Empire of AI, an appreciation
The nightmares that make generative AI possible
What I reject is the dangerous notion that broad benefit from AI can only be derived from—indeed, will ever emerge from—a vision for the technology that requires the complete capitulation of our privacy, our agency, and our worth, including the value of our labor and art, toward an ultimately imperial centralization project.
(Karen Hao, Empire of AI, p. 413)
Is your anger pure?
The late Alexander Cockburn, when he was writing for The Nation, made a point of asking his interns “Is your hate pure?” Cockburn was referring to their hatred for the powerful and corrupt, and claimed that the young Ed Miliband, an intern at The Nation, was “in shock” when asked the question, and could only stammer “I… I… don’t hate anyone.” This, as Cockburn predicted when Miliband became leader of the British Labour Party, was “all you need to know. English capitalism will be safe in his hands.”
To my everlasting shame, I cannot remember whether as an adult I ever went so far as to hate anyone, and this is probably why, deprived of the internet’s instantaneous powers of recall, I would have misquoted Cockburn’s question as “Is your anger pure?” To that question I would confidently have replied in the affirmative. There have rarely been so many good and obvious reasons to be angry. My colleague Peter Woit has extended his blog’s reach beyond his usual audience of scientists and mathematicians and is now deeply appreciated among Columbia University’s many departments, by expanding his repertoire in parallel with the explosive growth of occasions for anger.1
And yet anger can be experienced in various degrees of purity. I was reminded of this while reading Karen Hao’s Empire of AI, a deeply critical history of OpenAI and of its protean CEO Sam Altman. Hao’s book is subtitled “Dreams and Nightmares in Sam Altman’s OpenAI,” and the author displays a more than journalistic skill in making the nightmares vivid. That was an understatement. Two chapters in particular, entitled “Disaster Capitalism” and “Plundered Earth,” were so literally sickening that while reading them I had to put down the book repeatedly to process what I had just read.
Feeling physically ill is not an experience I can nor wish to share with readers of Silicon Reckoner, but I would be gratified if some of these readers were to finish today’s episode with anger of a purity warranted by the stories Hao’s book reveals. The chapters in question deal with the environmental and social costs, respectively, of the training process that has given rise to the generative AI models that made OpenAI famous. The need for an especially pure anger has become more urgent in the wake of recent reports on mathematicians’ interaction with such models.2 Whatever the merits of these models as measured by industry benchmarks, none of these reports, even the least credulous,3 has raised the question of what the world and the people in it had to sacrifice, voluntarily or otherwise, in order to enable these companies to produce and train their models.
Hao’s book, specifically in its two nightmare chapters, focuses on precisely this sacrifice. It shares this focus with Crawford’s Atlas of AI, with Couldry and Mejias’s Data Grab, with How Data Happened by Wiggins and Jones, and with AI Snake Oil by Narayanan and Kapoor, to limit myself to books I’ve read since the beginning of 2025. But Hao, a professional journalist, unlike the authors just mentioned, is experienced in taking the anger that visibly motivates her throughout this book and turning it into a strategically crafted language capable of unlocking the pure anger that her readers harbor, whether they know it or not, as a resource for concerted action.
Some passages from the two chapters in question illustrate what I have in mind.
On data annotation for self-driving cars (from Chapter 9):
When faced with economic collapse, Venezuela suddenly checked off the perfect mix of conditions for which to find an inexhaustible supply of cheap labor. Its population had a high level of education, good internet connectivity, and, now, a zealous desire to work for whatever wages… when Fuentes began to experience signs of a serioius health problem, she ignored the symptoms and continued working. All she could think about was putting in as many hours as possible in the final stretch of her employer’s operation.
The doctor later told her that had she waited any longer, she likely would have died. A coworker… had brought her to the hospital shortly before her body started convulsing and her pulse stopped for a full minute.…
Venezuelans became one of Scale’s top recruiting priorities. “They’re the cheapest in the market,” the former employee said.… As the pandemic hit, compounding the economic crisis, Venezuelans flocked to [Scale AI’s] Remotasks Plus en masse… Once Scale held the dominant position, its promises to workers faded.… After two hours of completing a tutorial and twenty tasks, Hernández earned eleven US cents. …
Scale proceeded [in other countries] to repeat the playbook it had developed in Venezuela again and again. It offered high earnings in each new market to attract workers and throttled those earnings as it settled in.… One group of eight workers in North Africa said Scale reduced their pay by more than a third in a matter of months.
On data centers in Chile (from Chapter 12):
"Welcome to the Quilicura Urban Forest," a project, it explains, that Google began in 2019 to give back to the community for hosting its data center. The sign includes a diagram to elaborate on the "forest's" benefits: on the left side is an illustration of Quilicura as an industrial zone, packed with factories producing greenhouse gases and air pollution; on the right is an illustration of the forest flourishing under the generous rain pouring out from a big cloud labeled "SMOG."
