Book Review: ‘The Optimist’ and ‘Empire of AI’

Two new books try to unmask Sam Altman and OpenAI, to varying degrees of success

Image: World Economic Forum

When OpenAI’s board tried and failed to oust Sam Altman in November 2023, rumors abounded as to why. At the time, the board simply said Altman had not been “consistently candid”, words seemingly designed to invite speculation. Was it an effective altruist plot? Was it because Altman had done something illegal? Was it Ilya Sutskever’s attempt to seize control?

Two new books, both out tomorrow, finally put the matter to rest. Together, they paint a picture of a leader with a concerning habit of dishonesty, deception, and manipulation; a person savvy enough to never leave a fully-smoking gun on the table; and a man capable of terrifying his own leadership team into silence.

He was, in other words, not consistently candid.

The Optimist, by Wall Street Journal reporter Keach Hagey, is a brisk, compelling account of Sam Altman’s rise. Hagey’s tightly constructed narrative adeptly surfaces how Altman came to be the man he is today — or more accurately, how Altman always was the man he is today. For the most striking aspect of Hagey’s telling is the consistency of Altman’s character over decades. As a child in St. Louis, Altman tells Hagey, he was “thinking that someday the computer was going to learn to think.” At the beginning of college, Altman supposedly scrawled a list of what he wanted to work on: “topping it, in order, were AI, nuclear energy, and education.”

But it’s not just Altman’s interests that have proved resolute. Hagey shows how the entrepreneur’s early career included many episodes oddly reminiscent of what was to come at OpenAI. At Loopt, his location-sharing startup, “much of the senior management” complained to the board that “Altman sometimes said things that weren’t true”. Later, “a dozen of the company’s senior leaders” called for him to be replaced as CEO.

Loopt’s approach to the societal downsides of its products also looked remarkably similar to OpenAI’s. Altman, we’re told, dealt with privacy concerns by “running straight at [them].” “Rather than downplay concerns, he seemed to relish talking about them to anyone who would listen,” Hagey writes. “He played up these fears to pitch Loopt as particularly safety-conscious, with strict safeguards to prevent misuse.” Altman took this tactic all the way to Washington, where “the idea was to suggest to lawmakers that they could regulate the industry now, and that Loopt could be a helpful partner in deciding what the regulation should be.” The similarities to OpenAI’s lobbying efforts are striking.

At Y Combinator, Altman upset yet more colleagues. Jessica Livingston, the accelerator’s founder, was reportedly “surprised and hurt” when she learned how much time Altman was spending on OpenAI instead of his day job — so hurt, in fact, that she reportedly did not speak to him for years afterward. Another YC partner, offering a defense of Altman, could only manage this: “He is moral. He tries to do the right thing. But he is conflict avoidant, not a great communicator, and sometimes he moves so quickly that he breaks trust.”

The Optimist depicts, in gory detail, the most spectacular instance of that playing out: the OpenAI board’s attempt to oust him. The book shows that the move to oust Altman was instigated not by the independent board members, as commonly assumed, but by Mira Murati and Ilya Sutskever. Murati, we’re told, complained in great detail about Altman’s “toxic” management style, while Sutskever “felt that Altman was dishonest and causing chaos”. Many of the problems stemmed from Altman’s manipulative nature: as Murati supposedly told Toner, he would first “say whatever he needed to say to get you to do what he wanted, and second, if that didn’t work, undermine you or destroy your credibility.” He would not address conflict between different executives at the company, most notably when it came to Greg Brockman (who had a pattern, Sutskever alleged, of “bullying”). Sutskever, Hagey writes, “was terrified of what would happen if Altman found out they were talking to the board”.

These employee concerns, combined with numerous instances of Altman lying directly to board members, led the board to conclude that his position at OpenAI was untenable.

Virtually everything in Hagey’s book is corroborated, and often expanded on, by journalist Karen Hao in her own book, Empire of AI. Hao writes that “the bottom line was Altman, sometimes with Brockman, had treated many people similarly over the years, manipulating and lying to people so habitually that at times he said things that he didn’t even seem to believe himself.”

