The Great Unreality

There is a particular kind of photograph that has begun to dominate LinkedIn. A man in his early forties, jaw too symmetrical, teeth too white, standing in front of an office window that looks out onto a city that does not exist. He is wearing a suit that fits him the way clothes only fit in advertisements. Beneath the image, a caption begins: I used to think success was about hustle. Then my grandmother said something that changed everything. What follows is eight hundred words of confected wisdom written by a machine, posted by a man who did not write it, read by an audience increasingly aware that none of it is real.

This, more than the keynote speeches and the obesity cures, is the actual texture of the AI age. A man who is not quite a man, telling a story that did not happen, to an audience that is not entirely human, in front of a window that has never existed. And we are told, by people whose personal wealth depends on the story holding together for the next four years, that this is the future, and that the alternative is to be left behind.

Left behind by what, exactly?

The money doesn’t work

Begin with the economics, because everything else follows from them. The hyperscalers, Microsoft, Google, Meta, Amazon, are collectively spending somewhere in the region of three hundred billion dollars a year on AI infrastructure.

That is more than the upper estimate of what economists say it would cost, each year, to end extreme poverty worldwide.

All of it being spent on data centres whose primary commercial output, as of this writing, is chatbots that write LinkedIn posts and image generators that produce the photographs described above.

The defenders will tell you this is investment in a transformative future. Perhaps. But the chips inside these data centres are not fibre-optic cable laid in 1999, sitting patient in the ground, waiting for YouTube to be invented. They are GPUs, among the most rapidly obsolescing pieces of hardware in computing. A graphics processor bought today is expected to be economically uncompetitive within five years or so, not because it has broken, but because the electricity bill required to run it will exceed the value of what it produces. The bubble in this technology, when it comes, will not leave behind useful infrastructure. It will leave behind warehouses full of scrap metal and very large electrical substations.

The investment thesis, then, requires the revenue case to materialise inside a roughly five-year window. The internet took about a decade. The personal computer fifteen years. The mobile phone twenty. There is no clear historical precedent for a technology of this scale repaying its capital expenditure on the timeline its financiers require. None I can find, anyway. And the people inside the industry, behind closed doors, know this. They are not building for the medium term. They are building for the narrative, for the next funding round, the next earnings call, the next quarter in which the appearance of inevitability has to be sustained long enough to sell stock.

That is not the financing pattern of a transformative technology. It is the financing pattern of a confidence trick. The dot-com bubble at least had the decency to crash before it consumed the energy budget of a small country.

What the tools actually do

The honest case for AI, stripped of the millenarian rhetoric, is that it produces a measurable productivity gain, perhaps ten to twenty percent, on a narrow set of bounded tasks. Drafting routine emails. Writing basic code. Summarising documents one would otherwise have to read. This is useful, but it is also by historical standards, modest. The spreadsheet did more. The word processor did more. None of these technologies required reorganising the global electrical grid.

But the modest, true version is not the version being sold. What is being sold instead is that this technology will replace lawyers, doctors, teachers, writers, designers, and the entire white-collar class (and the blue collar jobs with AI robots) within the decade. That human creativity is a soon to be obsolete biological accident. That version requires the technology to do things it has not yet demonstrated it can do, on a timeline nobody has yet justified, funded by capital nobody has yet shown how to repay.

In the gap between the modest truth and the grandiose pitch, an entire industry of slop has bloomed.

The aesthetics of fraud

This is where the LinkedIn man with the symmetrical jaw comes back into focus. The reason his image is unsettling is not that it is fake. Humans have been faking photographs since photography was invented. It is that it is fake without conviction. Fake without art. Fake in the manner of something assembled by a process that does not know what a face is for, generating an image that does not know what a man is, to be consumed by an audience the algorithm does not know is human. Fakery all the way down, and nobody involved in the production has stopped to ask whether the result is worth looking at.

The same disease has spread to the prose. The grandmother who said the thing that changed everything. The five lessons learned from a failed startup. The thread that begins, “Yesterday I fired my best employee. Here’s why.” All of it written by the same machine, in the same voice, with the same false-intimate cadence, the same fabricated specificity, the same final line that lands like a verbal punch from someone who has never thrown one. Counterfeit feeling, manufactured at industrial scale, by people who have decided that authenticity is a friction cost to be optimised away.

