story short: they established a virtual interdisciplinary research center, Hebrew University’s acclaimed Interdisciplinary Center for Neural Computation (ICNC). ‘It was the first in the world (alongside a similar initiative in Caltech) with an agenda to study the brain and teach brain science from multiple academic angles. We established a Ph.D. program with its own special curriculum. Every student had to learn about neurobiology, cognitive science, computer science, and physics to understand the dynamics of the nervous system and how they are linked to functions. And to this day, this PhD program is quite unique.’
ICNC has been replaced nowadays by the modern Edmond and Lily Safra Center for Brain Science, which maintains and nourishes the multidisciplinary spirit of research and education.
Haim Sompolinsky is very passionate about fostering new generations of researchers. ‘Being a researcher is a journey, and there are frustrations and obstacles along the way. When you read my “success stories,” you should know that I have also faced setbacks, failures, and ideas that didn’t hold up. That's science, and that’s why you have to be dedicated and persistent. You can’t just research until four o'clock and then go play golf. It’s a lifelong mission.’
In Haim Sompolinsky’s essay in Danish Medical Journal, he includes an image from a light microscope showing neural circuits from a fruit fly's brain. That image represents one of the highlights of his research career: the experimental verification of his theory about neurons’ activity related to spatial orientation.
‘I developed my theory of the neural circuit for spatial orientation 30 years ago, a theory known as the “ring model,” and exactly 20 years later, the ring circuit was directly observed in an experiment in the fly spatial navigation system. It was almost too good to be true—to see not just the activity of the neurons along the ring but also the pattern of their synaptic connections as predicted.’
‘When you read my “success stories,” you should know that I have also faced setbacks, failures, and ideas that didn’t hold up. That's science, and that’s why you have to be dedicated and persistent. You can’t just research until four o'clock and then go play golf. It’s a lifelong mission’Haim Sompolinsky
The fruit fly also plays a crucial role in the explosive development in neuroscience that Haim Sompolinsky has witnessed. ‘Now we have new tools to precisely map the connections between neurons in large networks. So, it was a milestone for our community when we recently finally got a complete mapping of a fly’s brain with all its 150,000 neurons and their connections, the so called “fly connectome”.’
But the ‘revolution’ also presents a dilemma. The overwhelming amount of data is both a blessing and a challenge. ‘It’s fantastic that we now have the complete mapping of the fly brain. But who can interpret this map? We can see how the fly’s neurons are connected, but we have only fragmental knowledge of how these connected circuits generate the animal’s ability to fly, navigate, hunt prey, or court a mate.’
‘Similarly, in the cognitive domain, we still don’t know how neural circuits generate emotions, thoughts, planning or language. And this huge gap between knowing a lot about structures but very little about their links to functions is a challenge for understanding the brain. And this gap gets larger the more data we get at the microscopic level.’
That is why Haim Sompolinsky also writes about ‘the power of theory’ in his essay. ‘Without theory, there is no way to close that gap. The role of theory is to explain how these building blocks act together as a system.’
Until now, Haim Sompolinsky has researched neural circuits, but artificial intelligence, also inspired by neuroscience, opens up entirely new possibilities. ‘Advances in neuroscience have provided a model for developing artificial neural networks and will continue to serve as a generator for much more powerful AI in the future, but we can also use artificial intelligence as a generator for neuroscience,’ he says.
‘Until just ten years ago, we could explain relatively simple cognitive functions and visual perceptual tasks, such as determining the position and orientation of an object or the position of the animal in an environment, using the theoretical and computational tools available at that time. But not how the visual system recognizes an object like a pen on a cluttered desk or a car in a busy street. That involves more sophisticated mechanisms and requires understanding information processing in multiple layers of neural circuits.’
‘But with AI, we now have a tool to build large-scale neural network models that more or less follow the structure of the visual system and can perform visual tasks at a high level, sometimes at human level quality. In fact, we can now begin to model information processing in the complete visual system, which makes up a large part of our brain. So, AI makes it possible to vastly expand and upgrade our models of the brain.’
Another new possibility AI offers is understanding how the brain processes language. ‘If you had asked me a few years ago whether I could figure out how the brain generates language, I would have dismissed it outright. It’s too complicated—too many parts of the brain are involved, and it involves a hierarchy of so many subcomponents that we might as well give up trying to understand it. So, we left all the language stuff to the linguists. But now, with the Large Language Models, it’s possible to begin approaching it as a theoretical neuroscience subject. By probing the artificial intelligence neural networks that generate language, we can begin to uncover the underlying principles of how language is comprehended and produced by the human brain. So, it opens up a whole new area. It’s another revolution, and we're just scratching the surface.’
Since young Haim Sompolinsky began venturing into the field of neuroscience, his perspective on the brain has changed. And the most important lesson is that the brain is almost incomprehensibly complex. ‘Understanding the brain means understanding the richness and complexity of the world because it is this world that is represented in the brain by patterns of neural activity. It’s astonishing that these complex neural codes can do all that. And that’s why it’s so fascinating. That’s why I also talk about solving the mysteries of the neural code—because it’s perhaps the biggest mystery of science that a biological tissue can represent so much richness.’
Facts
Haim Sompolinsky is 74 years old and ready for the next chapter in his research career, which he is confident will bring new major discoveries thanks to artificial intelligence. Maybe even the solution to the mystery of the neurons’ language.
He doesn’t think about age. Neither did his father. David Sompolinsky wrote his last scientific articles in his mid-nineties and only reluctantly retired at 94. He never forgot his time in Copenhagen and the Danes’ efforts to save the Jews. David Sompolinsky was sharp until the end and died at the age of 100 in 2021.
By that time, his son had long since taken up the mantle as a researcher and made a name for himself internationally. And with the award ceremony in Copenhagen, Haim Sompolinsky feels that a circle has been completed.