Join me, if you will, on a rapid passage through the varied ambiences of AI. Our entry point is the architecture at the heart of all of contemporary artificial intelligence, the artificial neural network. Absorbed into one of these vast and cavernous computational abstractions, the structure arrayed before us is layer upon layer of stacked units, each connected to counterparts in the previous layer by shimmering threads of digital signals, each unit pulsing as it boosts the signal to identical artificial neurons in the next layer. Stacked thousands of stories high, the network is as dizzying as a 19^th century power loom scaled to the size of a skyscraper. We pass beneath as it’s being trained, and the crashing symphony of calculations washes back and forth overhead until the final model emerges with optimised grandiosity.

And yet, as we try to discern the emergence of intelligence from this frenzied activity, all we hear are mumbled repetitions and distorted echoes. It seems that the function of the mighty machine that surrounds us is mere pattern recognition; that the purpose of chaining together myriad signals in giant matrices is simply a ping of recognition at the end, or at most an emulation of what came in as training data. If we entered this geography in the spirit of explorers, searching for the source of a deep river of emergent cognition, we are momentarily and somewhat vertiginously stranded. Can AI simply be a giant correlation engine, a remixer of learned patterns, void of any actual intelligence?

Still seeking clues, we step onto one of the passing scrapers that gather data on behalf of our AI and are swept through such a greater variety of internet arcades than any flâneur could possibly imagine. Our steed scoops up everything we pass through, from the wood-panelled and carefully tended libraries of Wikipedia to the gossiping and often sordid taverns of Reddit. It smashes into art collections, seizing everything without asking, then forges out to the urban wastelands of moribund social networks and abandoned personal blogs that flap in the breeze like forgotten flyposting. Here in the raw data, we see the evidence of intelligence, broadly defined, but it’s still unclear by which means it is ordered for the process of pattern recognition; how mathematical formulas are separated from racist graffiti, or how images of traffic lights are demarcated from those of bare street corners.

Responding to our confusion, our psychogeographic shuttle fires us out of another portal in the megamachine, where we race at two-thirds the speed of light along submarine cables to Africa to emerge in a city block in Nairobi. Every part of the building is filled with the crowded cube farms of local workers toiling with mouse and screen in the heat of the poorly air-conditioned afternoon. Stepping carefully along the narrow gaps between the rows of workers we observe that some are indeed labelling images with attentive pedantry, painstakingly tagging every object in frames from a dashcam video, demarcating pedestrians, bicycles and roadside trash, adding to training data for self-driving cars they will never get to ride in. In their well-thumbed printouts of guidance, they’re told that images of people lying on benches should be labelled as “sitting” because the existence of homelessness is not relevant for this purpose. The more unfortunate workers, perhaps in a separate floor of the tower block, are filtering out the worst descriptions of bestiality, child abuse and fascistic fantasies from samples of text intended for large language models like ChatGPT. The results are used to train a different kind of AI called reinforcement learning, which stands guard like Cerberus at the gates of the subsequent training process. These workers’ efforts are paid at $1.30 an hour and will leave them deeply traumatised with no recourse to counselling or support.

So far, our dérive has exposed us to a chilling emptiness inside the models underpinned by the morally-compromised backstreets of data scraping and crowdsourced labour. Somewhat shaken, we strike out for the places where AI is actually applied. Surely, on the application side of our abstract city there must be a carnivalesque parade of image and text generation whose bubbling potential for good relieves at least some of this somber ambience. Getting there, it turns out, means passing through the terraced streets of applied machine learning, even as they are being cleared to make way for AI’s soaring towers. In these narrow passages we find the recipients of welfare benefits sitting disconsonantly, clutching algorithmically-generated letters accusing them of fraud or worse. In passing we note how many seem to be people of colour, but we’re unable to process this as, further down the street, police officers are kicking in more doors on the recommendation of the machine learning. It’s a relief to arrive at the futuristic towers of AI where a facial recognition system opens the sliding glass doors for us as we approach. Settling into the well-lit reception area, we sit in comfy modernist chairs to watch the welcome video. At last, the AI we were promised; a flow of patients is examined by benign-looking clinicians who, consulting their AI screen, direct for them to be wheeled to the next stage of their healthcare journey. But wait; according to the brochure, this AI is optimised for the healthcare insurance company and is alerting the hospital to the earliest plausible moment that patients’ care can be completely cut off.

Throwing the glossy brochure to the floor we flee from our virtual journey to the material reality of a nearby school, certain at least that we will find smiling young faces whose learning potential is being unlocked by the personalised attention of educational AI. Stepping carefully as we enter to avoid the collapsed roof which the school’s already-stretched budget can’t fix, we can’t help detecting an air of desperation. Harried teachers are muttering “well, it’s inevitable isn’t it” as they read the latest management guidance about applying AI to generate lesson plans, one of the last parts of teaching they actually found fulfilling, while students are messaging each other about how to get AI-generated assignments past AI detection software, software which in any case ends up accusing those whose first language isn’t English or whose neurodivergence makes their writing style more stilted than average. Loitering for a moment in an actual lesson, where a passionate young teacher is talking to students about climate change, we’re momentarily struck by her mention of AI. What was it she said – will AI save us from climate change, or will all the data centres make it worse? That sounds like real psychogeography; a drift around the computational cathedrals of AI data centres whose well-lit publicity images show server racks in converging lines like an early Renaissance painting’s experiment with perspective.

