they want (you) to believe

on Karen Hao's Empire of AI and the religion of AGI

This is an unpaywalled (and slightly updated) version of a post from October last year, which I am now making free because I want to convince you to buy this book. It is out in paperback on May 19th in the UK, and you can pre-order it here, from one of my favourite independent bookshops, or from yours, or you can ask your local library to stock it. I truly do think this is one of the most important nonfiction books of our times, a vital piece of journalism that stands against so much of the AI hype we’ve been fed by multimillionaires with nefarious agendas. Please do read it.

---

When I was taking my first philosophy classes at college, twenty-two years ago now, one of the things that thrilled me most was the asking of questions I had not ever thought to consider before. I had an amazing Northern Irish tutor called Barry Kelly, who once missed a week of classes around St Patrick’s Day and told us that he’d been ill and had followed doctor’s orders to ‘drink a crate of Guinness’. We had two (very nice) lads in the class who would literally wear berets (I know), sat at the back and tended towards long black overcoats; once, when they arrived late and had to swish their way to their seats after everyone else had sat down, Barry asked: are you here to take communion?

One of the first questions on our Philosophy of Mind course was a seemingly simple one: what is a mind? This is one of those things that you can put to a group of people in the pub and they will act as if it’s the stupidest question in the world: we all know what it is, we’ve all got one. But then the follow up questions must come: well, then, how is the mind—nonphysical—connected to the physical brain? Is the mind nonphysical? Does the brain affect the mind or does the mind affect the brain, or both? If the content of your mind is completely different at age two compared to age 80, then when constitutes your particular mind, exactly? What’s the through-line? How about with animals? What makes us intelligent in a different way to them? What is intelligence, when it comes to it? And how does that relate to the mind?

Even when you’ve been thinking about these things for two decades, you can forget just how divided opinions are, on even the fundamentals. At the start of the pandemic, having touched on these topics with me before, my partner asked if we might get through first lockdown by having a regular ‘cocktail and philosophy hour’ at 5pm by our bay window; gimlets were the order of the day then, and it would take us away from our phones and give us something to think about beyond a sweeping plague. I figured that the mind-body problem, as it’s known, was a good place to start; because the mind and the body are two different things I said, with confidence, only to be immediately met with the response: are they? I don’t think so. Five years later we still don’t really agree (though we did stop cocktail hour quite quickly, on realising that lockdown was not just going to last a few weeks and we might end up irretrievably dependent on alcohol).

It is because of this question that I am so interested in AI, or rather, that I am interested in the claims made about AI.

For all of my adult life, we’ve had striving, erudite white tech capitalists making quite insane claims about what computer science will achieve within the next fifty years; it was one of these claims, in particular, that led me to write my first nonfiction book. The tech industry has long been in the business of hype, because it is hype that drives venture capital investment and it is investment that makes billionaires, regardless of whether or not a product works or even exists; look at Elizabeth Holmes. Despite it being very unclear exactly how generative AI will bring about the things it claims it will, in terms of advancement or profit-making (from the Guardian: ‘A recent Massachusetts Institute of Technology report revealed that 95% of companies investing in generative AI have yet to see any financial returns’), in March 2025, OpenAI—the developers of ChatGPT—held the most successful private funding round in history, bringing in $40 billion. It dwarfed that, and broke the record again, in early 2026, this time raising $122 billion. The creation of hype, in many ways, is the product. Anything they manage to create after this is almost incidental.

But when it comes to AI, the things that are being promised make deep assumptions about things we do not yet even profess to understand nor agree on, things that are fundamental to the progression of these technologies, such as whether a mind is an emergent property of the brain, and if so, how and when this emergence occurs—and, most importantly for that industry, how we might even begin to replicate it. Technologist and cloud architect Dwayne Monroe, who has been working in this field for thirty years, summed it up neatly in an interview with Michigan Abolition and Prisoner Solidarity in late 2025 (which is well worth a read), saying:

The phrase, “artificial intelligence,” should be understood as marketing. We do not yet understand how thinking works in humans or any of our fellow creatures (this is why cognitive science is a serious area of study).

Considering this lack of knowledge, it’s absurd to say that tech companies have produced an artificial version of what is not understood.