Google boasts about this forest on its website and in its PR releases, When I ask the company's country spokesperson for an interview about Google's development in Chile, she sends me instead some polished briefing materials, later adding that the company creates a community impact program for each new data center to support local projects such as in education, sustainability, internet access, and health; for Quilicura, Google has invested over $1.2 million. In her materials, the part about the forest talks about residents using the green space. There are no residents. The place is too far from any bus line, and there are no homes in the surrounding area to speak of. Outside the modest plot, too small to fit Google's data center itself, a dozen stray dogs meander around, barking and rummaging through the trash. The spokesperson said the forest is being updated to "evolve the experience for the community."…
That summer, as Google filed a report with Chile's environmental agency for approval of its data center-a largely rubber stamp process-MOSACAT, a water activist group, began combing through all 347 pages of the filing. Buried in its depths, Google said that its data center planned to use an estimated 169 liters of fresh drinking water per second to cool its servers. In other words, the data center could use more than one thousand times the amount of water consumed by the entire population of Cerrillos, roughly eightyeight thousand residents, over the course of a year. …Not only would the facility be taking that water directly from Cerrillos's public water source, it would do so at a time when the nation's entire drinking water supply was under threat.4
A clear conscience
I had been primed for anger by yet another conversation in which it was pointed out, correctly,5 that, in comparison with whatever ecological or social devastation may have been wrought on vulnerable countries or populations in the course of building the infrastructure or training the generative models that are capturing the attention of the mathematical research establishment, any additional devastation due to the exploitation of these models by mathematicians will be vanishingly small. To put it gently, this is a cop out. Karen Hao, who has done as much as anyone to bring these destructive tendencies of machine learning to public attention, has followed an increasing number of scholars by choosing the framework of colonialism as a way of clarifying the continuity of tech industry practices with the dynamics of the development of the global economy over the past centuries.
Nick Couldry and Ulises A. Mejias… argued that Silicon Valley’s pervasive datification of everything was leading to a return of disturbing historical patterns of conquest and extractivism. (p. 104)
When she began reporting on OpenAI, Karen Hao was the senior AI editor at the MIT Technology Review.6 It’s fair to assume, therefore, that she is not an opponent of technology in general nor of AI in particular. The lead quotation reproduced in the present post, which appears near the end of Empire of AI, refers to a potential “broad benefit from AI.” By that point she had devoted the body of her book to explaining why she believes this potential will not be realized on the current model of generative AI. For example, on the well-rehearsed claims that the explosive growth of data centers is not a problem, because “AGI will solve climate change once and for all,” she quotes Sasha Luccioni, climate lead at Hugging Face:
Generative AI has a very disproportionate energy and carbon footprint with very little in terms of positive stuff for the environment. (p. 276)
Can the community of mathematicians be counted on at least to acknowledge that, for a growing number of tech writers and engineers, talk about AI without referring to its environmental and social costs is irresponsible?7
The book’s afterword, entitled “How the empire falls,” describes a hopeful vision of a “formula for dissolving empire” through a “redistribution of power” along three axes: knowledge, resources, and influence. Whether or not subcultures of the mathematical community will be joining these efforts to redistribute power — I see no reason for optimism about top-down action, but I look forward to being surprised by spontaneous activism — much of the damage recorded in Hao’s two nightmare chapters is irreversible. And even if mathematical research makes a negligeable contribution to future damage,8 any research that builds on the models currently being promoted is retroactively complicit in the colonialism on which these models were themselves built. The ethical implications of this situation are no different from what those of us face who use the tools — like Substack — that were made possible by centuries of colonialism. What response could begin to compensate for the irreversible damage caused by colonialism?9
Here is one you may have missed.
The best account I’ve read of some of these recent developments is Aravind Asok’s guest post on Peter Woit’s blog.
The contrast could not be starker between the measured language of Asok’s report and the recent Scientific American article on the FrontierMath benchmark, which has become notorious since Ken Ono, quoted extensively in the article, published a brief disclaimer on LinkedIn (“I don't agree with the tone of the article. Mathematics isn't dying.”) And yet none of these reports makes the slightest mention of the FrontierMath scandal — OpenAI’s secret funding of and exclusive access to the FrontierMath benchmark. The scandal was revealed in January of this year, was widely discussed — it had the most views on this newsletter of the past three years, cumulatively, by a wide margin — and then simply vanished without a trace. Empire of AI explains through numerous examples how OpenAI has repeatedly made its scandals evaporate; but how mathematicians so universally succumbed to such a radical memory lapse remains a mystery to me.
UPDATE July 17, 2025: Hao devotes a separate section to the environmental costs of mining in Chile’s Atacama desert for the lithium and copper needed for most AI infrastructure. An article published today in the Guardian asks:
Mining companies are pumping seawater into the driest place on Earth. But has the damage been done?
