Hacker News Reader

Comment by stale2002 | Hacker News Reader

stale2002

> If you go and study a work of fiction in a college class, unless that material is in the public domain, you're gonna have to pay for your copy,

No you wont!

It is only someone who distributes copies who can get in trouble.

If instead of that you as an individual decide to study a piece of art or fiction, and you do no distribute copies of it to anyone, this is completely legal and you don't have to pay anyone for it.

In addition to that, fair use protections apply regardless of what the creative works creator wants.

View on HN

Making a profit off variations of someone's work isn't covered under fair use.

Gotcha.

I wasn't talking about someone creating and selling copies of someone else's work, fortunately.

So my point stands and your completely is in agreement with me that people are allowed to learn from other people's works. If someone wants to learn from someone else's work, that is completely legal no matter the licensing terms.

Instead, it is only distributing copies that is not allowed.

AI isn't a human. It isn't "learning"; instead, it's encoding data so that it may be reproduced in combination with other things it has encoded.

If I paint a painting in the style of Monet, then I would give that person attribution by stating that. Monet may have never painted my artwork, but it's still based on that person's work. If I paint anything, I can usually point to everything that inspired me to do so. AI can't do that (yet) and thus has no idea what it is doing. It is a printer that prints random parts of people's works with no attribution. And finally, it is distributing them to it's owner's customers.

I actually hope that true AI comes to fruition at some point; when that happens I would be arguing the exact opposite. We don't have that yet, so this is just literally printing variations of other people's work. Don't believe me, try running an AI without training it on other people's work!

Every waking second humans are training on what they see in their surroundings, including any copyrighted works in sight. Want to compare untrained AI fairly? Compare their artistic abilities with a newborn.

No. That is NOT what humans do unless you somehow learn grammar without going to school. Most of a human's childhood is spent learning from their parents so that they can move about and communicate at least a little effectively. Then, they go to school and learn rules, social, grammar, math, and so forth. There's some learning via copyrighted works (such as textbooks, entertainment, etc.), but literally, none of this is strictly required to teach a human.

Generative AI, however, can ONLY learn via the theft of copyrighted works. Whether this theft is covered under fair use is left to be seen.

Clearly going to school did not help you learn the meaning of theft. If you keep repeating the same incorrect point there is no point to a discussion.

First: in your opinion, which specific type of law or right is being broken or violated by generative AI? Copyright? Trademark? Can we at least agree it does not meet the definition of theft?

I was taught as a kid that using something that doesn't belong to me, without their permission is theft... and it appears courts would agree with that.

> which specific type of law or right is being broken or violated by generative AI?

Namely, copyright. Here's some quick points:

- Generative AI cannot exist without copyrighted works. It cannot be "taught" any other way, unlike a human.

- Any copyrighted works fed to it change its database ("weights" in technical speech).

- It then transforms these copyrighted works into new works that the "original author would have never considered without attribution" (not a legal defense)

I liken Generative AI to a mosaic of copyrighted works in which a new image is shown through the composition, as the originals can be extracted through close observation (prompting) but are otherwise indistinguishable from the whole.

Mosaics of copyrighted works are not fair use, so why would AI be any different? I'd be interested if you could point to a closer physical approximation, but I haven't found one yet.

> Generative AI, however, can ONLY learn via the theft of copyrighted works.

That's not true at all. Any works in the public domain are not copyrighted, and there are things that are not copyrightable, like lists of facts and recipes.

Generative AI could be trained exclusively on such works (though obviously it would be missing a lot of context, so probably wouldn't be as desirable as something trained on everything).

That's not a fair statement to make. It can influence a judge's decision on whether something is fair use, but it can still be fair use even if you profit from it.

The doctrine of fair use presupposes that the defendant acted in good faith.

- Harper & Row, 105 S. Ct. at 2232

- Marcus, 695 F.2d 1171 at 1175

- Radji v. Khakbaz, 607 F. Supp. 1296, 1300 (D.D. C.1985)

- Roy Export Co. Establishment of Vaduz, Liechtenstein, Black, Inc. v. Columbia Broadcastinig System, Inc., 503 F. Supp. 1137 (S.D.N.Y.1980), aff'd, 672 F.2d 1095 (2d Cir.), cert. denied, 459 U.S. 826, 103 S. Ct. 60, 74 L. Ed. 2d 63 (1982)

Copying and distributing someone else's work, especially without attributing the original, to make money without their permission is almost guaranteed to fall afoul of fair use.

Reply on news.ycombinator.com