Taken to a data centre by an obliging Uber driver, who tells us on the way how his take-home pay is being progressively eroded by the AI behind Uber’s app, we’re surprised that our first encounter is with serried rows of industrial cooling towers, each emitting tendrils of water vapour into the autumn air. Entering the anonymous looking warehouse of the data centre, through several layers of humourless security checks, we immediately understand. So many computers! Each running at full blast! Never mind the gentle warmth that a laptop shares with our thighs as we sit and type, this is fully overclocked heat generation of acres of computers along with their cabling and power supplies. No wonder the place has to pump in thousands of gallons of cooling water a day to keep it from melting. And the noise! The chorused hum of thousands of cooling fans emanates outwards from the data centre, mostly staying below the civic regulation meant to limit late–night parties, while being as persistent as a distantly circling jumbo jet, generating nausea, migraines and depression in unlucky local residents. Have we really found the much-fabled Cloud? Unlike actual clouds, which transport life-giving moisture from the seas to the land, this one clearly sucks up scarce water resources at a voracious rate. Perhaps, instead of “the Cloud,” we should call it “the Drought?” Exiting with relief from a side door, we discover that this data centre has its own electricity substation. A passing worker tells us this is one reason why the building is on the site of a former steel plant, which had its own connection to the grid. “That, and the way the local authority gave us a 10-year exemption from local taxes!” he says with a wink. The lure was a promise thousands of local jobs for this deprived post-industrial area, but it turned out they were mostly just to build it and only a hundred or so engineers like him are needed to run the place. He had been hoping to move to a new housing development to be near work, but that’s on hold as there’s no capacity left in the local electricity supply after the data centre has taken its cut. Heading back inside, he says with a final wink and a cynical chuckle, “Never mind the eye-watering CO2 emissions, we’re covered by carbon offsets!”

Rediscovered by the PR flack whose job it is to herd random psychogeographers who turn up at the gate, we return to the visitor centre and are handed VR goggles to watch a promo video about the brave new world of data centres that the company is establishing at breakneck speed. “Growth is what drives the economy, after all!” he bleats. “But why so many?” we naively ask, briefly removing the goggles, and he runs the numbers for us – the exponential growth in the model size of generative AI and the corresponding explosion in demand for computing power to train and run them. “After all, it’s not for nothing that Nvidia’s share price overtook Apple and Google last year” he says. Admitting our confusion, he points to the glittering chip in a glass display case at the centre of the room; one of the Nvidia H100 GPU microprocessors that are needed in their hundreds of thousands to keep the world of chatbots and image generators afloat. Re-donning the headset, it briefly glitches to show pixelated footage of coltan mines in the eastern Democratic Republic of Congo, where smoke from armed conflict scars the horizon even as the miners hack from the earth the minerals vital to the functioning of GPUs. Wrenching off the headset we run from the bland, low-rise building, hidden in this industrial estate, out past the thrumming wind farm whose renewable resource is fully diverted to boost the data centre’s green credentials, and onwards, panting and breathless, back to the city itself.

Our psychogeographic dérive ends, as so many before it, in a canal-side public house. This one is called “The Sir Geoffrey Hinton” and the pub sign rocking gently in the breeze shows a gold Nobel medal with the inscription “No Insight Here”. Drowning our disillusionment with a pint of the landlord’s finest, we reflect on the day’s events. Can it really be that AI’s giant edifice of sci-fi claims and venture capitalist investment is nothing but a spectacle, completely unable to fix any of society’s obvious failings while intensifying existing injustices in the process? Staggering to the toilet, we spot a scrawled message the cubicle door; “AI is not a collection of algorithms, rather it is a social relation among people, mediated by algorithms” – signed G. Debord. If our marker pen philosopher is right, we may need to construct some situations to disrupt this apparatus. As soon as we leave the refuge of the pub we’ll be plunged back into a world where each datafied action we take risks crossing the invisible decision-boundary of a predictive algorithm, where we never know which apparently human interaction is steered by a script from generative AI, where, indeed, we don’t even know if the regional accent on the other end of the helpline is being synthesised by an AI agent. Outside is where a government desperate for automatised productivity and the populist vote is fine-tuning its algorithmic systems to target migrants and other minorities, cutting costs through preemptive AI pathways that designate some lives as relatively disposable, oblivious to the lingering smell of smoke from recent pogroms. Returning to our seat in despair, we pass by the pub’s back room. Its door is slightly ajar and we can hear the hum of purposeful conversation from what sounds like a large group of people. A cheerful if ruddy face appears abruptly around the door. “Cheer up mate” he says, “it may never happen. Come and join our jolly banter. My names Ludd, by the way, but you can call me Ned.”