Book Review: The Infinity Machine, Sebastian Mallaby

Like hovering your hand over a sparking electric fence, there is an inherent tension one can pick up that runs through a deeply researched biography. In writing it, the author must balance their own admiration for or disdain of their protagonist, choosing what details to reveal, emphasis or blur. As they read the work, the reader can sometimes begin to see the author’s own identity projected onto their subject too. Suetonius, Plutarch, McCullough - it is a tale that persists through the millennia. It is what makes the best biographies electric.

This tension becomes even more stark when the protagonist of the work is still alive, and is intensified further when the subject sits at the frontier of a technology which may change how many of us live, or if we in turn, as a species, remain alive. Apply to this potent premise the distorting influences of power, capital, and factional elite politics, and you arrive at The Infinity Machine, a book that often raises complex questions of the human spirit dealing with a complex technology with an unknown spirit.

The Infinity Machine is the most comprehensive biography of Demis Hassabis to date. Hassabis co-founded and runs DeepMind - now Google DeepMind (GDM) - one of the handful of artificial intelligence (AI) companies at the frontier of this multi-trillion dollar technology. The AI systems GDM and its competitors such as OpenAI, Anthropic, or DeepSeek produce could play a role in heralding a new era of humanity.

The book is brilliant on several fronts. It explains complex AI technical terms in clear fashion, weaving their originators and advocates into the offered examples. It speeds through decades of history and science in under 500 pages, introducing the reader to a wide cast of lively characters and ideas who go on to populate many of the best moments of the work. It ties together capital flows, scheming in Silicon Valley, and the advantages and tensions of academia and corporations with ease. And it is clearly deeply researched: the footnotes alone could sustain a separate book, and the author has dozens of conversations with many of the leading figures in this space over several years. The Infinity Machine deserves to be read widely.

I am neither an AI nor technology technical expert, although I work around those who are. My own background is far more focused on the politics of decisions - by which I mean the way information, emotion, and power permeate through critical human moments. What follows then is both an analysis of this book primarily through this lens - what drives these people in power, beyond just Hassabis - and a series of open questions the work led me to consider. All views are mine in a personal capacity.

Let us go forward with this premise. We are dealing with what appears to be a novel technology that may redefine humanity, or which may lead to our extinction, with a huge number of options in between. To this end, the people who control this technology matter immensely, and how we study them matters too. Understanding these people allows us to see why they are racing towards Artificial General Intelligence (AGI), a somewhat loosely defined term, and gives us the opportunity to judge if we think they are the right person or organisation to be doing so. How do they behave? What are their incentives? What fundamentally drives them? As Mallaby notes, “when you are building a machine of infinite potential, the stakes are so high.”

What Drives A DeepMind Founder?

What drives DeepMind? Mallaby unpacks this question through both the prism of its founding team and the ethos and culture the company tries to create.

We are introduced to a young, utterly brilliant mind. A chess champion and prodigy who seemingly wins whatever game he turns his attention to, we follow Demis Hassabis and his driving desire to work on AI and unlock many of the secrets of the universe as he makes his way through school, his first job, and university. Hassabis embodies the archetype of an ‘Ambitious Affable Scientist’ you would find in one of the dozen or so science fiction works mentioned in this book, driven by pure love of the game and a boundless hunger to push the scientific frontier forward, not money and power.

As he learns throughout the book, to push this frontier forward and benefit humanity, he will need to be at the table, buying a seat through capital and power. His ultimate incentives are articulated best in the epilogue during a conversation with Mallaby: ‘“AGI should be gifted to the world eventually. I mean, AGI is infinitely bigger than a company or a person or a set of owners. It's bigger than capitalism and national economies. It's humanity-sized, really. "It's humanity's invention and it's going to affect all humanity. So humanity should run it. Unfortunately the problem is, what are the right institutions? "Until we figure that out, I do need power, at least a bit of it. "It's like with money. I don't care about money at all, but I need some of it."”

Described as having a “magpie” approach to gathering information, Hassabis weaves together disparate ideas and intellectual PhD wanderers, convincing them to join DeepMind or work harder through “Jedi” mind tricks, depth of intellect, and raw charisma. Rather than “pitching [Peter Thiel] with yet another start-up story…he hooked Thiel with chess, observing that there was a deep tension between the bishop and the knight, with the two pieces carrying the same value yet possessing vastly different capabilities.” When Thiel and his team push for DeepMind to move stateside, Hassabis refuses, because he is “a British patriot who bristled at the presumption of American pre-eminence.” Within DeepMind itself, there was “a culture that brought out the best from its scientists. With his habit of collecting ideas from everywhere - movies, books, chance acquaintances in the university bar - Hassabis understood instinctively how to find the special quality in each team member.” Hassabis has a strong preference for technical experts and an initial scepticism of Large Language Models (LLM.)

