I’ve been working off and on for months on my own position regarding the use of artificial intelligence both at work and in my everyday life. I’ve seen other essays of this type, but have largely avoided reading them just so I’m not regurgitating other people’s takes. (An AI hallmark!) This is about 2000 words, so TL;DR AI bad, no thank you. But whether or not you, yourself use it, I hope you’ll read and understand that for many of us, this isn’t a knee-jerk reaction or a culty, cosmic doomer rant. It’s about money, power, and quality of life for me and the people I share a planet with.

Every discussion of AI requires some situating. In this post, when I write “AI” I mean “generative AI,” which excludes things like computer vision, speech recognition, translation, and other, largely older, models that do not necessarily rely on large amounts of compute or tons of pilfered intellectual property. More specifically, I’m talking about large language models (LLMs) like Claude, ChatGPT, Gemini, et al.

I’m relatively fine using a model like Nvidia’s Parakeet, for example, on my own machine, to transcribe or translate audio content (though I will also do an edit pass of my own to make sure any captions or transcripts I output are correct). The difference can be enormous, in terms of the fully-loaded compute cost (including training): a general-purpose model like ChatGPT can do lots of tasks, but it will usually do them more poorly and use more GPU time than a custom-built model. Wasting compute is, in a word, bad. At least that’s what I learned in computer science classes.

Each version of each model is also tied to an enormous training bill, which can cost hundreds of millions or even billions per run. Even if you’re just downloading a model off of Hugging Face to run on your own machine, it’s still a product of the original sin of millions of hours of GPUs running full blast in a data center, training on stolen content, adding noise and pollution and draining useful water from humans living nearby. Using these models means you not only accept the added electricity you’re using to ask it a question, but also a share of all the model training costs that led up to it, not to mention the infrastructure being built to train newer and even larger models.

The AI industry is a state machine that so far has turned over a trillion dollars of capital and millions of tons of natural resources into lies and heat.

What AI is really for

In 2024, I was chief technology officer of a startup that was trying to close a funding round. As CTO, it was my responsibility to determine the extent to which we should use generative AI. The answer was obvious. It wouldn’t help us at all. There was nothing we were doing that couldn’t be done in a spreadsheet. And that’s fine: a system doesn’t have to be made unnecessarily complex for it to be valuable. Any student’s SAT answers could fit on a few punch cards, and yet millions of college admissions over the years wouldn’t have happened without them.

What we found during those meetings with angel investors or incubators was that they couldn’t care less about whether generative AI was a fit for our needs. All they knew was that gen AI companies were getting eye-popping valuations, and they wanted to push their chips in on companies that sounded like they would get eye-popping valuations, too. There were funds that would offer us six figure sums in compute credits, in exchange for a chunk of the company. But investing enough to keep a half-dozen employees going for another six months, even if it could be profitable inside a year? No thanks. Why get a mere double or triple your investment when you have a small chance at an OpenAI-sized multiple?

Generative AI serves one more very important purpose to investors, that I think has gone largely unexamined. Why would they give us compute money, but not salary money? It turns out to be pretty straightforward. It’s messy, employing us meat bags. Not only do you have to pay us once or more every month, we also need health insurance, and sick time, company swag, even the occasional pizza or donut. More importantly, startups pay less in salary in exchange for offering shares in the company. Both investors and executives keep an eye on the company’s capitalization table to make sure that, as the company grows, their potential payout isn’t diluted away to nothing.

However, if you don’t have employees, just a bunch of agents running riot, there’s nobody to pollute your cap table, and your founders and investors don’t have to share. The compute-based funds just take equity for that transaction, cutting out all those pesky, needy, equity-hogging workers. Put simply: generative AI is designed to ensure that those really big payouts are divided between founders, venture capitalists, and infrastructure providers, cutting the working class out of those big paydays they occasionally ended up earning. So inefficient, wasting capital on the unwashed. It’s also no accident that thousands of the layoffs pinned on AI are mid- to late-career employees of large companies—exactly the kinds of people with large amounts of unvested stock grants and options that companies can claw back.

Put it all on double zero

It was during that fundraising round, which ultimately failed, that I realized AI, to most of the capital market, is merely a lottery ticket. They don’t care how it works. They don’t even care if it works. What they care about is that, when you talk about AI, you have investors’ attention. It is the largest growth markets, nearly the only one, in an otherwise dismal market environment. AI must succeed, because if it falters, there is no Next Big Thing to take its place, and the long-overdue market correction will destroy trillions in net worth. Those who built their stacks by hyping AI (looking at you, Sam Altman) or who are highly leveraged on AI assets (and you, Larry Ellison) risk being wiped out entirely. It must succeed, because for them, there is no plan B.