This lack of a fundamental understanding is downplayed in every sphere; most people you speak to will just accept and repeat the tech industry’s claim that if we just ‘technology’ enough, intelligence will occur. It’s such a given that few people even think to question it. And there I am, philosophy-pilled, reading book after book on AI (we even own the AI book by murdering warmonger Henry Kissinger, for my sins), poring through dense engineering explanations and information about GPUs and scaling and still finding absolutely nothing that answers the question that I often scrawl madly into a book’s margins and on more than one occasion have barked out loud to an empty room, like a maniac: BUT HOW??!

It has been intensely sanity-restoring, then, for me to read Karen Hao’s truly incredible book Empire of AI. Hao is a journalist with an AI specialism; she graduated from mechanical engineering at MIT hoping to enter Silicon Valley and tackle climate change, before quickly realising that Silicon Valley had absolutely no intention of doing that. Instead she moved into journalism, being one of the key voices covering the rise of generative AI, and specifically the company OpenAI. Hao gained three days of access to the company two years before the release of ChatGPT; she was quickly shut out for asking questions the company did not like. She is uniquely positioned to clearly lay out what underpins the AI bubble: the belief that someone is going to achieve Artificial General Intelligence (AGI), or human-level intelligence, sooner rather than later. That we really can create a mind, by building larger and larger supercomputers—and that this must be justified at any cost.

It has been sanity-restoring, to read this book, but it has also scared me to my bones—not because I believe that the billionaire overlords who control our politics and our technology can actually achieve this, but because of the ways that their determination to do so entrenches our most destructive and supremacist tendencies, and undermines democracy, at the expense of everyone else. One of the most terrifying realisations of my adult life has been that the tech industry, now maniacally-centred on producing this fabled AGI—which it cannot define, and for which it has no practical roadmap, beyond the borderline religious idea that it will just occur when their current models are pushed to some mythically large scale—is willing to expend seismic amounts of money, climate-shattering amounts of energy, and indeed almost all the planet’s resources on their vague and contestable methodology, with no obvious end in sight. Or, perhaps more realistically, that they are maniacally-centred on convincing us that they believe in AGI, because they want to amass seismic amounts of money. And we, the little people, probably cannot stop them.

---

OpenAI HQ

ChatGPT arrived into the world with a bang at the end of 2022, and by January 2023 had become the fastest-growing consumer software app ever. Now, if you sit in a room discussing generative AI with a random group of people, you quickly realise that you are pretty much the only one not using it; some people seem to use it all day every day, for things they previously had no issue doing themselves: writing work emails, putting together their weekend to-do lists, writing itineraries for upcoming holidays abroad. Generative AI has been folded into pretty much every piece of software we use. I can’t so much as open WhatsApp without having to navigate off some conversation with an AI that I don’t want and didn’t start (and acting as a barrier to me engaging in my actual human relationships).

ChatGPT was an attempt to build a consumer use case out of what was, originally, a much bigger proposal. OpenAI, the company that produced ChatGPT, was founded as a nonprofit in 2015 by Elon Musk, Sam Altman and a small team of others, with $1 billion in funding pledged from a group including Amazon Web Services, comic book villain Peter Thiel and Musk himself; only a fraction of this money ever materialised, but Musk was insistent on the figure as a marketing tactic. It was not to produce any commercial products but was to focus on research, which would be open-source, hence OpenAI. Within two years they had decided that the project would require stupefying amounts of money; by the time Hao started covering them in 2018, Musk had left, and by 2019, they had created a for-profit company and abandoned any idea that their work would be open-source. That year they began partnering with Microsoft for an initial $1 billion; by the end of 2024, Microsoft had invested $14 billion into the company but had begun to disentangle themselves from it. In September 2025, OpenAI was valued at $500 billion, one of just seven companies that compromise a trillion-dollar AI market valuation.

Yet all of this was predicated on one initial idea: that AI could potentially murder people to achieve its goals. According to Hao, Musk was so worried about Google’s DeepMind project achieving an entirely hypothetical AGI, and this AGI subsequently getting out of control, that he accepted Sam Altman’s invitation to dinner to discuss it, and very quickly afterwards pledged a billion dollars towards their own counter-project, OpenAI. ‘The future of AI’, he said, ‘should not be controlled by Larry [Page, Google co-founder].’ Altman agreed that it was impossible to stop humanity from developing AGI, and had proposed that they should get ahead, to stop anyone ‘evil’ from taking control of it. As if this wasn’t deeply ironic enough, Altman termed this the ‘Manhattan Project for AI’, naming his mission after the US’s drive to create the atomic bomb, which resulted in the killing of almost a quarter of a million people at Hiroshima and Nagasaki and the start of the nuclear age, which has since plunged humanity into several existential crises of global proportions.

