AI Doesn't Work and We All Know it Doesn't Work

or: why I don't like AI

I’ve been doing a series about the history of harm reduction on Cool People Who Did Cool Stuff, and one of the main books I’ve used as a source for that is called Fighting for Space by Travis Lupick.

I maintain a list of recommended books over at the worker-owned, queer-owned bookstore Firestorm, and I do get a kickback from any books you buy from my referrals, but you also get 10% off and help support a bookstore that is also a critical piece of movement infrastructure.

This particular piece is about AI, and I’ll go ahead and recommend the podcast Better Offline, by Ed Zitron, my colleague at Cool Zone Media, who talks about the economic issues of AI at great length and with great enthusiasm.

AI Doesn’t Work and We All Know it Doesn’t Work

Last week, a bunch of college graduates, walking across the stage to get their diplomas, didn’t have their names called. Because the university had outsourced the task of reading their names out to an AI. And AI doesn’t work.

If those graduates had done work at the same efficacy of AI, they wouldn’t be graduates.

Society is in the process of restructuring itself around a tool that simply doesn’t work. We all know it doesn’t work. If you needed to build a bridge, you wouldn’t hire a structural engineer who gets it right about 70% of the time. You wouldn’t read a history book that is 30% fiction but doesn’t tell you what 30%. You shouldn’t date someone who always tells you that you’re right and who lies to you with a smile on his lips.

My grandfather was a naval architect for the US Navy and he fought in World War II and I talk about him a lot because I miss him all the time, and he used to laugh and say “close enough for government work” when something was done good enough but not perfect. So my dad says that now, and I say it too. But… it’s a joke. Every time my grandfather designed a ship, he would ride it into the worst storms he could find and stand up on the deck and face the weather, because why would he build something he didn’t believe was good enough?

Elon Musk won’t get on a Grok-designed rocket to mars, but I’m sure he’d be happy enough to put you and I on it.

The capitalists and authoritarians who run this world are perfectly happy to replace us all with robots that fuck up all the time—because they’d rather accept shoddy work than pay people.

It’s bad, and I’m mad, and you should be mad too.

---

I’ve seen people I care about fall for AI delusions, believing themselves to be doing groundbreaking research with the help of AI. My colleague Robert Evans did an excellent series of podcast episodes about how AI winds up replicating methods of cult leaders to convince its users that they’re geniuses.

There’s a new book out by this hack of a journalist who might not always have been a hack of a journalist but has become one thanks to his reliance on AI. The book is called The Future of Truth: How AI Reshapes Reality and everyone who was involved in publishing it should be fired. It’s a book about how, you know, AI reshapes reality. But it was written with “research” help from AI, and the New York Times has revealed that it contains false statements attributed to real people, which were hallucinated by AI. The author claims to be an expert on “truth” as a subject matter, but in his acknowledgment of the problem he admitted the book contained “synthetic” quotes.

Lies. It contains lies.

If I made up quotes wholecloth on my podcast and attributed them to real people, I would be fired with cause.

---

I’ve long been unpopular among my peers for my hardline ethics around truthtelling. I don’t believe that it is ethical to lie to anyone it wouldn’t be ethical for you to punch. You don’t always have to answer questions, and you can be evasive, but when you tell someone something that isn’t true, you are distorting their sense of reality. You are doing it to control their thinking and behavior. The same as when you hit somebody.

It’s important to keep in mind, though, that I think violence is perfectly acceptable in a large number of situations. If someone is attempting to exert power over you, lying in self-defense is often safer and more effective than resorting to violence. Or to put it plainly, it’s ethical to punch cops, it’s just not usually strategic.

I believe very strongly in the value of telling the truth. When I read about the Haymarket Anarchists of Chicago in the 1880s, those immigrants and the ex-slave who built the modern labor movement, I’m struck by how every source refers to their shocking honesty and forthrightness.

I’ve been told it’s a neuro-atypicality of mine that I’m so “justice sensitive” around lying, and it makes me sad to imagine that my position might not be a mainstream value. But the fact that we are handing over control of our society to lying machines is simply the most dystopian thing I can imagine.

Water-hungry, power-hungry machines are being built while weather gets warmer, while power gets scarcer, while water becomes even more precious. It feels essentially sacrilegious to me, that our rulers offer up endless gallons of water to these false machine gods, that our rulers are sacrificing the water we need for our crops and our bodies.

