I'm going to start by saying that I hate copyright, patents, and the way US courts work. I know there's a few of you sperglords out there who will see me argue in one direction one time and immediately assume I'm your enemy. No, I don't love copyright because I think AI training isn't fair use because it competes with the market. I also don't love AI because I think training might be transformative.
This sounds like some gay retard shit. There's nothing illegal about scraping JewTube.
If they only scraped Youtube (like I do every day) the lawsuit would have absolutely nothing. They're arguing that Nvidia used the work commercially without a license and now have to make a defense that it falls under fair use. There's no real precedent for any of this. It's still unclear whether training an AI even counts as using all of someone's work or none of it, if training an AI is fully transformative or not transformative at all, and in this specific case if these Nvidia are models used for research or commercial purposes. I personally don't think AI hurts the original market much because of how bad it is at what it does, but everyone else seems to and you could make a case to convince a judge that AI is explicitly being sold to replace the original sources regardless of it being AI coders, AI artists, AI youtubers, whatever. They're also trying to prove financial harm of these models downloading videos in the millions because of lost ad revenue. I personally think advertisors should just be shot.
As a thought exercise. Lets say Im a lazy arts teacher and its my job to teach some pottery technique to a group of students. I do this by showing my students a pottery YouTube video, not the entire video, just a segment. Have I as a teacher now infringed on the pottery YouTubes copyright?
If I hired human animator, voice actors, and scriptwriters and showed them YouTube videos of the kind of vibe I'm going for then it wouldn't even occur to anyone that I was somehow violating copyright. I don't see how replacing them with a computer changes that, if the content is significantly transformed then it is not violating copyright.
There's a view that massively increasing the volume of something changes the nature of it. Using a computer instead of a human can scale something up a million or billion times. As an example, if a random person just so happens to get you in the background of a photo, no reasonable person has a problem with that. If a company installs cameras at every street corner that just so happen to catch every path you take, that's mass surveillance. It doesn't really work like mathematical induction, where because you can do x, and x+1 is also fine, and infinite scaling is also perfectly fine. The difference between "license plate scanners" and mass surveillance really is just numbers. Fair use also considers the amount of a work used, and it's definitely arguable whether AI models are using very little of the original work, or all of it in the training process and they should require a license to the work. Showing segments of a video may be different than downloading all the videos and using some algorithm on them to produce something you're selling as a replacement to those videos.
There is no way to retrieve the original content upon which the AI generated content was trained, and it can no way be considered a market replacement because it does not contain any of the original content whatsoever.
I don't think AI can replace anything because it's still absolute dogshit, but the people selling it absolutely want you to believe it's a market replacement. That's the only reason stocks involving AI are going crazy right now. Saying there's "no way" it could be considered a market replacement is silly when that's an explicit stated goal of the companies selling AI. I also have no idea if this "using content" applies to just the training, distributing models, or just the output of the AI.
You don't need a license to use someone's work unless you are explicitly using it in your own work.
That's also not necessarily true in the US. There are also statutory damages in the US for violating licenses. Haven't you seen the scary messages that pop up on dvds warning you about getting a million dollars in fines for copying a video? You're not using it in your own work, but merely pirating a movie can get you huge fines. AI also isn't any of the categories that typically see more protection under fair use. Most models are explicitly for profit, and
advertised to replace the market it's training off of. That's why the lawsuit is trying to establish that these models are used for commercial purposes and not exclusively research.
The training is not transformative.
It is unlicensed use.
It's
probably both transformative and unlicensed use. The real issue is that US courts haven't actually decided on this issue so it's just ideas that are floating around. People are looking at pre-AI precedent and arguing over how it applies to AI without an actual rulings. When people say that AI is definitively transformative, or definitively not transformative at all, they have to be crazy because that hasn't been determined yet.
In another case where quantity actually matters, under US copyright rulings, it also considers how much of the material is used. Just because you can use 1 page doesn't mean you can use another page, or 100 pages. It's taken on a case by case basis and every time you defend, you're making an affirmative defense: Yes, I did use copyrighted materials without a license, but I'm allowed to because of [some bullshit that's going to cost a lot of money to prepare].
AI uses the complete material before (possibly) transforming it, and it's not unreasonable to think a judge would consider using
all of someone's videos, or perhaps all videos on a platform, to be less narrow than using whatever videos some artists can personally watch.
They're also going for statutory damages in the lawsuit where they don't EVEN have to prove damages to get their money. They're using the nintendo strategy: $50,000 fine per offense, you pirated a million videos, you now owe us the entire economy in damages. No one ever gets awarded ten billion trillion dollars (unless you're suing infowars), but it's never a bad place to start.
I still think copyright and patents should be set on fire, but anyone speaking with certainty on how old laws handle new technology is being a silly billie. Whether AI is transformative in any upcoming case could be decided by whether some geriatric ward patient had breakfast. Just like how in the real world laws don't exist if they're not enforced, nobody actually knows what the burgerland laws actually mean they've been ruled on by a judge.
Here's a quick example: Congress wrote the civil rights act that prohibits "discriminating based on age ... in all terms of employment". It doesn't matter what the text actually says though, because in cases like "General Dynamics Land Systems Inc. v. Cline" the supreme court reinvented "age" to mean the state of being aged and completely invented this legal precedent that discrimination based on age is perfectly fine as long as it's only to younger people. You can imagine how this sort of logic regarding "reverse discrimination" can extend to any "protected class" under the civil rights act. This is also why I hate judges in general, since they have the most power when times are easy and can easily siphon power off clearly written laws or too easily block the executive branch from exercising power they rightfully have.
And remember, you can still be completely right, win every lawsuit, and still go bankrupt like Bleem! if you're lucky. I think we all know this isn't going to happen to Nvidia either way.