I have also begun to notice something in psychotherapeutic work that I think is connected. Younger patients, late teens, early twenties, increasingly describe their internal lives in the manner of the content they consume. The sentences are shorter, and the self-narration is more polished. The reflections arrive pre-packaged, as if rehearsed for an audience that isn’t in the room, and I do not think this is always a coincidence. If you grow up swimming in synthetic emotion, your own emotional vocabulary starts to borrow from it. The slop economy is not just a marketplace; it is becoming an inner voice.

The authenticity reaction

Every aesthetic excess produces its opposite. The ornament of the Victorian drawing room produced Bauhaus. The slickness of mid-century advertising produced punk. The polish of stock photography produced the entire visual language of independent journalism in the 2000s, with its grain and its handheld camera and its insistence on the visible imperfection of the real. We are now living through the gilded age of synthetic content, and somewhere, in a notebook, on a four-track recorder, in a darkroom, its opposite is being prepared.

You can already see the early signals. The resurgence of film photography among people who were not alive when film was the only option. The premium being placed on handwritten correspondence. The growing audience for podcasts in which two people simply talk to each other, with no production, no script, for three hours at a time. Writers who publish their drafts with the typos intact, because the typos are now the proof that a human was present.

I would guess, though I cannot prove, that the market for the unmistakably human is going to expand rather than contract as AI-generated content saturates the rest. A handwritten letter is going to mean more in 2030 than it did in 1990. A photograph taken on actual film, of an actual face, in an actual room, will become the new luxury good. A novel written slowly by a person, with the marks of struggle still on the page, will command a premium over the novel generated in four minutes. Not because the human version is necessarily better in some measurable way, but because it is evidence. Evidence that someone was there. Evidence that the time of a living person was spent on the thing you are now consuming.

The companies pouring three hundred billion a year into this have not priced this reaction in. They cannot. The reaction is, by its nature, a refusal of the categories in which they think.

What comes after the bubble

The technology will not disappear. It will find its proper, modest scale. It will be a useful tool for bounded tasks, deployed quietly, in the background, the way spellcheck is deployed now. The data centres built for the speculative version will be partly stranded, partly repurposed, partly abandoned. The people who became billionaires on the way up will mostly remain billionaires on the way down, because that is how these cycles work. The people who lost their jobs in the meantime will not get them back. The carbon released to power the buildout will remain in the atmosphere.

And the rest of us will be left to do what we always do after a great craze. Take stock of what was real, what was hype, and what, if anything, was gained. Rebuild the institutions of trust that the slop economy corroded. Learn, again, to recognise the human in what we read and see, and to value it because it is rare.

The man in front of the impossible window will fade. The grandmothers who never spoke will fall silent. And the actual people who were here all along, flawed, slow, expensive, irreplaceable, will get back to the work of being alive.

The ones who lost their authenticity will be the ones left behind.

Suggested readings:

• The sharpest single piece on why the AI bubble will burst, by the journalist who has been ahead of every major beat on this story. Indispensable on the financial mechanics. Zitron, E. (2024). The Subprime AI Crisis. Where’s Your Ed At. https://www.wheresyoured.at/subprimeai/

• A key essay on what large language models actually are: a compression of the web, dressed up as comprehension. Chiang, T. (2023, February 9). ChatGPT Is a Blurry JPEG of the Web. The New Yorker. https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web

• The clearest argument for why the slop reaction is coming, and what it will mean to prove you are human on a web flooded with generated content. Appleton, M. (2023). The Expanding Dark Forest and Generative AI. https://maggieappleton.com/ai-dark-forest

• The establishment view, from inside one of the institutions most exposed to the trade. Goldman Sachs’ head of global equity research argues the technology is not built to solve the problems that would justify its costs. Goldman Sachs. (2024, June 27). Gen AI: Too Much Spend, Too Little Benefit? Top of Mind, Issue 129. https://www.goldmansachs.com/insights/top-of-mind/gen-ai-too-much-spend-too-little-benefit