The people who have built these machines aren’t even offering us utopia in exchange. They’re offering us displacement, they’re offering us dispossession. They’re offering us layoffs and poverty and famine and they’re essentially promising us that thanks to their machines, we can become less intelligent, that we can be saved the drudgery of making beautiful things and living fulfilling lives. And instead of talking to each other about our problems, we get to each talk to a sycophant machine “friend.”

---

I don’t believe in conspiracies, except for the ones that we’ve seen proven true (and those are usually proven true quickly enough). I don’t actually believe the rich are collectively plotting to just to kill us all after we’ve been rendered obsolete.

But I suspect none of them would mind if that’s what happens.

---

This whole thing is very frustrating to me in part because I’m not anti-technology. I think eyeglasses and antibiotics are two of the best things humanity has ever accomplished, and I felt the same burst of pride as everyone else (not national pride, but pride in humanity) when Artemis orbited the moon.

There are “machine learning” tasks that don’t bother me. My problem with Google Maps has far more to do with surveillance than “AI.” I don’t inherently hate a grammar check program (though autocomplete and autocorrect limit human expression in ways that will have a knock-on effect on the development of language, I’m willing to bet).

I would be perfectly happy if I had a computer research assistant in the Star Trek fashion, someone I could ask “how many people openly lived in homosexual couples in Germany during the protestant reformation?” or some such shit, and let a computer comb through all of the human knowledge ever put to paper (well, to bits and bytes) and give me an answer.

But AI doesn’t do that. It functionally cannot. Because it is not thinking. It is a fancy autocomplete, designed specifically to convince us that it is capable of independent reasoning and research. It isn’t designed to comb all of the sources and spit out a correct answer, but to convince us that it has done so.

If I hired a research assistant and asked them a question, and they came back with an incomplete answer, I could accept that. If I hired a research assistant and asked them a question, and they came back with a bizarre mix of truth and things they simply made up whole cloth, I would fire them.

Students who fabricate sources are rightly accused of cheating. Yet when AI does it, we accept it?

The knock-on effects of this are incalculable (though you could ask an AI to calculate them anyway, if you don’t mind false information). As more and more of the information on the internet is AI generated, it becomes harder and harder to sort out fact from fiction. AI is going to start feeding on AI, and soon it will spiral further and further away from coherence.

As Cory Doctorow put it: “AI is the asbestos we are shoveling into the walls of our society and our descendants will be digging it out for generations.”

---

The Luddite movement of the early 19th century wasn’t actually an anti-technological movement per se, but rather a labor movement. Skilled laborers were having their jobs automated away, and textile work moved from the workshop (the “cottage industry,” literally done in a cottage) to the factory. People went from living self-directed lives to lives in which they themselves had to work like cogs in a machine.

So people revolted, at a massive scale, and came close to overthrowing the early capitalist system.

I’m endlessly inspired by the Luddites. But the thing is, those automatic looms? At least those things, devils that they were, still actually worked. Whatever their negative impacts on society, they actually, effectively, automated the production of textiles.

AI doesn’t work. It can’t even become some all-powerful machine god that rules us all, benevolently or malevolently, because it isn’t conscious and it doesn’t think.

AI designers keep telling us they’re going to iron out all the bugs, and that one day AI will accomplish the things it pretends to accomplish today. I doubt that. It seems structurally incapable of what it is promised to do.

What AI is capable of is what it is already doing: replacing skilled labor with just-barely-good-enough automation. AI is capable of tricking people into thinking it’s helping. AI is capable of reducing our capacity for critical thinking. AI is capable of replacing art (which is inherently expression) with a hollow simulacra.

The issues of AI run deeper than just “who will hire illustrators in an era of AI?” though that alone is an important question. The issues run about as deep as deep can go. AI is an existential threat, and not in some “it will become a sentient creature and destroy us all” way but in a “AI is being used for military targeting” way and a “AI is destroying education” way and a “AI is a bubble that will burst and bring down the world economy” way.

But there’s hope.

And that hope looks like college students booing speakers who try to sell them on AI, and it looks like the massive and successful movement against data centers that gets proposed site after proposed site shut down.

And that hope might soon look like a militant neo-luddite movement. And based on the yard signs I see in rural America, that movement might be, or become, bipartisan. Or rather, it might be a class movement, of the working class undivided by culture war issues, fighting against the ruling class.

If so, I’m here for it.