Demis Hassabis and the quest for a Universal Learning Machine

On a single afternoon in Seoul in 2016, the history of human intelligence was rewritten by a ghost in a machine. When AlphaGo made "Move 37"-a placement so bizarre and "inhuman" that it stunned the world’s greatest grandmaster into silence-Demis Hassabis wasn't looking at the board. He was looking at the future of the scientific method.

Hassabis, the co-founder and CEO of Google DeepMind, views games as the foundational language of intelligence. To him, a game is not a toy; it is a highly compressed, high-dimensional simulation of reality. If you can build a machine that can master a game without being told the rules, you haven't just built a player. You’ve built a Universal Learning Machine.

"Intelligence is the most important thing we will ever discover," Hassabis often says. "If we can solve intelligence, we can use it to solve everything else."

Hassabis is a "meta-genius," a man who has successfully navigated three distinct lives: first as a world-class chess prodigy, then as a pioneering video game designer, and finally as the neuroscientist who would go on to build the most influential AI lab in history. In 2024, the path reached its ultimate culmination when he was awarded the Nobel Prize in Chemistry for solving the 50-year-old "protein-folding problem"-a feat that was achieved not by a biologist in a lab, but by an AI that had "learned" the physics of biology.

Today, as DeepMind leads the race toward Artificial General Intelligence (AGI), it is worth examining the "Infinite Game" of Demis Hassabis-and why he believes that the same logic that defeated a Go champion in Seoul is the only thing that can save the human race from itself.

Part I: The Chessboard and the "Theme Park"

The prophecy of the Universal Learning Machine began in London in 1976.

Hassabis was born to a Greek-Cypriot father and a Chinese-Singaporean mother-a background that he describes as a bridge between the analytical West and the holistic East. But his true nationality was mathematical. By the age of four, he was obsessed with chess. By thirteen, he was the second-highest-rated player in the world for his age group, trailing only the legendary Judit Polgár.

"Chess taught me that the board is never what it seems," Hassabis recalls. "It’s a tree of possibilities. You aren't just looking at the pieces; you’re looking at the intent behind the pieces."

But the chessboard was too small for him. He wanted to build the board.

At fifteen, after finishing his A-levels early, Hassabis began working at Bullfrog Productions, the legendary studio led by Peter Molyneux. He was the lead programmer on Theme Park, a simulation game that sold millions of copies.

The "Rosebud" moment of his life happened during the development of Theme Park. Most games at the time were "scripted"-if the player did X, the computer did Y. Hassabis wanted something else. He wanted the park visitors to have their own "intelligence." He programmed them to react dynamically: if you put more salt on the fries, they would buy more drinks. If the drinks were too expensive, they would get angry and leave.

"I realized that I didn't want to program the behavior," Hassabis said. "I wanted to program the system that produced the behavior. I realized that if you could build a system that could learn from experience, it would be more powerful than anything a human could ever hard-code."

Part II: The Reinforcement Learning Revolution

Hassabis left the game industry to complete his PhD in cognitive neuroscience at University College London. He wanted to understand the "biological hardware" of intelligence. He published landmark papers on memory and imagination, proving that the same parts of the brain that allow us to remember the past are the parts that allow us to simulate the future.

In 2010, he founded DeepMind. His pitch to investors (including Peter Thiel and Elon Musk) was short and terrifyingly ambitious:

1. Solve Intelligence.
2. Use it to Solve Everything Else.

He began with Atari games. He didn't tell the AI how to play Breakout or Space Invaders. He simply gave it the raw pixels of the screen and a single command: Maximize the score.

This was the birth of Deep Reinforcement Learning. The AI would play millions of games in a digital loop, failing, learning, and eventually developing strategies that no human had ever imagined. In Breakout, the AI discovered that if it tunneled through the side of the wall, it could trap the ball behind the bricks and score points at a geometric rate.

"It was a moment of pure, raw intelligence," Hassabis says. "The machine wasn't just following rules. It was inventing its own."

Part III: Seoul and the Ghost in the Machine

In 2016, the Infinite Game moved to a hotel ballroom in Seoul, South Korea. DeepMind’s AlphaGo was set to play Lee Sedol, the greatest Go player of the modern era.

Go is a game of near-infinite complexity-there are more possible positions on a Go board than there are atoms in the observable universe. It is a game that requires "intuition" and "feeling." Most experts believed it would be a hundred years before a machine could defeat a human grandmaster.

AlphaGo won 4 to 1.

But the victory wasn't the most important part. The most important part was Move 37 in Game 2.

The AI made a move so bizarre, so "inhuman," that the commentators initially thought it was a mistake. It placed a stone on the fifth line, far away from the center of the action. Lee Sedol stood up and walked out of the room, stunned. He later said he felt like he was playing against a "god."

"Move 37 wasn't a calculation," Hassabis explains. "It was an intuition derived from playing millions of games against itself. The machine had seen a pattern in the chaos that no human in 3,000 years of Go history had ever perceived."

Part IV: AlphaFold and the 50-Year Solution

For Hassabis, AlphaGo was just a "proof of concept." He didn't want to win games; he wanted to solve the "Grand Challenges" of science.

The biggest challenge was Protein Folding. Proteins are the "nano-machines" of life. Their 3D shape determines everything from how a virus infects a cell to how a muscle contracts. For fifty years, biologists had struggled to predict the 3D shape of a protein from its 1D amino acid sequence. It was a problem of combinatorial explosion-there are trillions of ways a protein can fold.

In 2020, DeepMind released AlphaFold 2.

Using a "Transformer" architecture-the same logic that powers LLMs-AlphaFold treated the protein sequence like a sentence and the 3D shape like the "meaning" of that sentence. It solved the problem with "experimental accuracy."

By 2022, DeepMind had released the structures of almost every protein known to science-200 million predictions. It was the "Big Bang" of digital biology. Researchers are now using AlphaFold to develop new vaccines, design plastic-eating enzymes, and discover drugs for diseases that were previously thought "undruggable."

"AlphaFold is the ultimate validation of the DeepMind thesis," Hassabis says. "We used the Universal Learning Machine to unlock the fundamental chemistry of the universe."

Part V: The Stewardship of the End Game

Today, Demis Hassabis is the CEO of Google DeepMind, leading the development of Gemini and the push toward AGI. He is no longer just a scientist; he is a steward of the most powerful technology ever created.

He is acutely aware of the risks. He was one of the first to call for "AI Safety" as a core engineering discipline, and he remains a vocal advocate for international regulation.

"We are building a tool that will be smarter than us," he says. "We have to make sure its goals are aligned with ours. We have to make sure the Infinite Game has a winner, and that winner is humanity."

In 2026, as the "Ghost in the Machine" begins to solve problems in nuclear fusion and materials science, Demis Hassabis remains the calm strategist from the London chess clubs. He knows that the board is still unfolding. He knows that Move 37 was just the beginning.

"The universe is just a very complex game," he says, a slight smile touching his lips. "And we are finally learning how to play."