Demis Hassabis, Father of Google's Artificial Intelligence

Part 3. The Brain, the Blueprint of the Mind

7 Becoming a Student Again

Kim Kyung-ran, Kim Kyung-jin

To build artificial intelligence, one must first understand natural intelligence. UCL Doctoral Program in Cognitive Neuroscience. On a cold morning in 2005, Demis Hassabis stood gazing out the window of his office at Elixir Studios in central London and made up his mind. He was already a successful entrepreneur in his early twenties, running his own game company with dozens of employees, and the games he created had sold millions of copies. But there was always an unquenchable thirst inside him.

He felt it was time to return to the fundamental question he had asked as a twelve-year-old boy, sitting in front of a chessboard, wondering 'What is thinking?' and buying his first computer. The game industry at the time was increasingly consumed by flashy graphics and cheap thrills, and Hassabis's dream of creating 'virtual characters with intelligence' kept running into technical limitations and commercial logic. He brushed aside the objections of those around him and shut down his company.

Then he slung a single bag over his shoulder and walked onto the campus of University College London (UCL). Hassabis's decision to study cognitive neuroscience rather than computer science for his pursuit of artificial intelligence was both deeply strategic and philosophical. He had noticed that AI technology at the time relied on nothing more than mathematical calculations and statistical probabilities.

To create true intelligence, the kind of general artificial intelligence (AGI) that adapts to new situations, learns on its own, and produces creative solutions just as humans do, he believed one first had to understand the human brain, the only entity in existence that already performs these functions flawlessly. UCL, one of London's most storied universities, boasted the world's foremost authority in neuroscience, the field dedicated to mapping the brain. There, he set aside the glamorous title of CEO and became a late-starting graduate student, sitting in the library early each morning and poring over thick neuroscience textbooks.

During his doctoral program, Hassabis was not content with simply earning a degree. He dug into the fundamental principles of how different regions of the brain communicate with one another and how the information we see and hear gets converted into knowledge. He was especially drawn to how humans remember past experiences and how those memories serve as the basis for envisioning future events.

This work later became the foundation of 'reinforcement learning' and 'experience replay,' the core algorithms behind the AI that DeepMind would build. He did not view the brain merely as a biological organ but as a kind of 'algorithmic machine,' the most efficient information processor in the universe. These four years of academic immersion marked the decisive turning point that would place him in the most unique position in the history of artificial intelligence. Advisor Eleanor Maguire. The person who greeted Hassabis when he knocked on the door of the UCL lab was Professor Eleanor Maguire.

Professor Maguire was the researcher behind the world-famous 'London taxi driver study.' She had scanned the brains of taxi drivers who memorized London's labyrinthine streets and discovered that the hippocampus, the region responsible for spatial memory, was physically larger in these drivers than in the general population. When Hassabis joined Maguire's research team, he was electrified by the realization that the brain is not a passive warehouse for storing information but an active engine that constantly draws maps and runs simulations.

Professor Maguire valued Hassabis's extraordinary focus and his programming skills, and the meeting of these two minds ignited a tremendous spark where brain science and computer science converged. Under Maguire's guidance, Hassabis began an intensive study of the hippocampus, the structure nestled deep within the brain. The hippocampus gets its name from its resemblance to a seahorse, and it serves as the central headquarters for 'episodic memory,' the system that lets us recall what we ate yesterday or where we went on vacation last summer.

Together with Professor Maguire, Hassabis investigated how the brain connects these countless memory fragments and reconstructs them into a single vivid scene. In the lab, he analyzed vast amounts of data, from brain imaging scans of taxi drivers to case studies of patients who had lost their memory, while simultaneously working to translate the brain's operating principles into mathematical models.

Professor Maguire taught Hassabis scientific rigor: no matter how elegant a hypothesis, it means nothing unless it is proven through meticulous experiments and data. Together with his advisor, he stayed up late into the night designing experiments, scanning subjects' brains with functional magnetic resonance imaging (fMRI), and tracking the subtle signals firing within.

What Hassabis absorbed during this period was the perspective of systems neuroscience: the understanding that no single brain region works alone, but that multiple regions collaborate like an orchestra to produce intelligence. Through his collaborative research with Professor Maguire, Hassabis grew from a mere engineer into a scientist who could read the brain's blueprints, and the papers the two published together appeared in the world's top scientific journals, drawing enormous attention from the academic community. Applying the Brain's Principles to AI Architecture. On the wall above Hassabis's desk in the lab hung a famous quote from the physicist Richard Feynman.

'What I cannot create, I do not understand.' This sentence captured Hassabis's fundamental approach to studying the brain. As he examined the phenomena that neuroscientists had uncovered, he constantly asked himself one question.

