Copyright, trademarks, and AI: Machine learning, Generative AI, LLMs, etc. - Is scraping fair use? Is image gen transformative?

[NoReturn]

In light of the recent lawsuit where a number of newspapers are claiming that AI datasets are a violation of copyright law (see below) and the ongoing screeching from the art and music spheres about usage without permission, I wanted to know your thoughts on the topic of copyright and generative AI, and what arguments you expect to see being presented in the future.

---

Quick reminders from the last 20 years of this discussion:

The Four Factors of Fair Use

1. The purpose and character of your use
2. The nature of the copyrighted work
3. The amount and substantiality of the portion taken
4. The effect of the use upon the potential market.

+ The "Fifth Factor" of "Are you popular?"/"Do people like/support what you made?" (e.g. Original Appalachian Artworks, Inc. v. Topps Chewing Gum, Inc., 642 F.Supp. 1031 (N.D. Ga. 1986).)

Other Possible Copyright Infringement Defenses

• De Minimis defense: Using too little to "count" in the new product (e.g. Images in one of those collages that makes a different image).
• Innocent infringer defense: The accused infringer had no reason to believe that their acts constituted an infringement of copyright.
• "It's not copying" defense: A copyright infringement accusation has to prove the two works are substantially similar from the viewpoint of the average observer. If they aren't, then it's not a copy.

 

[Stephanie Bustcakes]

I don't know even see how this applies to Generative AI. What is it exactly that people are accusing it of doing that violates copyright and trademarks?

 

[Colon capital V]

Too Long; Didn't Read, Total Copyright Death.
>Oh you didn't like me reposting your meme with your name watermarked on it? Sue me then faggot.

 

[NoReturn]

I don't know even see how this applies to Generative AI. What is it exactly that people are accusing it of doing that violates copyright and trademarks?

Sorry. For some context, I was inspired to make this because we're seeing more things like this:

This private information is unavailable to guests due to policies enforced by third-parties.

I'll edit the OP.

 

[gagabobo1997]

I don't think this debate will even matter in the long run. If it turns out that you cant use copyrighted material during training, AI companies will move training to other countries that allow it.
Open source will be impossible to stop too, for obvious reasons.
And eventually synthetic/public domain/licensed data will be more than enough for good models.

Anyway I agree: total copyright death NOW

 

[Stephanie Bustcakes]

If the lawsuit is over gaining access to content that that Microsoft and OpenAI obtained illegally, then I can see this being a legitimate lawsuit. However, if it's simply "We didn't give them permission to train their AIs on our data" then tough shit. There isn't any law about this that I know of. It would be like if a young artist learned how to make abstract art by studying Picasso's art, then Picasso suing that young artist because the young artist shows some stylistic influence from Picasso's art, even if the resulting art isn't copies but wholly original in content.

 

[GunCar Gary]

I don't know even see how this applies to Generative AI. What is it exactly that people are accusing it of doing that violates copyright and trademarks?

from what i know about trademark law, i think there's an argument that AI is frequently being used to violate trademark. go onto any AI model sharing site and there are tons of LoRAs that replicate popular TV and movie characters, naming the character and the franchise they come from directly. some of the LoRA creators are even linking to their kofi, creating a direct link between copying another artist's protected work and receiving money for it

however, i don't think this is really all that different from artists, AI or not, creating fan art of characters and being paid money via patreon for doing so. most of the problems artists have with AI are like this where the problems apply equally to AI models and human, and for some reason it's bad when a machine steals IP but good when scum-sucking bottom feeder twitter artists draw that IP getting gang banged and then slap their patreon link on top

copyright-wise i don't believe it's possible to copyright an art style or a writing style, so any complaining now is by people wanting to retroactively get money for something they didn't know they could gatekeep and charge for

 

[Soyjak Enjoyer]

I notice a very large overlap between people who claim to not believe in intellectual property and people who claim AI art is bad because it infringes upon intellectual property

 

[Absurdist Laughter]

I wanted to know your thoughts on the topic of copyright and generative AI, and what arguments you expect to see being presented in the future.

I sure hope the news papers lose this case. It would mean that language itself can be bought. Imagine having to pay a news agency 10 dollars in royalties to say good morning. What really needs to happen is our education system needs some updating quick. The ignorance surrounding how AI works is what causes a lot of the fear and the fear response to it. These models break things we perceive such as images or words down to a point of annihilation with a vague understanding that if you take some elements of this and that it'll create what was requested. It's why "prompt engineers" are looked for. When you prompt a generative AI, your words are annihilated as well into vague ideas that the algorithm matches to it as best as it can. My overly flowery description aside, regardless text is text and you can't copyright it. You can copyright an idea though, but an idea isn't a bit of code that can type faster than me. (I'm sure you can, but it is just one more reason to hate copyright and patent laws)