Demis Hassabis, Father of Google's Artificial Intelligence

Part 4. DeepMind

9 The Birth of DeepMind

Kim Kyung-ran, Kim Kyung-jin

The Gatsby Computational Neuroscience Unit at University College London (UCL) was a fierce battleground where some of the world's brightest minds gathered to decode the secrets of the human brain through mathematics. Founded by Geoffrey Hinton, it was where Demis Hassabis was finishing his doctoral work in neuroscience. Shane Legg, a tall, quiet New Zealander whom Hassabis kept running into in the hallways, was a figure who both fit the atmosphere of the place and stood oddly apart from it.

Legg had already spent time at the IDSIA research institute in Switzerland, digging deep into theoretical AI models under the supervision of Juergen Schmidhuber. In academic circles at the time, the phrase 'artificial intelligence' was practically taboo. Most researchers hid behind safer labels like statistics or machine learning, devoting themselves exclusively to narrow AI that solved specific problems. Legg was different.

He was one of the few people who spoke the words 'AGI (artificial general intelligence)' without fear, the very phrase Hassabis had been carrying close to his heart. Every lunchtime, the two would sit on a bench near the lab or in a London pub, debating for hours how human intelligence could be defined mathematically and what the world would look like if machines could think like people. Where Hassabis favored an intuitive, neuroscience-driven approach, Legg insisted on logical and mathematical rigor.

Their meeting was like combining fire and ice; each compensated for what the other lacked. Legg would later play a central role in warning about the dangers of superintelligence for humanity and in building safety mechanisms against it.

The person who injected real-world energy into this intellectual partnership was Mustafa Suleyman. Suleyman was not a scientist. He was a close friend of George Hassabis, Demis's younger brother, and had been a 'neighborhood kid' who came and went from the Hassabis household as if it were his own since childhood.

He had studied philosophy and theology at the University of Oxford before dropping out. From his teenage years, he had been an activist deeply engaged in social issues, serving as an advisor to Islamic organizations and to the Mayor of London. While Hassabis and Legg explored the nature of intelligence inside the lab, Suleyman was thinking about the outside world: how this technology could transform society, and how to gather the people and capital needed to make this enormous dream real. Suleyman reminded Hassabis that 'narrative' and 'persuasion' mattered just as much as scientific vision.

He argued that DeepMind had to become more than a research lab; it had to be a company that changed the world. He would later demonstrate his business acumen by leading the acquisition negotiations with Google and spearheading the DeepMind Health project. Then another figure joined the group: David Silver, a Cambridge classmate of Hassabis who had lived through the rise and fall of Elixir Studios alongside him.

After Elixir shut down, Silver had returned to academia and carved out a singular position in the field of reinforcement learning. Hassabis sensed that reinforcement learning, the brain's own method of learning, would become the core algorithm of DeepMind, and he could not let his old friend, now the field's foremost authority, slip away. Silver's arrival was the final puzzle piece that completed DeepMind's technical foundation.

He would later astonish the world as the lead researcher behind AlphaGo's victories. In November 2010, the team rented a small office near Russell Square in London and hung a sign reading 'DeepMind Technologies.' The startup scene at the time was consumed with social media and mobile app development. Silicon Valley investors sneered: 'You could build a photo-sharing app and make money in six months. Why bother with AI research that might take twenty years?'

Hassabis and his co-founders did not waver. Their mission was clear. 'Solve intelligence.' And then, 'Use it to solve

everything else.' This was less a corporate motto than a manifesto for scientific revolution. Like the Apollo program that sent humans to the moon in the 1960s, they drew up an ambitious plan to concentrate the best talent and capital in one place and conquer the uncharted territory of intelligence.

Instead of pitching revenue models to venture capitalists, Hassabis talked about a future in which artificial intelligence would cure diseases, combat climate change, and drive new scientific discoveries. Visionary investors like Peter Thiel were drawn to what looked like a reckless dream precisely because of its sheer scale. The birth of DeepMind was not just another tech company opening its doors. It was a signal flare announcing the end of the AI Winter and the arrival of a new spring.