'What would it look like if I implemented the brain's functions in computer code?' For him, brain research was a process of reverse engineering, aimed at designing a perfect artificial intelligence. Observing how the brain learns by adjusting the strength of connections between neurons, he found inspiration for how to tune the 'weights' in artificial neural networks.

He noticed that the hippocampus replays the day's experiences at high speed during sleep. If you applied this to a computer, the math was clear: an AI could become far smarter by repeatedly learning from past data instead of needing fresh data at every moment. This became the prototype of the 'experience replay' technique that DeepMind's AI would later use when learning to play games.

By observing how the brain operates, Hassabis imagined the 'architecture' an artificial intelligence should have. Not a system that simply executes commands, but a structure in which multiple modules interact, set their own goals, and build an understanding of the world, just as the brain does. Hassabis often told his fellow researchers:

'True AI should resemble the way the human brain solves problems. That is the most elegant and efficient path.' Even while writing his doctoral thesis, he kept jotting these ideas down as code on scraps of paper.

How the brain perceives space, how it forms abstract concepts, how it predicts an uncertain future: all of these found their way into his emerging blueprint for artificial intelligence. As Feynman had said, Hassabis harbored the ambition to build intelligence himself in order to fully understand it. His doctoral program at UCL gave him a deep conviction about the nature of intelligence, and he was now ready to prove that conviction to the world by founding the tool that would make it possible: DeepMind.

The Hippocampus: The Link Between Memory and Imagination. A Groundbreaking Discovery That Patients with Hippocampal Damage Cannot Imagine the Future. In early 2007, Hassabis published experimental results that stunned the world. He gathered patients suffering from amnesia alongside healthy volunteers and posed them a very particular question: 'Imagine you are going to the beach tomorrow morning.

Can you describe in great detail what you would see and what would happen there?' We take it for granted that we can imagine the future, but the responses from patients with hippocampal damage were shocking. They could not take a single step into the future.

They said things like 'Well, there would be the sea, I suppose... and sand. But that is all. I can't picture anything,' visibly distressed. The screen inside their minds was a blank sheet of paper. This experiment completely overturned the conventional understanding that humanity had held about memory and imagination. Until then, scientists had assumed that memory was for the 'past' and imagination was a separate faculty meant for the 'future.'

But Hassabis proved that a person who cannot remember also cannot imagine. The hippocampus was not merely a diary that records the past; it was the 'mind's workbench,' pulling out fragments of past memory and assembling them into virtual scenarios of the future. Through this discovery, Hassabis became certain that the core of intelligence lies in prediction and simulation.

The reason we remember the past, he found buried deep in the brain, is a law of survival: to better prepare for what lies ahead. Hassabis analyzed the patients' short, fragmented sentences and felt anew the greatness of human intelligence. When we imagine 'having lunch with a friend tomorrow,' our brain recombines countless past memories in the blink of an eye to construct a virtual reality. Hassabis wanted to transplant this ability into artificial intelligence.

What if an AI did not just process current data but imagined the future based on past experience and simulated the best possible choices in advance?

This groundbreaking discovery was published in Nature, and media around the world began to take notice of Hassabis, calling him 'the young genius who revealed that memory and imagination are one and the same.' Development of Episodic Memory and Scene Construction Theory. Hassabis took his research on hippocampal-damage patients a step further and proposed an innovative theory called 'Scene Construction.' Analyzing why patients with damaged hippocampi could not imagine the future, he noticed that they were not simply unable to remember events; they could not construct a three-dimensional space inside their heads.

When we remember or imagine something, our brain first sets up a spatial backdrop, like a stage. Where the table sits, where the window is, which direction the sunlight comes from: a three-dimensional background. Hassabis discovered that the hippocampus acts as the construction foreman who builds this '3D virtual stage.'

According to this theory, every experience (episode) we go through is placed on that stage. Just as actors cannot perform without a stage, if the spatial backdrop is not constructed inside the brain, neither memory nor imagination is possible. Using fMRI, Hassabis visually demonstrated that the hippocampus fires intensely when people imagine something. He took this as a critical clue for AI design. His vision was that artificial intelligence should not merely list words or images but should grasp the world as a three-dimensional structure and predict its own actions within it.

Scene Construction theory served to unify memory theories that had long been considered classics in psychology and neuroscience. Hassabis argued that 'memory is not playback but reconstruction.' Every time we recall the past, the brain builds a new stage and arranges the pieces of memory on it.

This flexible way of thinking later became the driving force behind DeepMind's AI understanding the space of a Go board and calculating win probabilities through countless simulated games. By explaining the brain's complex mechanisms through the concepts of 'scene' and 'construction,' Hassabis had found the North Star for where artificial intelligence needed to go.