This should not be taken for granted. As Rodrigo Ochigame reminded me, talk of applying AI in the interests of mathematics cannot be easily separated from the well-publicized plans of the industry to develop artificial mathematical reasoning algorithms in the quest for (ambiguously defined) artificial general intelligence. Much of Hao’s book centers on the history of this quest, but she does not stress the role the industry has assigned to mathematics; see this article, for example. With the reduction of the NSF mathematics and physical science budget by 66.8% and the drastic curtailment of federal government support for the sciences and for universities in general, we can expect to see a growing proportion of mathematics graduates opt for jobs in the tech industry. And we can expect these jobs to entail all the environmental and social costs of generative AI that Hao describes in her nightmare chapters.
As a subscriber to the MIT Technology Review I have access to the transcript of Hao’s recent interview with the Review’s executive editor Niall Firth. Here is a relevant excerpt. Firth had just asked
The data center costs are absolutely extraordinary, right? Like the data behind it is incredible. And it’s only gonna get worse in the next few years if we continue down this path, right?
Hao replies:
[According to a] McKinsey report … if we continue to just look at the pace at which data centers and supercomputers are being built and scaled, in the next five years, we would have to add two to six times the amount of energy consumed by California onto the grid. And most of that will have to be serviced by fossil fuels, because these data centers and supercomputers have to run 24/7, so we cannot rely solely on renewable energy. We do not have enough nuclear power capacity to power these colossal pieces of infrastructure. And so we’re already accelerating the climate crisis.
And we’re also accelerating a public-health crisis, the pumping of thousands of tons of air pollutants into the air from coal plants that are having their lives extended and methane gas turbines that are being built in service of powering these data centers. And in addition to that, there’s also an acceleration of the freshwater crisis, because these pieces of infrastructure have to be cooled with freshwater resources. It has to be fresh water, because if it’s any other type of water, it corrodes the equipment, it leads to bacterial growth.
And Bloomberg recently had a story that showed that two-thirds of these data centers are actually going into water-scarce areas, into places where the communities already do not have enough fresh water at their disposal. So that is one dimension of many that I refer to when I say, the extraordinary costs of this particular pathway for AI development.
A propos of “this particular pathway,” early in the interview Firth identifies
one of the themes of the book: the idea that technology doesn’t just happen because it comes along. It comes because of choices that people make. It’s not an inevitability that things are the way they are and that people are who they are.
The lesson is lost on most of the authors of articles like the Scientific American piece cited in an earlier footnote, and on too many of the technology’s mathematical boosters.
[Added May 2026] There’s some reason for optimism. A few months after this post was published, Ernie Davis published a full review of Empire of AI in SIAM News, the news journal of the Society for Industrial and Applied Mathematics, that treats these questions with the seriousness they deserve. Davis’s review includes the following memorable passages:
OpenAI delivered surprisingly powerful AI products with astonishing speed. With equally astonishing speed, it abandoned its initial idealism.…The overall picture was of an organization that was achieving extraordinary successes at the cost of a working environment that was simultaneously messianic, frantic, and paranoid.
and
The empire metaphor that Hao uses in the title of her book is very apt. Powerful tech companies and their leaders do indeed resemble historical empires and empire-builders in many respects, including their insatiable greed for wealth and power, alleged dishonesty and unscrupulousness, conviction that they are ultimately bettering the world, and blindness and lack of concern with any resulting damage.
and refers to “the fevered world of sociopathic billionaire technocrats.” One hopes that the Notices of the American Mathematical Society will also find space for serious reviews of books like Hao’s.
See note 4.
Thanks are due to the PIs and advisory board members of the NSF project “Normalizing Ethical Reasoning in Mathematics as a Foundation for Ethical STEM,” who helped me find the right language and ethical framework for some of the ideas expressed in this post.
Pure hate! A brilliant concept. Being bought up Catholic I was indoctrinated with the view that hate was a verboten emotion. I got over that in my twenties. There's so much to hate about the modern world: its immense greed, its vast inequity, and its obsessive deification of arseholes. Hao's book is essential reading for our times. Altman is surely one of the greatest A-holes of the century. Hao was terrific on Democacry Now. https://www.democracynow.org/appearances/karen_hao
And I'm glad to hear Peter Woit is being widely appreciated at Columbia. His views on physics were so spot on and he was for so long dismissed. He played an important role in my thinking when I was writing my first 2 books.
The conclusions drawn in this review are innately Marxist in nature, and I reject them wholesale because of their flawed belief that class envy is somehow pure in principle. The irony is implicit in the clearly intended nature of redistribution supposedly made possible by AI. One of the reasons it will collapse on itself and most likely result in a Venezuelan-like scenario, no matter where it goes, is because of the implicit belief that colonialism is as damaging as it is clearly implied to be. The belief that the environmental impact, at the core of the arguments presented in the analysis of this book, based on the idea that the predictive models currently used to understand greenhouse gas emissions and carbon footprint, are accurate, much less reliable, for the future prediction of the nature of this world's ability to operate, are highly arguable.
AI needs to be used responsibly in an ethical capacity. This has nothing to do with redistribution, and has everything to do with being a good steward of resources.