This doesn’t stop anyone from viewing or scraping the work, though, so in no way is it DRM. It just causes certain methods of computer interpretation of an image to interpret it in an odd way vs. human viewers. They can still learn from them.
>Nightshade's goal is not to break models, but to increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative.
Which feels similar to DRM. To discourage extraction of assets.
Sure. Just like how video game drm impacts performance and watermarks on images degrades the image. Drm walks a tight line that inevitably makes the result worse than a drum-free solution, but also should not make the item completely unconsumable.
So, do you want to define drm by intent or technical implementation? I'm doing the former, but it sounds like you want to do the latter. Also keep in mind that legalese doesn't necessarily care about the exact encryption technique to be deployed either.
Both. Changing an image is done all the time prior to publishing them. In fact, no image you ever see on the internet is a raw sensor output. They are all modified in some manner. The images processed using this method look the same to every person and computer that views them. That’s very different from DRM which encrypts things and prevents access to unprivileged users.
This is effectively the equivalent of someone doing really crappy image processing. As other commenters have mentioned, it does alter how images look to humans as well as machines, and it can be “mitigated” through additional processing techniques.
>That’s very different from DRM which encrypts things and prevents access to unprivileged users.
Well you can call it a captcha if you want. The point here is to make it harder to access for bots (but not impossible) while inconveniencing honest actors in the process. It doesn't sound like there's a straightforward answer to "are captchas DRM" either.
That's true of almost all DRM, isn't it? Even for the most annoying form that is always-online DRM, everyone is provided the same access to the bytes that form a game. You and I have the same bytes of game.
It's the purpose of some of those bytes that turns it into DRM.
No, it's not the same. The game is non-functional without the proper keys/authorization whereas images run through this algorithm are still images that anyone and any computer can view in the same manner without any authentication.
An analogy that springs to mind is the difference between an access control mechanism such as a door with lock and key versus whatever magical contrivance that prevents entry to a dwelling by vampires uninvited.
That may be an ok analogy, but magic isn’t real. Maybe a better analogy would be to speak or write in a language that you know certain people don’t natively understand, and using lots of slang and idioms that don’t translate easily. Someone could still run it through Google Translate or whatever, but they won’t get a great understanding of what you actually said. They’d have to actually learn the language and the sorts of slang and idioms used.
I agree that the analogy is strained. My goal was in elucidating the distinction between:
the goals of artists and the developers of OP,
versus
the goals of AI engineers,
and how it seems to me similar to the is/ought disctinction.
In my original analogy, it’s generally considered lawful to have a lock on your door, or not to do so, and the issue of a lock or lack thereof is moot when one is invited to enter, just as it is lawful to enter a premises when invited or during exigent circumstances, such as breaching entry to render lifesaving aid by emergency services or firefighters.
By that same token, no amount of locks or other barriers to entry will prevent ingress by a vampire once invited inside.
To me, much of the ballyhoo about OP seems like much ado about big cats eating faces, like a person publicly decrying the rise in vampire attacks after inviting that same vampire inside for dinner. It’s a nonstarter.
Copyright law is broken, because of the way that the law is written as much as the way that it’s enforced, and also broken because of the way that humans are. Ideas are not copyrightable, and while historically their implementations or representations were, going forward, neither implementations nor representations are likely to receive meaningful/effective protections from copyright itself, but only from legal enforcement of copyright law.
After the recent expiry of Disney’s copyright on Steamboat Willie, the outpouring of praise, support, excitement, and original work from creators shows me that copyright law in its current incarnation doesn’t perform its stated goals of promoting the creation of arts and sciences, and so should be changed, and in the meantime ignored and actively disobeyed, as any unjust law ought to be, regardless of what the law is or does.
In light of our obligation to disobey unjust laws, I applaud efforts like OP to advance the state of the art in computer science, while at the same time encouraging others working on AI to actively circumvent such efforts for the selfsame reason.
I similarly encourage artists of all kinds to make art for art’s sake while monetizing however they see fit, without appealing to red herrings like the legality or lack thereof of end users appreciating their art and incorporating it into their own artistic output, however they may choose to do so.
Like all art, code and its outputs is also First Amendment protected free speech.