But at the same time, as Hao outlines, Sam Altman was fiercely ambitious; after selling his first (failed) company for $5 million at 26, he was disappointed, because Steve Jobs was worth so much more at the same age. For all the ways that Altman and Musk might have genuinely believed in the threats of AGI, they were perhaps more concerned with the real business of making themselves rich beyond measure. By 28, Altman was the president of Y Combinator, the start-up accelerator that arguably decides the direction in which major US tech can afford to go, bringing millions in investment to selected companies. It was his combination of ambition and belief in the possibilities of AI that convinced a group of people to join him, to become the team behind OpenAI. One of these people was Ilya Sutskever, poached from Google Brain (which would later merge with DeepMind to become Google DeepMind).

It is hard to grasp the depth of this group’s belief in the inevitability of AGI without reading Empire of AI; this is one of many reasons that I implore you to read the book. Hao shows over and over that if the co-founders were not convinced at the beginning, they became not only convinced but evangelical, as did many (if not most) of their employees. What’s not obvious from the book is whether this was a sincere belief, or a convenient one; like all religions, there are both true believers and those who understand that the power of belief can be harnessed to cynical ends, including the enrichment of those who guide the flock.

Whether sincere or not, this group was insistent on the public sentiment: that AGI is inevitable, and they must be the first to achieve this goal. Of course, there was one key thing missing: they didn’t have a clue how to get there. Because, of course, there is no clear way to achieve real artificial intelligence; we do not know how it occurred in us. We do not know where it resides, we do not understand it nor agree, really, on what it is. How does a company or a team create a construction map for something they cannot take apart? As Hao puts it, ‘OpenAI had no idea what it was doing.’ But Sutskever had an idea, and it was based on the possible level of ‘compute’ (which is just horrific tech industry terminology for ‘computational power’; having formerly worked for a North American tech startup as a copywriter, I really hate what these people do to language):

compute, Sutskever felt, was king. And if it were possible to scale compute enough to train an AI model at human brain scale, he believed, something radical would surely happen: AGI.

You would think, if you have been reading AI literature and following its claims and trends, that this was a given: that there was only ever one theory for achieving AGI, and it was based on scale. But this is not the case. It is not even accepted by all AI researchers that scaling is the best way forward, not least because of the intense resource depletion that is necessary to achieve it. There are various theories about how humans acquire intelligence, and these are carried forward in AI research; some think language is the key (hence large language models) and others think it is symbols that are most important; neurosymbolic AI, an alternative to the LLM+scaling method that companies like OpenAI follow, is much more energy- and data-efficient. But this has been sidelined by a maniacal commitment to scaling, the negative effects of which were obvious to the OpenAI founders back in 2019, when Hao gained unprecedented access to the company. Pressing for comment on the massive environmental impact of the company’s mission, Hao was told that the benefits of reaching AGI would include it coming up with a solution for climate change (which, of course, we already have—the issue is that no governments will commit to it). As she put it to Sutskever:

What you’re saying is OpenAI is making a huge gamble that you will successfully reach beneficial AGI to counteract global warming before the act of doing so might exacerbate it.

Of course, Hao received no answer to this. She would later be told that the biggest barrier to achieving AGI is that no one knows what it really looks like. And yet in the four years that followed her days with OpenAI, their models were massively, almost incomprehensibly scaled upwards. As Hao puts it, referring to the GPT model released in 2023:

Under the hood, generative AI models are monstrosities, built from consuming previously unfathomable amounts of data, labour, computing power, and natural resources. GPT-4, the successor to the first ChatGPT, is, by one measure, reportedly over fifteen thousand times larger than its first generation, GPT-1, released five years earlier.

In April 2024, the CEO of Anthropic told the New York Times that by 2025/26, the price of training a single competitive generative AI model would be between $5 billion and $10 billion. In late 2025, Sam Altman, CEO of OpenAI, made clear he plans to build the company 250 gigawatts of computing capacity by 2033, which could necessitate spending 10 trillion dollars in the next seven years (despite their 2025 revenue being a reported $13 billion, which, multiplied by seven, results in less than 1% of the ten trillion he’s committing to spending). The modus operandi at OpenAI, and at every company that wants to beat them to their goal, has been vast expansion. The more data they use to train their LLMs, they believe, the closer they will come to creating AGI. But this data requires processing power, computational power. It needs massive data centres, and huge tracts of land on which to build them, and it needs huge amounts of water—potable water, as Hao points out—to cool them. Hao outlines cases in which data centres use more drinkable water than the local population; she quotes proposals from Google that state they need to use 169 litres of fresh drinking water per second to cool their data centres in parts of Chile where the residents cannot rely on access to it. In the last few years, OpenAI and other tech companies have worked hard to cover up the catastrophic environmental impact of their processes, but this was not always the case. Hao quotes Sutskever in conversation with Cade Metz for the book Genius Makers:

I think that it’s fairly likely that it will not take too long of a time for the entire surface of the Earth to become covered with data centres and power stations.

