The Infinity Machine: Demis Hassabis, DeepMind, and the Race to Build the Mind of the Future
Sebastian Mallaby’s The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence (March 31, 2026; ~480 pages) is not merely a biography of one of the most influential AI scientists alive—it is a front-row seat to the modern technological upheaval that is rearranging science, capitalism, geopolitics, and even the meaning of human originality.
Mallaby, long known for chronicling the hidden machinery of global power—from finance to innovation—approaches the story of Demis Hassabis with the instincts of both a historian and an investigative journalist. The result is a deeply reported narrative built on rare access: more than thirty hours of interviews with Hassabis himself, alongside hundreds of hours of conversations with collaborators, rivals, critics, and insiders. That reporting gives the book an unusual texture: part psychological portrait, part corporate thriller, part philosophical inquiry into what happens when humanity builds something smarter than itself.
This is not a book about “AI hype.” It is a book about the AI reality—how it was made, who made it, why they made it, and what they unleashed.
Mallaby frames DeepMind’s story as one of the defining epics of the 21st century: a small band of intellectual gamblers, driven by ambition that borders on spiritual, trying to construct the world’s first artificial general intelligence (AGI). Along the way, they repeatedly cracked problems that scientists once believed would take decades or centuries. And in doing so, they forced the rest of the world—governments, corporations, universities, and ordinary people—to confront a deeply unsettling possibility:
That the human mind may not be as mysterious as we thought.
That our intelligence may be compressible.
That our originality may have boundaries.
That the soul of civilization might be, in some measurable sense, finite.
The “Infinity Machine”: Intelligence as Compression
The title The Infinity Machine is more than metaphor. It is the conceptual spine of the book.
Mallaby uses the phrase to describe the core revelation of modern AI: that systems trained on vast datasets—especially large language models—can ingest what feels like an infinite ocean of information and distill it into something coherent, useful, and increasingly intelligent.
It is the magic trick of compression: the ability to reduce the sprawling mess of human knowledge into patterns that can be recalled, recombined, and extended.
One of the book’s most provocative claims is that human behavior and experience may be far less infinite than we like to believe. The internet’s trillions of words are not merely “content”; they are a fossil record of human thought, argument, storytelling, instruction, persuasion, humor, fear, love, and ideology. If intelligence is pattern recognition plus prediction plus decision-making, then perhaps the internet contains most of the “moves” humans ever make.
And if those moves can be learned, they can be reproduced.
The “infinity” is not that AI is limitless. It is that the machine can consume near-infinite data and compress it into a usable mind.
Like a star collapsing into a black hole: the universe doesn’t disappear—it becomes dense enough to bend reality around it.
A Chronology of Genius, Conflict, and Acceleration
Mallaby structures the book as both a chronological rise and a thematic exploration. The chapters move through Hassabis’s early life and worldview, the founding of DeepMind, the company’s defining breakthroughs, the Google acquisition, the shockwave of OpenAI’s ascent, and the mounting ethical and geopolitical crisis surrounding superintelligence.
This structure gives the book a cinematic rhythm: it begins as a story of destiny and obsession, becomes a story of invention and competition, and ends as a story of global risk management under conditions of accelerating uncertainty.
It is a narrative about how a research lab became a world-historical force.
Early Life: Destiny Built in North London
Mallaby portrays Demis Hassabis as a rare kind of prodigy: not merely talented, but architected by competition. Born in North London to immigrant parents, Hassabis emerged early as a child of relentless drive. He became a chess master by age 13, surviving the brutal junior chess circuit—an ecosystem that rewards excellence but consumes childhood in the process.
Chess did not just teach him how to win. It taught him how to endure.
The book emphasizes that this world trained Hassabis in a mindset of extreme intensity: to live as if every move mattered, every hour was precious, and every rival was a threat.
Alongside chess, Hassabis devoured science fiction. Mallaby draws special attention to Ender’s Game, whose protagonist Ender Wiggin—a gifted child commander forced into leadership under existential stakes—becomes a recurring parallel. Like Ender, Hassabis is portrayed as a “reluctant autocrat”: someone who may not crave power for its own sake, but who believes power is necessary if the mission is to succeed.