According to Hao, several senior OpenAI executives “had described Altman’s behaviors as abuse and manipulation; most had highlighted his lack of honesty and their inability to trust what he said.” Altman seemingly lied, Hao writes, even while the board was confronting him about his lying. Both books thus buttress the board’s claim that Altman made meaningful governance impossible. “Bit by bit,” Hao writes, “Altman was trying to cloud their visibility and maneuver in ways that prevented the board from ever being able to check him.”

But while the two books converge in their damning assessment of Altman, they differ markedly in style and focus. The Optimist is the slimmer volume but the heavier hitter, a zippy work that serves not just as a biography of Altman, but also a miniature history of the past twenty years of tech. If you want to understand the forces behind Altman and OpenAI, this is the book to read.

Hao’s meticulous reporting in Empire of AI, meanwhile, is frequently undermined by the book’s polemics and lack of focus. It aims to be a socioeconomic critique of the entire project of modern AI, spending whole chapters interrogating the entire notion of “intelligence”, railing against the scaling hypothesis, and uncritically platforming critiques from many of the biggest names in AI ethics and accountability.

Unfortunately, the arguments are scatter-shot and often flimsy. Hao uses the fact that we can’t cleanly define “intelligence” to dismiss the entire idea of AGI as absurd, overlooking that most definitions of the technology rest on its impacts (e.g. systems that “outperform humans at most economically valuable work”, as OpenAI defines it). In dismissing the technology, Hao also fails to examine its potential effects, refusing to entertain the notion that a system which can do most economically valuable work would have vastly greater impacts for the economy and innovation than current systems do. Catastrophic‑risk scenarios, meanwhile, are waved away as speculative and a distraction, as though risks shouldn’t be addressed until they’re on our doorstep. The present-day harms caused by open-weight systems — a technology Hao is clearly a fan of — go virtually unmentioned.

Hao’s economic analysis is particularly one-sided. Citing Acemoglu and Johnson’s Power and Progress, she argues that technological revolutions “promise to deliver progress”, but instead tend “to reverse it for people out of power”. Yet this contradicts substantial historical evidence that technological advancement has, over time, improved fundamental measures of wellbeing — from life expectancy to child mortality to basic living standards — across all segments of society, even during periods of uneven economic distribution.

To support her claims, Hao makes distasteful comparisons, likening the modern AI boom to the development of the cotton gin, which “only served to intensify slavery”. AI does indeed involve many exploitative practices, and Hao is at her best when humanizing the often-invisible labor that enables these systems, particularly in her vivid depictions of content moderators in Kenya who suffered psychological trauma while labeling toxic content for OpenAI’s filters. But comparing this to the horrors of slavery and colonialism, as Hao repeatedly does, leaves a bitter taste in the mouth, and risks trivializing very dark portions of our history.

Empire of AI is not without merit. Its reporting is exemplary, its human vignettes are memorable, and its outrage occasionally lands a punch. It is also more damning: The Optimist, especially in its telling of the OpenAI NDA scandal, is sometimes too quick to give Altman the last word. Hao, to her credit, is unsparing in demonstrating how the incident was yet another example of Altman lying.

Yet on balance, The Optimist is the better book. With its tighter focus and more detailed history lessons it is sharper, leaner and — crucially — better at connecting the dots between youthful fibs and trillion‑dollar stakes.

From both books, though, a similar picture of Altman emerges: that of a leader whose talents are undeniable but whose character flaws are profound. As the November 2023 board crisis demonstrated, even those who worked closely with Altman eventually concluded that he should not be entrusted with developing AGI. We may all come to regret that he still is.

The Optimist: Sam Altman, OpenAI, and the Race to Invent the Future, by Keach Hagey. Norton; 368 pages; $31.99. Amazon US; UK.

Empire of AI: Dreams and Nightmares in Sam Altman’s OpenAI, by Karen Hao. Penguin Press; 496 pages; $32.00. Amazon US; UK.