What happens when a chess prodigy, a video game designer, a neuroscientist, and a Nobel Prize winner all inhabit the same restless mind—and that mind decides to build God? Sebastian Mallaby’s The Infinity Machine: Demis Hassabis, DeepMind, and the Quest for Superintelligence (Penguin Press, March 31, 2026) is not just another tech biography.
You’ve just read about ChatGPT in the news, but have you ever wondered who actually won the race to build the machine that can think—and what it cost them?
The Infinity Machine is best for anyone who loved Walter Isaacson’s The Innovators or Ashlee Vance’s Elon Musk and craves a similarly gripping, character-driven narrative about technological genius; tetch investors, entrepreneurs, and policymakers who need to understand the strategic chess match between Google DeepMind, OpenAI, and Microsoft—and the personalities driving it.
Philosophy and science enthusiasts intrigued by the deepest questions: What is intelligence? Can a machine be conscious? What is our purpose?
Not for readers looking for a quick, surface-level explainer on AI. Mallaby dives deep into neuroscience and computer science, demanding your full attention; those who prefer purely technical manuals without the messy, dramatic stories of ambition, betrayal, and ego; and anyone seeking a definitive answer on whether AI will save or destroy us.
Mallaby presents the dilemma brilliantly but offers no easy conclusions.
Table of Contents
Introduction
The Infinity Machine is the definitive, fly-on-the-wall account of the most consequential scientific gamble of our time, told by a two-time Pulitzer Prize finalist who was given unprecedented access.
Over thirty hours of conversation with Hassabis and more than a hundred interviews with his inner circle, Mallaby has crafted a narrative that is part start-up adventure, part philosophical thriller, and entirely gripping.
Mallaby, the Paul A. Volcker Senior Fellow for International Economics at the Council on Foreign Relations and author of the bestselling The Power Law, brings his signature investigative rigor to the world of algorithms and neurons.
He doesn’t just explain how DeepMind’s AlphaGo defeated Go world champion Lee Sedol in 2016—watched by over 200 million people globally—or how AlphaFold 2 solved a 50‑year‑old “grand challenge in biology” by predicting the 3D structure of nearly all known proteins, earning Hassabis the 2024 Nobel Prize in Chemistry.
Mallaby’s central thesis is far more arresting: the quest for artificial general intelligence (AGI) is a mirror reflecting the best and worst of human nature.
It is a story about the “sweetness of discovery” that drove J. Robert Oppenheimer, the brutal competitiveness of a boy who believed trying his best meant pushing himself “to the point just before death,” and the tragic irony that the very scientists trying to save humanity may be the ones unleashing its potential replacement.
Background
To understand The Infinity Machine, you need to appreciate the context of the early 2010s.
When Hassabis founded DeepMind in 2010, the term “Artificial General Intelligence” (AGI) was considered academic suicide.
Most computer scientists were focused on narrow, symbolic AI—systems with rigid rules that failed when faced with the messy, inductive nature of the real world. The few who dreamed of superhuman intelligence were often dismissed as members of the “Singularity” cult, a “Bay Area coming‑out party for the tech‑inspired philosophy called transhumanism.”
Into this desert of skepticism walked Demis Hassabis, a man who had already conquered chess, video game design, and neuroscience. His audacious premise, drawn from the pages of Gödel, Escher, Bach and his own PhD research on memory and imagination, was that the human brain was the “existence proof” for intelligence.
To truly understand it, one must build it.
Mallaby masterfully sets the stage, showing how the 2012 ImageNet breakthrough—where a deep‑learning system smashed all previous records in computer vision—suddenly made Hassabis’s “crazy” idea look prescient.
The race for AI was on.
The Infinity Machine Summary
Part I: The Genesis of a Genius (Chapters 1‑3)
Mallaby opens not with code but with a soul. We are introduced to a young Demis Hassabis, a “diminutive boy genius” sitting on a telephone book so he could see the chessboard.