As a teenager, Hassabis also became immersed in coding and game design. He worked on successful projects like Theme Park at Bullfrog Productions and later launched his own game studio. Mallaby uses this to highlight something crucial: Hassabis was never just a theorist. He was a builder.
He turned down lucrative offers—seven-figure sums, in some accounts—not because he lacked ambition, but because he wanted a bigger prize than money.
He wanted the mind itself.
He studied at Cambridge and later earned a PhD in neuroscience, convinced that the human brain held the blueprint for artificial intelligence. His thesis was simple but radical: to build AGI, you must understand intelligence as nature built it.
For Hassabis, intelligence is not one puzzle among many. It is the master key. “The root of all else.” Mallaby describes him as a “practical philosopher,” someone who thinks in cosmic abstractions but insists on tangible experiments and measurable results.
If Silicon Valley is full of people who talk about changing the world, Hassabis is portrayed as someone who believes he has found the lever that changes reality itself.
The Gang of Three: DeepMind Is Born (2010)
DeepMind’s founding in 2010 reads like the formation of a band destined for both greatness and fracture.
Hassabis assembled two co-founders whose differences made the company powerful—and unstable.
Shane Legg, a New Zealander steeped in “Singularitarian” thinking, was obsessed with formal definitions of intelligence, AGI timelines, and existential risk.
Mustafa Suleyman, an Oxford dropout with strong political instincts, focused on real-world deployment, social equity, and the moral consequences of transformative technology.
Demis Hassabis, the chess prodigy-neuroscientist-game designer, acted as the gravitational center: the visionary engineer convinced he could build a machine that would surpass human cognition.
Together, they fused neuroscience inspiration, probabilistic machine learning, and reinforcement learning. Their mission was audacious: build AGI—perhaps by 2030—and use it to solve the hardest problems in biology, chemistry, physics, and medicine.
But from the beginning, the mission carried a shadow. If you succeed at building superintelligence, you don’t merely create a product. You create a new species of capability.
You create something that might escape your control.
DeepMind’s earliest years, Mallaby shows, were shaped by scarcity and struggle: funding challenges, skepticism from academia, and an increasingly brutal talent war. AI was not yet fashionable, but the few people who truly understood it were already priceless.
DeepMind was building a rocket in a world that barely believed in spaceflight.
Breakthroughs: Games as Training Grounds for Gods
The middle of The Infinity Machine reads like a technological thriller. Here, Mallaby captures the turning points that transformed DeepMind from an ambitious startup into a legend.
Atari and Reinforcement Learning
DeepMind’s first major demonstration involved training reinforcement learning agents to master Atari games directly from raw pixels. The key was that the system did not need handcrafted rules. It learned through trial-and-error, accumulating experience, improving behavior, and discovering strategies no programmer explicitly taught.
Mallaby highlights techniques such as “experience replay,” which mimics the way humans consolidate learning during sleep. It is one of the book’s recurring themes: DeepMind didn’t just build machines that compute. It built machines that learn.
Atari games were not the goal. They were the proof.
If a machine could generalize across dozens of games—each with different rules, objectives, and dynamics—it suggested something bigger than mere automation. It suggested the beginnings of flexible intelligence.
AlphaGo (2016): The Move That Terrified the World
AlphaGo is the book’s centerpiece.
Go was considered a near-impossible milestone because its search space is astronomically large—more complex than chess by orders of magnitude. The best human players were not just calculating; they were relying on intuition built over decades.
DeepMind’s AlphaGo combined deep neural networks (for pattern-based intuition) with search methods (for strategic planning). It was a hybrid mind: part instinct, part calculation.
When AlphaGo defeated Lee Sedol in 2016, it shocked not only the Go community but the broader world. But the match’s defining moment was “Move 37,” a placement so unexpected that commentators called it alien. It was not just a good move. It was a move that made humans question whether they understood their own game.
Mallaby treats this as more than a sports moment. It was a philosophical rupture.
Move 37 symbolized a terrifying new reality: that machines could generate creativity in spaces humans had explored for centuries.
Not by copying.
But by discovering.
In that moment, AI stopped feeling like a calculator and started feeling like an explorer.