We meet the New Zealander and DeepMinder-to-be Shane Legg, “an avid coder [who] incorporated mathematics into his programs [and saw] his performance in his school math class [go] from mediocre to outstanding…As Legg reflected later, the lesson of his youth was, "Write my own rules, because there's a lot of bullshit around. A lot of bullshit."” As far back as 2009, Hassabis and Legg discussed AI risk, and the former describes the latter as “like finding an oasis”. Notably, Legg is relaxed about Hassabis having 9x more equity than him when DeepMind is formed. His incentives are, like Hassabis, science-based rather than accumulating power and capital. The fact that the pair get on so fluidly - two scientists who overlap to the core - is worth noting.

Finally, via Hassabis’ younger brother, DeepMind’s other future co-founder Mustafa Suleyman strolls into our lives in a series of poignant anecdotes: “When Mustafa turned fifteen, …his parents split up and his mother followed her new partner to New Zealand. His father returned to Syria to remarry, leaving Mustafa in London to fend for himself and his fourteen-year-old brother. The boys had no home to go to, so they found shelter with friends. His father sometimes sent them money, but it was never enough.” Thankfully Suleyman is entrepreneurial, selling sweets and fixing up cars for income. “Hardship did not stop Suleyman from flourishing academically.” He is motivated through a deeply held belief in social justice (initially raised in a devout Muslim household and constantly helping his community, before later becoming an atheist), using the technology to benefit the public good (a chapter or so is devoted to DeepMind’s efforts to aid parts of Britain’s crumbling NHS) and eventually making sure AI safety is at the heart of the mission. Again, money does not play an overwhelming role in his drive.

The Scientist Versus The Corporate Suit

We meet several other important figures in the AI space, and in doing so, begin to learn of their perceived motivations and incentives for entering the frontier AI race. It is here that the book perhaps has a slight blind spot: for it rarely examines in depth the criticisms levelled by this motley crew against Hassabis.

Peter Thiel, the contrarian billionaire and legendary Silicon Valley figure, is an early important addition. Thiel “wasn't sure if he could trust Hassabis” despite financially backing the firm, citing him being a fellow champion chess player and masterful at the game Diplomacy. “Thiel thought that Diplomacy was essentially a test of how well you could manipulate people,” we are told, before moving immediately on. References to these games appear intermittently, such as when Hassabis comes back to Thiel’s Founders Fund looking for more money. “Thiel thought Hassabis was playing games, not least because he knew that Hassabis was a player of games, and if Hassabis was playing games, Thiel wanted to counterplay him.” Mallaby later notes that the relationship between the pair goes cold, with Thiel not taking meetings or sharing feedback. “Revealing a curious blind spot, Hassabis was slow to pick up on these signals. It was as though the Jedi, accustomed to winning people to his cause, couldn't adjust when he encountered an aloof cynic…“Demis would get up in front of the company and tell the team how he had met with Peter and briefed Peter and Peter believed in this and Peter believed in that," Suleyman recounted. I was like, 'Dude, you haven't even seen Peter' "I thought, You just can't say stuff like that. But Demis didn't think of it as lying or even exaggeration. It was just his reality.””

Interestingly, this brief handful of sentences should perhaps lead to further questioning - why did Hassabis feel the need to lie to his team? Is it correct to try and post rationalise this as good leadership and keeping confidence high rather than lying? Does it speak to a deeper personality trait? Instead, we move swiftly on.

Elon Musk appears at length. An early worrier over the potential of AI to decimate the human species, anecdotes arrive thick and fast. A “formidable space-and-cars tycoon, who was also clownish, jealous, and vainglorious,” Musk goes from friend to foe in the space of 100 pages. By a fire pit late on the night of his 44th birthday, Musk ‘terrified of superhuman intelligence”, speaks to Larry Page. Page told him not to worry, that “he looked forward to a time when people might merge with intelligent machines, or when machines might simply replace humans. Evolution would ensure that the best form of intelligence won out; if the best form involved fast silicon circuits rather than slow biological tissue, so be it. There was no point being sentimental about such things.” Furious, Musk pushed back. “Page retorted that Musk was a "speciesist" —a bigot with a soppy prejudice in favor of carbon over silicon." The speciesist epithet sent Musk over the edge.”

Why did Musk actually develop these views? What other factors were at play beyond the alleged jealousy or fear of a corporate getting to AGI first? Is Musk himself not quite similar to Hassabis in that he comes from a science and engineering background, and is obsessed with science fiction and intricate wide ranging theories of the world? Why do they differ so much in their motivations, and, based on their trajectories, is there a chance the former may become more like the latter if he gathers more wealth and power? Can the ‘Ambitious Affable Scientist’ remain uncorrupted by the capital and power he accumulates? Likewise, Musk’s efforts and incentives to grapple with the existential safety elements of the technology don’t fully receive the unpacking they perhaps deserve.

Humans are complex creatures with rich interior lives; few of us are binary. What drives us can be hard for our own selves to articulate, even more so given some of it would not be appropriate for the public domain, let alone for others to try and analyse. Yet it is critical, given the importance and directional potential of this technology, that we apply a level of complexity in trying to understand the people driving it at the very frontier.