I think this explains why much of the harm brought on by generative AI goes without examination, much less meaningful attempts at repair. That’s nothing new in tech, but the direct relationship between gen-AI outputs and injustices in the real world makes the apathy, or antipathy, toward corporate responsibility more legible. It’s one thing when a company can’t or won’t explain why its C-staff is all white and/or all-male; it’s another when you can ask a company’s large language model how to get away with discriminatory hiring practices, and it will cheerfully tell you.

Where do you draw the line?

Having followed the progress of generative AI since about 2020, including working with one team that developed a current model, I have been searching for a bright line, one where I can say that this subset of AI usage is ethical, or at least ethical enough for me to use. I have reached the point where I don’t think that line exists. As a researcher, I could use it for the purposes of providing a full and accurate critique. But as far as using it to write for me, or code for me, or design for me, or even talk to me, I cannot use it while maintaining my ethical standards.

I also have to confess here that I do not always maintain my ethical standards. I still use the services of a number of companies that I don’t want to support, as an example. However, I’ve determined that my need to remain connected to my parents, extended family, and to my community in Mexico, overrides my desire to rid myself of any connection to Meta platforms. Still, I don’t have a real need to make exceptions for generative AI. It doesn’t materially improve my life, so I don’t feel a need to compromise my values in any way.

Since Google’s search engine now inserts AI slop into every request, I have switched to an independent search provider. I am able to avoid triggering AI at least directly there, and the results are so substantially better than Google’s now that it’s worth the move. I also have AI turned off in my Gmail account, and eventually may migrate off of Google platforms entirely if the onslaught doesn’t stop. It is getting more and more difficult to walk a non-AI path when it seems in all directions the floor is lava, but if in the end it leads to less dependency on computing devices, that’s not a bad thing in itself.

By the way, and this should go without saying, but no AI was used in the production of this or any other issues of this newsletter, except for one image I generated two years ago, lampooning a certain retired AI booster. (We are none of us perfect.) I’m going to keep avoiding it for my own purposes, though as long as I stay up on tech, at least interfacing with it for work seems increasingly unavoidable. I don’t see anything that will change my position on the overall utility or desirability of AI in the short to medium term. As long as I have a choice, and the economic means to keep a roof over my head and a set of reasonably firm values, I am opting out of first-person interaction with AI. And if any of my situation changes for the worse, before I accept AI as my lord and master, I will go down swinging.

If you happen to be working somewhere that’s sold on AI, particularly one where words like tokenmaxxing are used unironically… I get it. People who have worked in a place that suddenly “pivoted to AI” often feel like the frog in a rapidly-warming pot of water. It’s not what we signed up for when we got into web design or project management. And I can understand that the change is happening on internet time, not on career time—in other words, the mandates are coming faster than your ability to find a new job as a bus driver. Statistically, you probably use AI at work. You might like it. You may even need it to keep your last pinky on the cliff’s edge of your tech career. But it’s worth confronting the truth: on balance, this is a bad deal. It’s more likely than not, in my opinion, that the collapse of an AI bubble will lead to an economic depression, and as these services companies are coming to depend on start to shutter, it will fall upon those of us who are still rubbing organic, artisanal neurons together to build what comes next.

I don’t think of the possible collapse of the AI stack as doom. Honestly, I think where we’re at, in no small measure due to AI but also to that founder/VC/institutional ouroboros that’s tied up all the capital today, kind of sucks. But if this ship is going to sink, I do not wish to tie myself to the mainsail. One thing that has become clear is that if we’re going to find a way to keep people alive, healthy and active in society, AI is not going to help us get there. It is merely a vehicle to extract capital from one end, in the hope of eventually collecting revenue from the other. And it has only proven itself to be good at the former.

Being one person with an axe to grind about AI feels like spitting in the ocean. But I really don’t love where I think this is going. I don’t even love the idealized world many AI boosters have, of AI teachers, AI doctors, AI farms. It’s not a future I want.

So, you can have this revolution. I don’t want to help make billionaires into trillionaires, or businessmen into nation-states. My ambitions are now much smaller—clean water, edible food, happy friends, stable governments—but they’re more important to me than helping to turn the world’s economy into a Monopoly game.