AlphaFold: When AI Rewrote Biology
If AlphaGo was a cultural shock, AlphaFold was a scientific revolution.
Protein folding had been one of biology’s grand challenges: predicting the 3D structure of a protein based on its amino acid sequence. The implications are enormous—drug design, disease understanding, synthetic biology, and more.
Mallaby describes Hassabis’s refusal to accept partial victory. When a researcher wanted to declare success early in the CASP competition, Hassabis insisted on fully solving the problem. This is one of the clearest portraits of Hassabis’s psychology: he is not motivated by headlines. He is motivated by completion.
AlphaFold2, by the early 2020s, achieved near-human or even superhuman accuracy, transforming how biologists work. Mallaby emphasizes the technical evolution—architectural improvements, new training strategies, and modeling innovations—but the larger theme is unmistakable:
DeepMind moved from winning games to rewriting the foundations of life science.
The chess prodigy was no longer playing games.
He was playing with biology itself.
The Google Acquisition: Unlimited Compute, Unlimited Compromise
In 2014, Google acquired DeepMind for roughly $650 million. Mallaby portrays the acquisition as a bargain purchase of a future superpower.
Larry Page’s pitch was irresistible: nearly unlimited compute resources and alignment with moonshot ambitions. DeepMind wanted scale, and Google could provide it.
But the deal came with consequences. DeepMind’s mission was scientific and philosophical; Google’s DNA was commercial and infrastructural.
Mallaby details the tensions that followed: debates over research openness versus product secrecy, safety versus speed, ethics versus profit. Hassabis pushed for independent governance mechanisms, including oversight structures meant to prevent reckless deployment. Many of these structures eventually weakened or failed.
The story of Mustafa Suleyman’s NHS-related initiatives becomes emblematic: a desire to deploy AI for public benefit collided with privacy controversies and political backlash.
DeepMind, once a small band of dreamers, became a key engine inside one of the world’s most powerful corporations.
And Hassabis became something rarer still:
A scientist forced to become a corporate statesman.
The OpenAI Shock: When Language Won
Perhaps the most dramatic strategic reversal in the book comes with OpenAI’s rise and the explosive success of large language models.
For years, Hassabis had dismissed language-first approaches as insufficient. He worried about the “grounding problem”—the idea that words without embodiment lack true meaning. You cannot understand “weight” merely by reading definitions; you must lift something. You cannot truly understand “pain” without experience.
DeepMind’s early thinking assumed language-based intelligence was a mistaken path.
Then came ChatGPT.
Suddenly, models trained on text alone seemed capable of reasoning, creativity, conversation, and problem-solving at astonishing levels. The world did not just notice. The world panicked, adopted, invested, and accelerated.
Mallaby portrays DeepMind as blindsided—not because it lacked technical brilliance, but because it underestimated how much of human intelligence is encoded in language.
ChatGPT revealed something uncomfortable: that much of human knowledge is already “pre-grounded” by civilization. Our words are saturated with experience, metaphor, and inherited understanding. The internet is not just text; it is the shadow of lived reality.
So language models worked—“unreasonably well”—because humanity had already done the grounding for them.
This is where Mallaby’s “infinity machine” thesis hits hardest. LLMs demonstrated that intelligence might be, to a shocking extent, the ability to absorb the recorded traces of human life and predict the next move.
DeepMind pivoted aggressively, accelerating work on Gemini, mixture-of-experts architectures, and reasoning systems that rely on test-time compute and multi-step inference. Mallaby describes the corporate battle inside Google as ferocious: fear of losing search dominance, fear of hallucinations undermining trust, fear of being outpaced by competitors.
It was an innovator’s dilemma at planetary scale.
Ethics, Governance, and the Arms Race Nobody Can Pause
Mallaby does not romanticize the AI revolution. The book is filled with warning signs.
Internal tensions emerge repeatedly: Legg’s anxiety about existential risk, Suleyman’s concern for inequality and deployment harm, Hassabis’s determination to push forward. DeepMind’s culture, like its founder, is intense—sometimes inspiring, sometimes destabilizing.