Incentives and Power

The sharpest criticism is reserved for Sam Altman, the current leader of GDM’s competitor OpenAI (OAI). Described to readers as “a silver-tongued networker and investor, a gifted opportunist defined less by his devotion to AI than by his general ambition”, and as a “flawed but formidable leader,” we learn that “in 2017, Altman had contemplated a run for governor of California,” a point delivered to us as a perceived flaw, because this incentive indicates a lust for power. When OAI goes through its own leadership and structural stresses, Altman is presented as an untrustworthy instigator in the shadows. In one instance, after the dust settles around a failed coup effort, he “quickly cozied up to the big man [Elon Musk], deftly ensuring that his own role in inciting the rebellion went unsuspected…He threw Brockman and Sutskever under the bus, telling Musk's trusted lieutenant, Shivon Zilis, that their remonstrations had been "childish.”” A later section notes Altman’s particularly weak flip flopping and sycophancy with regard to taking AI safety seriously, causing the exit of Amodei “along with several other dissidents; their numbers soon swelled to a bit over a dozen, representing about a tenth of OpenAI's research team.” [Much of this rings true given the public reporting on Altman and ongoing court cases that have revealed messages and texts from this period and others.]

To this end, Altman’s incentives appear as fairly superficial: wealth and power. Mallaby ponders “perhaps, on a generous reading, Altman was, like Hassabis, trying [italicised in the text] to be good, even though his potential to do good remained debatable.” Where Hassabis is an ‘Ambitious Affable Scientist’ who sometimes errs with poor strategic choices, Altman’s decisions are viewed at all times as cynical rather than strategic. His “commercial instincts, and his success in attracting money and talent, owed much to his embeddedness in Silicon Valley…Starting in his early twenties, Altman had established himself as a star in this constellation, making and soliciting investments, exchanging introductions and ideas, twisting threads of mutual interest and friendship into a sinewy lattice of connections…If Hassabis was a contrarian individualist, patriotically remaining in London, Altman stood at the center of a formidable network that circulated people and capital and buzz, all in the service of making the new future.”

Indeed, an effort to think through some potential policy prescriptions sees Altman’s work portrayed primarily as “an exercise in branding—as a tool for raising capital and luring talent-Altman's essay was masterly. On a substantive level, it was harder to know what to make of it. …talk is cheap. Given his Machiavellian tendencies, Altman's talk was particularly cheap, especially since he was in no position to ordain his proposed wealth tax.”

One doesn’t have to like Altman to feel that the above skillset requires precise and consistent honing and effort, and to get to the level he has got to, an almost superhuman devotion to the craft. Yet is written off as Machiavellian or shallow in an almost casual way, because the perceived incentives behind it - money and power - are viewed as sitting further down the purity hierarchy than scientific ambition or drive. That could be a fair belief - it seems sensible to want someone guided by higher ethics to be running the company that achieves this breakthrough first - but does it do a disservice to the drive behind Altman, or more interestingly, to those around him?

For example, if Altman is a known liar, manipulator and chameleon - again, all accusations which are justified in various public articles and court documents - why do so many intelligent people still work under him at OAI? They cannot all believe that he is truthful only when he deals with them. Is it that they believe OAI is most likely to reach AGI first, and they want to be part of this elite circle? As Mallaby notes, over 700 of the nearly 780 or so OAI employees wrote a letter asking for him to return following his ousting. Why do so many of his financial backers support him over his competitors? They cannot all believe that Altman will share power should one version of AGI be reached. Does Hassabis really differ from Altman in his view that he just needs a small bit of power as a vehicle to achieve AGI for the good of all humanity? These questions could have been teased out further.

Epilogue: A Question As Old As Time

There is a tension as old as humanity buried in The Infinity Machine, and Mallaby, to his credit, circles it repeatedly, even if he never quite pins it down, perhaps by choice. Who should steward power: the philosopher or the prince? The Scientist or The Suit? And what if this power could be utterly transformative for humanity? And what if it’s not?

Hassabis is rendered as a figure whose compromises are the reluctant concessions of a brilliant mind grappling with an imperfect and complex world. Altman’s compromises are presented as confirmation of pre-existing cynicism, a man interested only in himself and his ambitions, who when faced with these imperfections, muddies them further by behaving poorly. Thiel is written up as a shrewd but contrarian investor who mistrusts Hassabis thanks to his excellence in boardgames that require manipulation. Musk viewed him as working for a corporation that would not put humanity first if it achieved AGI. All of these are fascinating positions that speak both to their holders and to the incentives that govern or drive these powerful people, and how they interact with the protagonist of this book. Yet none are really deeply explored.

Ultimately, if we had a book as deeply sourced and beautifully researched on each of the AI companies' leaders or characters mentioned in The Infinity Machine, we would be in a better place to understand what drives this handful of people sitting at the frontier of humanity’s next era. In the meantime, this work helps illuminate the driving incentives and motivations of one of its key figures. And who knows: Hassabis may yet prove to be the right person at the frontier.