And all of this, of course, towards a methodology that has absolutely no certainty of working. AGI remains elusive, and undefinable, and in the meantime companies have to justify the relentless spending. The reason that we have AI encroaching on every aspect of our lives is that OpenAI and the companies like it need to force the consumer end of their product into every place they can in the desperate hope that it will become profitable. All that is sure is that these companies will continue to expand, spending unconscionable amounts of money on infrastructure and computational power despite the fact that they are still in the red. Despite its $13 billion in annual revenue, OpenAI will not be profitable until its income is almost ten times that amount, and that’s before you take into account Altman’s ludicrous infrastructure plans. Even Mark Zuckerberg is talking around the possibility of AI being an enormous bubble that will soon burst, sending the stock market plummeting. As someone who graduated into the 2008 economic crash, I for one am tired of living through these supposedly once-in-a-lifetime catastrophes which negatively affect everyone except for the billionaires who have safe places to take their new money.

As it stands, the market returns on this insane investment in AI cannot justify themselves—and so the need to position the search for AGI in spiritual terms increases. Yet not all computer scientists buy it. I laughed out loud—a semi insane, sort of manic laugh—when Hao quoted a 2022 tweet from OpenAI co-founder Ilya Sutskever—’it may be that today’s large neural networks are slightly conscious’—and an expert’s response to it:

One DeepMind scientist, specialised in the study of cognition and consciousness, replied in the comments, ‘… in the same sense that it may be that a large field of wheat is slightly pasta.’

---

seriously though, read it

When it comes to things that challenge your connection to reality, there is a point of no return: you must either stop and accept that you’re very far from what would be considered rational or logical, or you must recommit yourself anew, refusing fact and severing the cord that once connected you to the real world. This holds true if your belief is not sincere: at some point your flock needs to be directed off the cliff of credulity, and you must be the one to lead them. When it comes to the search for AGI, it seems you must fall away from the project or you must commit yourself to believing, no matter what the evidence says.

As you read your way through Empire of AI, it dawns on you that so many of the people involved with AI research have fallen into it as a religion in and of itself. It is either an adjunct to their existing beliefs or commands a new belief system of its very own. Hao writes:

At Google that spring, Blake Lemoine, an engineer on the tech giant’s newly re-formed responsible AI team, grew convinced that the company’s own large language model LaMDA was not only highly intelligent but could be considered sentient. He said this was not based on a scientific assessment but rather on his belief, as a mystic Christian priest, that God could decide to give technology consciousness. ‘Who am I to tell God where he can and can’t put souls?’ he wrote.

There is a moment in the book Love and Sex with Robots, the 2007 treatise from David Levy in which he contends that we’ll be fucking robots by 2050, in which he states that if a robot tells you it is feeling something, you have no reason to believe that isn’t true. This has stayed with me for many reasons, and this same sensibility is apparent in the assertions made by those on the OpenAI team: the insistence that if you create a piece of software entirely to ape human behaviour, and it does exactly that, you must for some reason deceive yourself that there is a human-like mind inside of it. This is like putting your hand inside a puppet, turning to face it and, while your hand is still inside, deciding to believe that it is in fact independently mobile—not only mobile, in fact, but sentient (what we might call the Mr Flibble Theory of Mind). You can see the cogs, you can see the mechanism, but still you believe in something from the beyond working its magic. What is this if not religion?

And what must you do if you are the figurehead of a religion that is growing in notoriety? You must preach. In a September 2024 blog post, Sam Altman made the sort of claims that you might find coming from a person screaming on Glasgow’s Buchanan Street before handing you a pamphlet about the rapture, detailing how gays should be stoned to death before judgement day comes:

I believe the future is going to be so bright that no one can do it justice by trying to write about it now; a defining characteristic of the Intelligence Age will be massive prosperity.

