Once again, we regret to inform you, it is AI

This week in death to AI some judge in some case ruled that AI companies using our books without permission qualifies as fair use because it's transformative, and not distinct from authors being influenced by other books themselves.

The judge did also say that AI companies pirating books to train their AI was illegal, so I guess hooray for that, but I'm here today to take issue with the claim that feeding books into an LLM to train it is the same as my reading books and then also writing books.

An LLM is, by design, imitative. It is imitating everything at once which is why in general the output sounds bland and uninteresting, but it can also imitate specific things if you ask it to. We all saw this with the harrowing Ghibli nonsense a couple of months ago.

It works by calculating the probability of each point of data it contains. If you ask it to write an essay on Shakespeare in the style of Joan Didion it will run calculations to determine what words you might want in that essay. The kinds of words in all the essays on Shakespeare it's eaten balanced against the kinds of words in all the Joan Didion its eaten get fed through the algorithm like scrap meat into a sausage grinder and spat out the other side in a clean intestine skin.

This is not how real artists influence each other. Real artists take influence from other people's work in several different ways. I will describe those ways now:

Firstly, of course every artist's work does grow out of all the other work they've experienced. The way I write is the product of every book I've ever read, every film I've seen, every piece of music I've listened to. In part. That sort of passive, subconscious building out of something I've forgotten – that's the only way in which real artists can be compared to AI.

But all that influence of all that enormous pile of work does not compare to the passive, subconscious influence of being a living human person who's experienced being alive in this world.

Art – all art – is essentially an eternal conversation about what it means to be human. You can't make it without being human. The influence of life itself far outweighs the influence of other works of art.

And that's just passive influence.

Because we also might take direct influence from other artists. I have done this several times in this very newsletter! And when we do that we think about it a lot. We think about what it is, in a particular work, that we want to allude to.

What is the purpose of the reference? Is it to draw a connection with the reader? If, for example, I reference going down into the pit of despair, that will spark memories of watching or reading The Princess Bride, giving the reader a shared experience with my characters and with me as a writer.

Or perhaps I want to do to a reader what listening to Sondheim does to me. How his lyrics jump around my brain, bubble up beneath my skin. So I'll look at his work and try to figure out how he does that, what grammatical and musical tricks he pulls to create those feelings. And then I'll think about what I can do with prose that might have a similar effect.

All this is very intentional and considered.

An LLM can't do that because an LLM can't think. It has no intentions, nothing it does is deliberate.

Of course we also might take negative influence from a work of art. We might read something and think I want to do exactly not that and then actively try to avoid any kind of allusion or reference.

An LLM can't do that either. It can't distinguish between different works, it can't tell what's true and what's false, it can't recognise satire, it can't be selective in any way. As Film Crit Hulk put it recently it can't read.

The fact that Meta and Open AI and Anthropic used piracy sites to get their hands on our books for training their LLMs is flatly illegal. Using them without our consent and without paying us licensing fees should be illegal too and the only reason it isn't is that copyright law is generally established in the courts, and we're only in the early stages of litigation.

I've said it before and I'll say it again: AI is loser shit. Made by and for losers who have no interest in actually making things of value.

Death to AI.