I recently walked past an old minicab stand which, during our younger student days, sallied my friends back and forth between the city’s nightlife and our more affordable suburban digs. It is gone now; only a dilapidated and cordoned-off shack remains of a once-thriving minicab empire. Like thousands of others across the western world, the business went into terminal decline the moment Uber appeared in our lives. Yet a simple check of the app today shows the price of an Uber is not substantially less than a minicab ride back then. Uber rocketed to celestial heights because it was heavily subsidized by venture capital – driving the value through the floor, bankrupting all physical competitors, before it then jacked up the price to create an instant monopoly. And this model, I fear, awaits many of the rest of us in the white-collar world.
As in the early days of $3 Uber rides, many of us are intoxicated by the promise of generative AI in the form of near-free LLM use; chatbots which offer the service, as one friend of mine recently claimed, of “being able to talk to God.” Regardless of the extreme grifting and often explicitly fraudulent nature of much of the AI economy, it is difficult not to feel at least some degree of amazement at the sheer power of this software. It can revolutionize data analysis, generate plausibly written Master’s level academic work in seconds and even schedule complex tasks through agents – a breakthrough in computational achievement.
It’s in the tech world that people are living what Thoreau called ‘lives of quiet desperation’
Yet these marvels are overshadowed not only by the catastrophic cannons of slop but because almost everyone I know relates to AI with a sense of extreme trepidation. The sense that AI is “coming to take their jobs” is on almost every working person’s mind these days. Are we right to worry? Will AI act like Uber, do our jobs for next to nothing and, when we’re obsolete, name its price?
The grotesque amount of hype in the industry sounds a worrying note. The AI economy requires enormous front-loaded capital expenditure on GPUs and data centers, among others, all of which demands galactic levels of fundraising. Justifying these financials only works alongside a bigger cultural narrative that AI will transform not only the economy but almost everything about the modern world – including the destruction of almost everything done in front of a computer. CEOs such as Sam Altman (OpenAI) and Dario Amodei (Anthropic) are, fundamentally, in the business of fundraising – and pushing hysterical claims about the end of white-collar work might just convince those with deep enough pockets to keep the gravy train moving.
They also know from the financial crisis of 2008 that there really is such a thing as being “too big to fail”: once a critical amount of the American economy is locked into the AI bubble, the government will just bail tech companies out if things go wrong. The obvious fact is that AI really is already affecting people’s jobs in quite substantial ways. Everyone from copy editors to cartoonists in my network are hemorrhaging business to chatbots which can simulate “good enough” versions of whatever blog post or image their clients once paid reasonable money for. Yes, ChatGPT-produced writing is awful, and yes, AI-generated images are lame, but for most businesses that simply doesn’t matter – only cost saving does. There will always be compensation for top-grade artistry, of course, but most creatives pay the bills not with their highest quality output, but the daily rote of their craft. This commercial opportunity is fast disappearing for writers and graphic designers; I can feel it myself.
Over in startup-land, AI has transformed many well-remunerated roles almost overnight. Anyone working in software engineering has had years or even decades worth of their investment in skills evaporate in an instant with the release of Claude Code – Anthropic’s LLM which can instantly generate almost any piece of software through judicious prompting from its human operators. Many in the industry are apocalyptic about this change and more miserable than I’ve ever seen them.
“Disruption” sounds great in investor decks, but here it means the mental anguish of millions of people fearful they won’t be able to keep a roof above their families’ heads by this time next year. It’s in the tech world today that people are living what Henry David Thoreau called “lives of quiet desperation.”
Which takes us back to the specter of Uber. Increasingly, staff at tech companies are being told they must use AI. They are given a set amount of “tokens” – units of data, approximately parts or whole words of text, used for processing by the models for predictive purposes – and judged by their managers at the end of the month. They get in trouble if they do not use enough tokens, running the risk of singling themselves out as politically resistant to the “efficiency” and “productivity” revolution that awaits anyone ambitious enough to use a chatbot to draft much of the busywork that keeps corporate office life ticking along. But many tech workers have even started burning the tokens on pointless tasks, “tokenmaxxing” as it is now called, in order to bump up their “AI adoption” metrics. One friend, a journalist, believes his company is training a model on his writing to eventually replace him. He’s not the only one. Almost everyone being forced to use these products thinks something similar is going on – and with Uber as a precursor, it’s obvious why.
Yet the most unnerving part of this story, in fact, is that even if we accept that ChatGPT can entirely replace you, hiring human employees is still far cheaper than getting a chatbot to do all the work. AI models, which are often free or available for something like a measly $20 per month, are enormously, monstrously subsidized by venture capital. If they were being sold at their genuine market rate, tech companies would very quickly go bankrupt.
So why keep giving it out so cheap? Well, as my old minicab owner friend might contend, it could be because AI companies are incentivizing corporations to replace their human workers with AI, promising them a cheaper, easier, more docile management experience. Then, once the meat-robots are out and the cyberbrains are in, they can jack up the price again. Uber was subsidized to the tune of $13 billion before it went public – but the AI economy receives $200 billion in investment each year. The consequences of such an exponential bet are simply unprecedented, even if the singularity does arrive sooner rather than later.
One friend, a journalist, believes his company is training AI on his writing to eventually replace him
Personally, I’m reluctant to call conspiracy on too much about what’s happening in Silicon Valley. I’m skeptical about the jobs apocalypse. But LLMs, although incredibly useful for many tasks, are still probably a net loss for humanity. They seem to me an invention of the same sort as plastic: a revolutionary advance for shopping bags and food wrapping, the downside being that they pollute the oceans which produce the fish to wrap up in the first place. But sticking with woven bags and freshly picked produce would mean indulging less in our near-infinite capacity for greed, new things – and sushi delivered in less than 40 minutes. So screw the ocean.
With AI, we have an internet “ocean” increasingly polluted with mindlessly generated drivel, inauthentic agent-bot networks, deepfakes. The ultimate outlet for a mentally atrophying humanity. Couldn’t we just cooperate in person and write our emails the old-fashioned way, rather than spending our brief time on Earth “optimizing workflows” for virtual robots merely to further enrich the owners of capital? Yes, possibly. But LLMs are cheaper, at least for now.
Comments