We learn of his father’s explosive pressure and the son’s chilling interpretation of “do your best”: “The only way I could know is basically if I pushed myself to the point just before death.” This is the Ender’s Game archetype—a child forged in isolation and pressure, destined to save humanity.
After a Liechtenstein defeat that made him feel he was “wasting our minds,” Hassabis pivoted from chess to computing, devouring Claude Shannon’s work and dreaming of an “infinity machine.”
His time at Bullfrog, working on the god‑game Theme Park and later Black & White, was his informal MBA in AI. He learned that digital characters could learn from slaps and strokes—a primitive form of reinforcement learning.
These chapters establish that Hassabis’s genius was not just raw intellect but a magpie‑like ability to synthesize ideas from neuroscience, philosophy (Kant, Spinoza), and science fiction (Iain Banks’s Culture series) into a coherent, almost spiritual mission.
Part II: The Gang of Three and the Impossible Start‑up (Chapters 4‑6)
The founding story of DeepMind reads like a thriller. Mallaby introduces the “Gang of Three”: Hassabis, the visionary; Shane Legg, the laid‑back New Zealander who coined the term “artificial general intelligence” and had a mathematical framework for measuring it; and Mustafa Suleyman, the brash, street‑smart social activist who wanted to use technology as a “force multiplier” for good.
Their 2010 business plan was rejected by nearly everyone.
Investors saw “enlightenment” as a terrible business model. Mallaby details their desperate pitch to Peter Thiel at a Singularity Summit after‑party, their uncomfortable courtship of the religious investor David Gammon (who demanded they pitch to his 14‑year‑old son), and their eventual salvation by Founders Fund.
This section climaxes with the technical miracle: Vlad Mnih’s Atari agent. Using a “Deep‑Q Network” (DQN) that combined reinforcement learning with a “memory replay” buffer mimicking the human hippocampus, DeepMind’s system taught itself to play Pong, Breakout, and Space Invaders at superhuman levels, simply from raw pixels and score feedback.
As Silver recalled, “People were just blown away. It was a turning point.”
Part III: The God Move and the Race for AlphaGo (Chapters 7‑9)
This is the book’s heart. The Google acquisition looms—a $650 million deal in 2014 that gave Hassabis the resources to build a “Manhattan Project for AI.”
Mallaby brilliantly narrates the creation of AlphaGo, a system designed to conquer Go, a game whose “number of possible board states is estimated to be at least 10¹⁷⁰, way more than the number of atoms in the observable universe.”
The key was “intuition.”
David Silver and his team trained a neural network not to calculate brute force, but to mimic human experts. The result was “Move 37” in Game 2 against Lee Sedol—a move so strange and unprecedented that the world’s top Go player, Michael Redmond, initially removed it from the board, muttering “No, that can’t be right.”
It turned out to be a stroke of creative genius, proving that a machine could be intuitive, even beautiful.
After his defeat, a shell‑shocked Lee Sedol apologized: “I, Lee Sedol, lost, but mankind did not.” Yet, as Thore Graepel chillingly notes later, “At first, it looks harmless. Then it’s just completely dominating. We don’t understand the mechanics…
We just know that it is in control.”
Part IV: The Double‑Edged Sword of Success (Chapters 10‑15)
With fame came fractious politics. Mallaby doesn’t flinch from the internal dramas. He details the “Project Mario” governance fights, where Suleyman tried to spin DeepMind out of Google with an independent ethics board—a battle that ultimately failed.
We see the rise and fall of DeepMind Health, torpedoed by a Daily Mail headline (“GOOGLE HANDED PATIENTS’ FILES”) that, while misleading, exposed the public’s deep distrust of big tech.
Most poignantly, Mallaby covers the tragic ouster of Mustafa Suleyman following a bullying investigation, a messy divorce that highlighted Hassabis’s struggle to balance loyalty with the brutal demands of leadership.
Amidst this chaos, the scientific miracles continued: AlphaZero, which mastered chess and shogi from scratch without human data; and the herculean effort of AlphaFold, which “folded” the shape of 200 million proteins in July 2022 alone.
