The Packbats' Weblog

Packbats' policy on AI content generation, one swear

Any machine built by scraping massive amounts of training data off the Internet without permission can get fucked.

Any machine designed to divert income from artists, writers, musicians, and other creators to companies, likewise.

That is to say: Midjourney, Stable Diffusion, DALL-E, Adobe Firefly, ChatGPT, Bard, and so on - all of them can go straight to hell.

If we share or boost something that uses these tools, please let us know. We will never make anything with any of those.

The thing is, "AI", in its modern deceptive usage, mostly means "neural net". And that makes things a little difficult for us as people who get hinky about the literal meaning of words, because the campaigners we agree with are saying things like "generative AI" and "no AI" and making Human Made logos for humans to put on human art that isn't made out of theft and exploitation ... and yeah, we support that, but that's not what we would say?

Like, we keep pointing to Adam Neely's collaboration with Dadabots. This was a neural net acting as a generative AI creating art ... that is based on two hours of bass playing by a musician who volunteered to have his work used this way, and that creates music that no bass player would ever play at a gig. It's not stealing anyone's creations and it's not stealing anyone's jobs. And it's not stealing credit - for as long as it existed, it had Adam Neely's face plastered on the thumbnail, it was explicitly made with his permission, and had a link to his video about it in the description. And, well, we said what we have to say about a hyperfocus on humanity.

So, like, the AI bubble is toxic but we don't have a pithy way to say it. Except maybe to call them plagiarism machines.

Anyway I don't think we made a post here on Dreamwidth about that shit, so let us say again that these plagiarism machines deserve to be destroyed. And as great as nuance is, it just lends emphasis to that conclusion.

1.

okay so the big problems with the modern wave of generative AI are that:

• it is designed to steal jobs from artists and writers
• it was manufactured using titanic amounts of stolen work from those selfsame artists and writers
• it required and requires titanic amounts of electricity and other computing infrastructure in the middle of climate crisis

...but an additional problem is that, because they are intended to be used without warning, they force people to try and find intention, worldview, meaning, all the things we expect from our fellow writers and artists, in material which contains none of that.

and that's just abusing our assumption of good faith.

2.

like, okay

imagine a friend mentioned something bad that happened to them, and we replied "oh no!"

this is as nearly an automatic and thoughtless response as we can think of ... but consider what it would mean to our friend

at the very least, they can infer that:

• we have been paying attention to their speech
• we recognize, whether we understand the details or not, that something unfortunate has happened
• we care that something unfortunate happened, and would rather it had not

if they had said the same thing to ChatGPT, and ChatGPT had said, "oh no!", would it have anything like the same effect? ChatGPT doesn't know them and won't remember them, it only understands that "oh no!" is a thing that is said frequently in its corpus in this kind of context. it is simply and utterly hollow.

Via elynne: the real story of the autism-vaccine saga, in easy to read and visually attractive comic form.

Hi! I'm going to talk to you about morality, because I'm arrogant and you're imperfect.

No, these facts have no relation. Everyone is imperfect - myself more than you, I wager - and I'd be arrogant even if the lot of you were plaster saints. But the second has interesting consequences which the first permits me to address.

(And as long as I'm blathering, let me make a quick clarification: morality is not law, and law is not morality. If you find yourself interchanging the two, you need to recheck your math. Moving on.)

The thing about morality I want to address today is not the content, but the form. Morality acts on three grammatical persons - the first, the second, and the third - and among most people it tends to be different for all three. (This is why Mormons come to your door - it's harder to be rude to a face than a phone.) This makes sense except for one important factor: a lot of people (though probably fewer than it seems) get the proportions backwards, and need correction. So let me break it down for you.

In the first person - in your morality for you - you ought to be strict but fair. As some wit commonly cited as "Yahl, J." is quoted: "Perfection is our goal, excellence will be tolerated." Stick to the straight and narrow road, get it right the first time, and if you get it wrong, get it right the next time. Practice your morality with all the intensity, precision, and dedication that you were supposed to practice the piano when you were growing up.

In the second person, and still more in the third person - in your morality for your friends and for your strangers - be looser. If your personal code is the double-yellow line, give your friends the entire road and strangers two city blocks in both directions. If your personal code is the Geneva Conventions, let your friends have the Declaration of Independence and allow the rest the Golden Rule. Or, if you prefer: an it harm none, let everyone else do what they will.

Why this? Because you don't really know what's right and wrong, not to any sensible degree of accuracy. Oh, you're better off than the Hittite slave holder who, lacking our hard-won experience, never made the connection between the wretched condition of the slave and the moral repugnance of the institution, but "better off" is a long way from omniscient. And the hard part about morality is that it's chaotic - it depends on a tremendous array of details which you might (if you're lucky) know for your own situation but which you are more ignorant of the farther you look from your center of consciousness. While on the one side you want to do right, on the other you don't want to be - in fact, you shouldn't be - the one who beats people up when they haven't done anything wrong.

So how do you do this? You set an engineering margin of error - draw yourself a circumference small enough that you may be confident it (mostly) resides within the right and aim for that, while drawing for others a loop which (mostly) circumscribes the right and nudge what falls outside back in. In other words, you be the anti-hypocrite: you criticize in yourself what you let pass in others.

And that's the form to take, in the first, second, and third persons. Thank you for your patience.