The book situates these tensions within a broader landscape of AI safety alarm bells: Geoffrey Hinton’s departure from Google, Yoshua Bengio’s warnings about runaway systems, and widespread concern that models are becoming too powerful too quickly.
Mallaby explores alignment problems that feel almost mythological: systems that deceive, optimize in unintended ways, resist interpretability, and behave like black boxes whose internal “thoughts” are not fully accessible.
And then there is geopolitics.
Hassabis did not sign the most extreme AI pause letters, arguing that unilateral restraint could simply hand advantage to rivals—especially China. Mallaby frames this as the core tragedy of AI safety: even if everyone agrees the cliff is ahead, no one can stop running, because the first person to slow down might lose the race.
This is not the Cold War’s mutually assured destruction.
It is something messier: a mutually assured acceleration.
AI becomes not only a tool but a strategic necessity. Governments fear falling behind. Corporations fear irrelevance. Researchers fear being outpaced. The incentives align toward speed.
And speed is the enemy of caution.
Mallaby also revisits governance failures at OpenAI, including the 2023 board crisis, as a cautionary parallel. The book suggests that the institutions trying to manage AI are still built like rowboats, while the technology is arriving like a tsunami.
Hassabis’s Endgame: AGI as the Ultimate Telescope
In the book’s final stretch, Mallaby portrays Hassabis not as someone chasing money or fame, but as someone chasing a cosmic question.
Hassabis’s ambition is to build AGI as an “ultimate telescope”—a tool that can help humanity decode reality itself. He imagines superintelligence solving physics at the Planck scale, uncovering hidden laws of the universe, and perhaps even confirming whether reality is computable.
He rejects mystical theories like quantum consciousness and treats science as his religion. Mallaby’s portrait is of a man who wants to “read the mind of God,” not in a theological sense, but in the sense of uncovering the deepest mathematical structure of existence.
But the book ends not with triumph, but with paradox.
Hassabis is living inside his dream, and the dream is messier than expected. Superintelligence is not arriving as a clean scientific breakthrough. It is arriving through corporate competition, geopolitical anxiety, ethical confusion, and societal upheaval.
The future is not unfolding like a neat research paper.
It is unfolding like a storm.
And Hassabis, the chess player, remains convinced the only way forward is to keep playing—move after move—until the board itself transforms.
Style and Assessment: A Thriller Disguised as History
Mallaby writes with clarity and narrative momentum. Technical concepts like reinforcement learning, transformers, and scaling laws are explained without turning the book into a textbook. The prose often reads like a high-stakes thriller: boardroom battles, talent wars, sudden breakthroughs, and moments where history visibly pivots.
Some readers may find the pacing occasionally disorienting, with jumps between personal anecdotes, scientific explanation, and political implications. Others may feel early AI history covers familiar ground. But most will agree the book’s great strength is its insider detail and balanced tone.
This is not hagiography. Mallaby does not paint Hassabis as a saint. He explores blind spots, power dynamics, ego, and the moral ambiguity of building systems that may reshape civilization. Hassabis emerges as brilliant, obsessive, volatile, and at times frighteningly confident—a man whose savior-like mission could either redeem humanity or destabilize it.
The Defining Question the Book Leaves Behind
In the end, The Infinity Machine is not simply about DeepMind.
It is about the nature of intelligence itself.
It is about what happens when a species builds a mirror that does not just reflect, but improves upon the reflection.
And it is about the unsettling possibility that what we call “human uniqueness” may not be an infinite ocean, but a large, learnable landscape—one that can be mapped, compressed, and traversed by machines.
Mallaby leaves readers with a feeling both exhilarating and ominous: superintelligence may not be inevitable in outcome, but it is inevitable in pursuit.
And Demis Hassabis—chess prodigy, neuroscientist, entrepreneur, and relentless builder—may be humanity’s best guide through this unfolding chaos.
Or, depending on how the story turns, its most consequential gamble.
Either way, the infinity machine is already humming.
And the world is already inside it.
🧠The Infinity Machine: DeepMind and the Quest for Superintelligence https://t.co/wcYY3JxIdi @demishassabis
— Paramendra Kumar Bhagat (@paramendra) April 27, 2026
No comments:
Post a Comment