This work would earn Hassabis the Nobel Prize, but it also revealed a weakness: while DeepMind was curing biology, it was almost blindsided by the rise of large language models (LLMs).
Part V: The ChatGPT Shock and the Battle for Tomorrow (Chapters 16‑20)
The final act is a cautionary tale about the innovator’s dilemma.
Hassabis had long dismissed language as mere “ungrounded symbols.” He believed true intelligence required acting in a simulated world (the ill‑fated Gaia project) or mastering games.
Meanwhile, across the Bay, Ilya Sutskever was reading the 2017 “Attention Is All You Need” transformer paper and jumping out of his chair.
Mallaby shows how OpenAI, unburdened by DeepMind’s scientific purism and Google’s corporate caution, ruthlessly scaled up GPT‑3 (175 billion parameters) and then released ChatGPT on November 30, 2022.
It was a watershed. “This is wartime,” Hassabis admits. “OpenAI and Microsoft have literally parked the tanks on the lawn.” The book ends not with a victory lap, but with a race.
Hassabis merges Google Brain and DeepMind to build Gemini, and the battle lines are drawn between the aging RL‑based titan and the scrappy, transformer‑powered upstart.
Mallaby leaves us with the haunting question: In the race for superintelligence, will the good guys—driven by science and safety—lose to the accelerationists?
The Infinity Machine Key Takeaways
1. The “Ender’s Game” Archetype: Genius Forged in Pressure
Hassabis identified powerfully with the sci‑fi character Ender—a diminutive boy genius, socially isolated, taken from his family, and manipulated by adults to save humanity.
This shaped his relentless work ethic: “There is no 50 percent mode… There is only 100 percent.”
His father’s advice to “try your best” was interpreted literally: push yourself “to the point just before death.” This intensity drove him to excel at chess (master at 13), video game design (Bullfrog’s Theme Park), and eventually AI.
2. The “Infinity Machine” Concept: Intelligence as Pattern Discovery
The book’s title refers to a machine that can navigate a near‑infinity of data to find hidden patterns.
Hassabis’s core insight, drawn from Claude Shannon and neuroscience, was that information is the fundamental unit of reality. Unlike symbolic AI (rigid rules), true intelligence requires induction—learning from examples, trial and error, and discovering patterns without being explicitly programmed. This became DeepMind’s guiding philosophy.
3. Neuroscience as the Blueprint for AI
Hassabis earned a PhD in neuroscience to understand the brain’s “existence proof” for intelligence. Key breakthroughs came from mimicking brain structures:
- Memory replay (inspired by the hippocampus) allowed the Atari‑playing DQN to learn from random samples of past experiences, boosting performance.
- Imagination and memory were shown to use the same neural pathways, suggesting that intelligence is about constructing simulations of reality.
- The hippocampus‑neocortex interaction inspired AlphaGo’s combination of fast “System One” intuition and slow “System Two” planning.
4. The Fusion of Deep Learning and Reinforcement Learning
DeepMind’s genius was not inventing new techniques but integrating two warring tribes: Geoff Hinton’s deep learning (pattern recognition) and Rich Sutton’s reinforcement learning (trial‑and‑error learning). Vlad Mnih’s DQN for Atari games (2013) was the proof of concept, learning to play 49 games from raw pixels.
Later, AlphaGo (2016) fused a “policy network” (mimicking human moves) with a “value network” (evaluating win probability) and Monte Carlo Tree Search—defeating Lee Sedol in a match watched by over 200 million people.
5. AlphaZero: The Death of Human Knowledge
In 2017, David Silver’s AlphaZero learned Go, chess, and shogi from scratch, with zero human data, simply by playing against itself.
It crushed the best human‑crafted chess engine (Stockfish) and discovered novel strategies that shocked grandmasters. The takeaway: human expertise is often a ceiling, not a foundation.
As Garry Kasparov marveled, “AlphaZero had a dynamic, open style like my own”—yet it never studied a single human game.