'The Simulation Engine of the Mind.' As he neared the end of his doctoral program, Hassabis crystallized one grand concept that ran through all of his research: the idea that 'the brain is a Simulation Engine of the Mind.' He believed that the reason humans have dominated every other species on Earth and built civilizations is that they could run tens of thousands of simulations inside their heads before ever leaping into real danger.

Before a primitive hunter set out, the act of imagining 'Will a lion appear if I go into that forest? Will I be safe if I hide behind that rock?' was itself the essence of intelligence. The brain was a sophisticated simulator that lives through virtual futures in advance and selects the safest and most rewarding path. For Hassabis, this concept was the ultimate goal that artificial intelligence had to reach.

The AI of that era was nothing more than a calculator that performed well within given rules, but Hassabis dreamed of an AI that simulates the world on its own. If an artificial intelligence could understand the physical laws and causal relationships of the real world and predict in advance what consequences its actions would bring, that would be one step closer to human intelligence. Watching the hippocampus and the prefrontal cortex cooperate to run this engine, he puzzled over how to embed this 'simulation loop' into the architecture of an artificial neural network.

This 'Simulation Engine of the Mind' concept later evolved into the technology that became synonymous with DeepMind: model-based reinforcement learning. An AI builds its own model of the world and then exercises imagination within that model to find answers. In his small London lab, staring at fMRI images, Hassabis was already picturing a distant future in which AI would simulate treatments for complex diseases and discover solutions to climate change.

He was already running his own brain's engine, imagining the revolution that DeepMind would bring. Selected as One of Science Magazine's '10 Breakthroughs of the Year' in 2007. Influential Papers Published in Nature, Science, Neuron, PNAS, and Other Top Journals. Demis Hassabis's academic debut was dazzling. Papers bearing his name poured out in a flood, appearing in world-class journals that most doctoral students would be lucky to publish in even once over an entire career. Nature, Science,

Neuron, and the Proceedings of the National Academy of Sciences (PNAS), journals that scientists regard as holy grails, competed to publish Hassabis's findings. He shed the label of 'former game developer' and established himself almost overnight as a rising star in neuroscience. His papers carried outsized influence because the questions they asked ran so deep.

He did not stop at analyzing one narrow function of the brain. Instead, he explained the grand operating principles of the human mind, memory, imagination, and intelligence, as a single organic system. The academic community described him as 'a rare talent who possesses the sharp logic of a computer scientist and the deep insight of a neuroscientist at the same time.' His citation counts grew exponentially, and invitations to lecture poured in from leading universities around the world.

The papers he published in 2007 completely changed the paradigm of memory research. His thesis, 'Memory is not merely a record of the past but a tool for the future,' redirected decades of memory research. The academic achievements and connections he built during this period later became a reservoir of trust when he founded DeepMind, enabling him to attract not only Silicon Valley capital but also the finest minds in academia.

By establishing scientific authority that cut to the essence of intelligence, Hassabis proved on his own terms that he was qualified to lead the coming revolution in artificial intelligence. Proving Through fMRI That Memory Recall and Imagination Share the Same Neural Mechanism. The most decisive evidence from Hassabis's research came from experiments using fMRI (functional magnetic resonance imaging). He had people lie inside the large cylindrical fMRI machine and gave them two tasks.

The first was 'Remember what happened last Christmas.' The second was 'Imagine what will happen next Christmas.' While the machine hummed and captured changes in blood flow through the brain, an astonishing scene unfolded on the monitor. The brain regions that lit up when recalling the past and those that lit up when imagining the future were nearly identical, as if reflected in a mirror.

This was a definitive proof that memory and imagination share the same neural mechanism. From the brain's perspective, recalling the past and imagining the future are fundamentally the same operation. A specific network centered on the hippocampus activates and recombines stored information. Based on this data, Hassabis arrived at a definition: 'Intelligence is the ability to simulate future possibilities using past experience as raw material.'

This discovery was honored as one of the '10 Breakthroughs of the Year' by Science magazine in 2007. The achievement gave Hassabis enormous confidence. He had confirmed how the brain converts intangible imagination into tangible intelligence.

The remaining challenge was to translate this seemingly magical process into digital code. Looking at the fMRI images, he was certain: just as the brain gathers memory fragments to paint the future, an AI could gather fragments of data to simulate the world.

By the time he finished his doctoral program, Hassabis had no reason to remain at the university. He had read the mind's blueprints thoroughly enough. Now, carrying those blueprints, he was ready to step once more into the rough world of business and build the machine that would change everything.

fMRI brain scan image, hippocampal region activation