The unusual combination of Hassabis, Legg, and Suleyman fused scientific rigor, philosophical depth, and practical drive into a single engine aimed at the towering goal of AGI. Merging Academia and Startup Culture. The question Demis Hassabis wrestled with most deeply when founding DeepMind was: 'What kind of organization should this be?' His years in academia at Cambridge and UCL had given him a bone-deep understanding of the strengths and weaknesses of university research labs.

Universities were places where the freest imagination in human history was possible, what people call 'Blue Sky Thinking.' They were the only spaces where you could ask fundamental questions without being shackled to immediate profits or deliverables. But universities were also slow.

Professors got buried in paperwork chasing grants. Researchers were trapped inside the silos of their narrow specialties, often with no idea what the colleague next door was working on. The game companies and startups Hassabis had experienced since his teenage years were the opposite. They had the speed and energy of a flat-out sprint toward a goal, and the desperation to produce results sparked innovation.

But the pace was too short for deep research. Hassabis dreamed of a 'hybrid organization' that captured only the strengths of both worlds. His model was Bell Labs, the American research institution that gave birth to the transistor, the laser, and information theory in the mid-twentieth century. An organization where the best minds gathered and debated freely, like Bell Labs, yet moved with the clear objectives and deadlines of a start-

up. That was the blueprint for DeepMind. The approach to hiring was different from the start. Hassabis personally sought out PhD-level talent in machine learning, neuroscience, mathematics, and physics from around the world.

It was not yet the era when Big Tech companies like Google and Facebook were vacuuming up every AI researcher in sight, but luring promising academics to a shaky startup was no easy task. Hassabis sold them not money but mission. 'Aren't you stuck at your university, chasing publication counts, solving small problems? Come with us and tackle intelligence itself. No worrying about grant funding. No teaching obligations.'

Those words pierced the hearts of brilliant minds thirsting for intellectual challenge. He was not just hiring engineers who could code well; he was assembling scientists who wanted to build a thinking machine. When it came to designing the organizational culture, Hassabis enforced 'convergence.'

He was vigilant against researchers retreating into the silos of their own specialties. He started by rearranging the office layout itself. Neuroscientists sat next to computer scientists. At set times each day, everyone gathered for tea to share their research. An AI playing a brick-breaking game would be shown, a neuroscientist would explain how it connected to the brain's dopamine system, a mathematician would formalize the explanation into equations, and an engineer would turn it into code.

It was in this environment that DeepMind's core idea was born: the fusion of reinforcement learning and deep learning. Fields that would never have crossed paths in academia, their members attending separate conferences, began to mix inside the crucible of DeepMind. Hassabis also transplanted the characteristic 'velocity' of startups into the research process.

Unlike typical academic research that stretches over years, DeepMind broke projects into short cycles and built rapid prototypes for validation. It was the Silicon Valley maxim 'Fail fast' applied to scientific research. But it was not a blind race for speed.

He guaranteed researchers the freedom to publish papers. This was meant to reassure academics who feared that corporate secrecy would keep their findings from reaching the world. There was one condition, though.

Before writing a paper, researchers first had to prove the technology internally. This instilled a healthy competitive spirit and laid the groundwork for DeepMind to later churn out cover stories in top journals like Nature and Science. The hybrid culture stumbled through plenty of trial and error in the early days. Free-spirited academics chafed at corporate discipline, and results-oriented executives found it hard to wait on basic research that showed no immediate progress.

Hassabis appointed himself the buffer between the two sides. He encouraged researchers by telling them, 'What we are doing is science,' while assuring investors, 'This science will eventually become an enormous business.' DeepMind was a strange space where academic intellect and corporate ambition coexisted.

The heat of algorithm debates running until three in the morning. Whiteboards covered in equations. A high-end gaming console humming in one corner. On this distinctive cultural soil, AlphaGo, the most powerful artificial intelligence in human history, was germinating. What Hassabis had built was not just a company but a new kind of 'factory' that manufactured intelligence. Hassabis's day is split into two shifts. During the day he runs meetings and manages operations as CEO. At night, in what he calls his 'second shift,' he reads papers and organizes ideas.

This pattern, stretching until three in the morning, is the backdrop to an academic record of more than 2,000 research papers and an h-index of 83.

DeepMind London office