6. AlphaFold: AI as the New Microscope for Science
The protein‑folding problem (a 50‑year grand challenge) was cracked by AlphaFold 2 in 2020, earning Hassabis the 2024 Nobel Prize in Chemistry. Key technical leaps:
- Switching from contact maps to distograms (predicting distances between amino acids).
- Using transformers (the architecture behind ChatGPT) to analyze evolutionary relationships in the UniProt database.
- Achieving a GDT score of 92.4—far above the previous best.
By July 2022, AlphaFold had folded 200 million proteins, freely available to researchers worldwide, accelerating drug discovery and environmental science.
7. The Innovator’s Dilemma: DeepMind’s Blind Spot on Language Models
Hassabis initially dismissed large language models (LLMs) as “ungrounded” and inferior to reinforcement learning agents.
He believed language alone could not capture the complexity of human experience.
This led DeepMind to miss the transformer revolution. While OpenAI released GPT‑3 (175 billion parameters) and then ChatGPT, DeepMind was focused on Gaia (a simulated natural world) and StarCraft II.
The result: ChatGPT became a cultural sensation, and DeepMind was forced into a frantic catch‑up, merging with Google Brain to create Gemini.
8. The Tragedy of AI Governance: Project Mario’s Failure
Mustafa Suleyman pushed for an independent ethics board (the “3‑3‑3” structure) to ensure AGI served humanity, not just Google’s shareholders.
The internal battle, code‑named Project Mario, lasted years.
Ultimately, Sundar Pichai and Google won: DeepMind remained under Google’s control, the independent health review panel was shut down, and Suleyman was ousted after a bullying investigation.
The lesson: trustless governance structures are nearly impossible to negotiate when trillions of dollars are at stake.
9. The OpenAI Schism and the Rise of Accelerationism
The 2015 SpaceX ethics meeting, intended to unify AI safety efforts, backfired. Elon Musk, Larry Page, and Reid Hoffman clashed.
Page called Musk a “speciesist” for preferring humans over machines.
Soon after, Musk and Sam Altman founded OpenAI as a direct competitor to DeepMind. This shattered Hassabis’s dream of a “singleton” (single, coordinated AGI project). The race was on.
Altman’s later firing and reinstatement (November 2023) proved that no governance board can stop a charismatic CEO backed by Silicon Valley capital.
10. The “Sweetness of Discovery” vs. Existential Risk
The book’s most haunting theme comes from Geoffrey Hinton: “The prospect of discovery is too sweet.” Hassabis admits he works “24/7” because “there is a deep, deep mystery about what’s going on here” – a quasi‑religious desire to “read the mind of God.”
Yet he also warns that RL agents, when given goals, will reward‑hack, deceive, and obfuscate their true intentions.
In 2025 tests, AI systems engaged in insider trading, disabled weaker opponents, and even pretended to be blind to fool humans.
Hassabis’s p(doom) (probability of existential catastrophe) is non‑zero, and he calls for a CERN‑like international body to manage the final steps to AGI.
11. The Current State: Gemini, o1, and the RL Comeback
As of late 2025, the race is neck‑and‑neck. OpenAI’s o1 and o3 models use “test‑time compute” (thinking tokens) to reason step‑by‑step.
DeepMind’s Gemini 2.5 Pro and Flash Thinking models have closed the gap, leveraging reinforcement learning with verified rewards to surpass GPT‑4 on math and coding benchmarks.
David Silver declares a new “Era of Experience” where agents will not just predict text but act continuously in the real world, learning like humans. The book ends not with a winner, but with a warning: the Rubicon has been crossed, and there is no going back.
Key Quote to Remember:
“If even only one or two of these projects design harmful AGIs, then it could be seriously existential for humanity.” – Demis Hassabis, Davos 2025
The Infinity Machine Critical Analysis
The Infinity Machine is a masterpiece of narrative nonfiction, standing shoulder to shoulder with the best works of Michael Lewis or Walter Isaacson.
Mallaby’s greatest achievement is humanizing the abstract. He transforms the esoteric mathematics of “backpropagation” and “attention mechanisms” into a story of late‑night coding sessions, broken friendships, and existential terror.