Although it will happen incrementally, astounding—fixing the climate, establishing a space colony, and the discovery of all of physics—will eventually become commonplace. With nearly-limitless intelligence and abundant energy—the ability to generate great ideas, and the ability to make them happen—we can do quite a lot.

You can read these things as sheer marketing, or you can read them as the pseudoreligious ramblings of someone who has not only drunk his own Kool-Aid but is trying to convince the gathered masses to drink it too, and to pay him several billion dollars for the privilege. In June 2025, Altman wrote a blog post that began with the phrase ‘We are past the event horizon; the takeoff has started’, and went on to claim that ‘we have recently built systems that are smarter than people’. Despite the facts of the matter—there is no reason to believe that such systems have any intelligence whatsoever, yet we are spending untold amounts of money on them—Altman insisted that the rapturous advent of AGI is both here and affordable:

Intelligence too cheap to meter is well within grasp. This may sound crazy to say, but if we told you back in 2020 we were going to be where we are today, it probably sounded more crazy than our current predictions about 2030.

May we scale smoothly, exponentially and uneventfully through superintelligence.

These people are the preachers of the present, the cult leaders of today; it does not matter that the predictions they make never materialise, because they are always around the corner. The rapture is always on the horizon; it is always the end of the world tomorrow.

Do they really believe? Some of them, I’m sure, do: we live in a post-sci-fi world where powerful people take ideas straight from novels and think they can speak them into being. But structurally, their belief only plays into already-existing methods of capital accumulation. As Edward Ongweso Jr puts it, ‘the tech sector has undertaken a wildly successful and sophisticated marketing campaign that has drowned out skepticism and critical analysis’. He continues:

A coalition of hyperscalers, venture capitalists, fossil fuel firms, conservatives, and reactionaries are engaged in a frenzy of overbuilding, overvaluing, and overinvesting in compute infrastructure. Their goal is not to realise AGI or radically improve life for humanity, but to reallocate capital such that it enriches themselves, transmutes their wealth into even more political power that imposes constraints on countervailing political forces, and liberates capitalism from its recent defects (e.g. democracy), consolidating benefits to its architectures regardless of the actual social utility of the technologies they pursue.

There is every reason to believe that the reason that the scaling model of AI has been chosen and funded to a completely unhinged degree is not because it has the best chance of achieving this mythical AGI, but that it is the best possible method of stripping money and resources from the lower rungs of society and funnelling them upwards to the 1%. The modern tech industry loves to pointlessly reinvent what we already have, from Musk inventing the futuristic concept of ‘tunnels’ to Uber privatising buses; by the end of Empire of AI, Hao has successfully argued that OpenAI is primarily reinventing the East India Company, revamping colonial expansion and resource accumulation for the twenty-first century. And since the publication of the book, what’s become clear is that these companies are (to varying degrees as dictated by how they are able to publicly position themselves against the other companies) already entrenched in the military-industrial complex, where the funding is bottomless, and what Ali Kadri calls the true processes of capitalist accumulation—waste and death—are created.

I kept describing the experience of reading this book, to my partner, as having a low level panic attack for about a week straight. How else should a person feel when they learn that data centres have been kept running in the wake of hurricanes, when hospitals have been shipping patients out because they needed to turn the power off? Or when you learn that 900 acres of Arizona have already been bought by Microsoft to house some of these data centres, which then will consume untold amounts of water? Over and over again we are shown that corporate profit, US military expansionism and the sci fi dreams of billionaire maniacs are prioritised over human life. And then you realise there is next to no impact when you tell people these things, because this is the logic of capitalism—that it must be allowed to grow endlessly, and quickly, at the expense of literally everything else—and with AGI, it has met religion once again. Capitalism must run unchecked so that a mythical, unproven goal can be attained—and that consent for this insane goal, the creation of AGI at all costs, has been manufactured by the rollout of generative AI models that promise to alleviate people’s work load and cognitive tasks when all they will actually do is to bring more work, worse pay and horrific conditions. It feels very much that the argument has already been lost. Because the mechanisms that already rule our world are committed to making us believe they have the answers to questions we haven’t even begun to solve, and that no price is too high for gaining this knowledge.

In Karen Hao’s words:

AGI, if ever reached, will solve climate change, able affordable healthcare, provide equitable education. OpenAI is the poster child for this line of thought. It cannot say how the technology will deliver on these promises—only that the staggering price society needs to pay for what it is developing will someday by worth it.