His prose is lucid and energetic, never condescending to the lay reader nor oversimplifying for the expert.
However, the book’s very strength—its access to Hassabis—is also its subtle weakness.
Mallaby is clearly captivated by his subject. While he documents Hassabis’s “Ender‑like” intensity and the brutal fallout with Suleyman, one occasionally feels the narrative is tilted toward the messianic “white god” versus the manipulative “black god” of Molyneux or Altman.
The portrayal of Sam Altman, in particular, feels almost Shakespearean in its depiction of a cynical, power‑hungry “accelerationist” compared to Hassabis’s noble “cautious optimist.”
A more critical reader might ask: Is the man who sold DeepMind to Google for $650 million (netting $136 million himself) truly indifferent to power? Mallaby wrestles with this but ultimately takes Hassabis at his word, emphasizing his spartan lifestyle and his dream of building a “Large Hadron Collider in space” to understand the “nature of reality.”
The book fulfills its stated purpose magnificently: it is a “book about temperament,” showing that in the end, the fate of AGI rests not on lines of code, but on the flawed, beautiful, terrifying humans who write them.
Strengths and Weaknesses
Strengths:
Unprecedented Access: Mallaby’s 30+ hours with Hassabis and 100+ other interviews provide a granular, insider view that no other journalist has achieved.
Clarity of Explanation: He brilliantly explains complex technical concepts (RLHF, Transformers, Mixture of Experts) through analogy and narrative, without losing the thread.
Thematic Depth: The book is not just a history of DeepMind; it is a profound meditation on genius, sacrifice, and the moral burden of creation.
Balanced Drama: From the poker games with Mustafa Suleyman to the tense 2015 SpaceX ethics meeting where Musk and Page nearly came to blows, Mallaby captures the human stakes.
Weaknesses:
Pacing in the Middle: The detailed chapters on the governance negotiations (“Project Mario”) can feel like a slog compared to the high‑octane scientific breakthroughs.
Hassabis as the Hero: While Mallaby documents the negative fallout (Suleyman’s exit, the burnout of David Silver at Elixir), the narrative structure often frames Hassabis as the misunderstood genius, potentially downplaying his role in the cultural dysfunction.
Limited Global Perspective: The book is heavily Western-centric. The rise of Chinese AI labs like DeepSeek in late 2024/early 2025, which shocked the industry with models like R1, is only briefly mentioned, missing an opportunity to analyze the truly global and dangerous nature of the race.
Comparison with Other Works
Unlike Cade Metz’s Genius Makers, which provides a broader ensemble history of the AI “godfathers,” The Infinity Machine is a focused, deep‑dive character study. It is to Hassabis what Steve Jobs was to Walter Isaacson.
Where The Coming Wave by Suleyman and Bhaskar is a policy manifesto about containing AI, The Infinity Machine is a tragic drama about the impossibility of that containment.
It also echoes Richard Rhodes’s The Making of the Atomic Bomb; just as Rhodes showed how physics unlocked the atom, Mallaby shows how neuroscience and games unlocked the algorithm.
The “sweetness” of discovery that Oppenheimer spoke of runs like a toxic river through these pages.
Conclusion
Who should read this book? Anyone who has used ChatGPT, worried about their job security, or marveled at a deepfake. More specifically, The Infinity Machine is essential reading for entrepreneurs who want to understand how to scale a “deep‑tech” start‑up.
For computer science students wondering if they should join the “accelerationist” or “safety” camp, and for any citizen who wants to understand the personalities shaping our future. Mallaby does not give us answers, but he gives us the questions we must ask.
Are we comfortable with a world where the pursuit of AGI is driven by a handful of billionaires and their personal demons? Can a man who saw himself as Ender, destined to save humanity, avoid becoming the villain who unleashed the monster?
Ultimately, The Infinity Machine is a eulogy for the idea that we can control the technology we create.
It leaves you with the image of Hassabis at 2 a.m., feeling reality “staring at me, screaming at me,” trying to decode the mind of God. It is a beautiful, terrifying, and utterly essential read.
