Live data from Hacker News

Nightshade: An offensive tool for artists against AI art generators

nightshade.cs.uchicago.edu

141–150 of 710 posts

Re: Nightshade: An offensive tool for artists against AI art generators

#141

Earlier quoted context omitted.

What are you saying? Of course it's subjective, e.g. 3 million years ago there were no 'different implications' whatsoever, of any kind, because there were no humans around to have thoughts like that.

I'm using "implication" as a synonym of "effect". If a human learns to imitate your style, that human can make at most a handful of drawings in a single day. The only way for the rate of output to increase is for more humans to learn to imitate it. If an AI learns to imitate your style, the AI can be trivially copied to any number of computers and the maximum output rate is unbounded. Whether this is good or bad is s…

Ah okay, I get your meaning now, I'll edit my original comment too.

Though we already have an established precedent in-between, that of Photoshop allowing artists to be, easily, 10x faster then the best painters previously.

i.e. Right now 'AI' artistry could be considered a turbo-Photoshop.

Re: Nightshade: An offensive tool for artists against AI art generators

#142

What the article doesn't illustrate is that it destroys fine detail in the image, even in the thumbnails of the reference paper: https://arxiv.org/pdf/2310.13828.pdf Also... Maybe I am naive, but it seems rather trivial to work around with a quick prefilter? I don't know if tradition denoising would be enough, but worst case you could run img2img diffusion. reply

The poisoned images aren't intended to be viewed, rather scraped and pass a basic human screen. You wouldn't be able to denoise as you'd have to denoise the entire dataset, the entire point is that these are virtually undetectable from typical training set examples, but they can push prompt frequencies around at will with a small number of poisoned examples.

Re: Nightshade: An offensive tool for artists against AI art generators

#143

To protect an individual's image property rights from image generating AI's -- wouldn't it be simpler for the IETF (or other standards-producing group) to simply create an AI image exclusion standard , similar to "robots.txt" -- which would tell an AI data-gathering web crawler that a given image or set of images -- was off-limits for use as data? https://en.wikipedia.org/wiki/Robots.txt https://www.ietf.org/

Entities training models have no incentive to follow such metadata. If we accept the premise that "more input -> better models" then there's every reason to ignore non-legally-binding metadata requests. Robots.txt survived because the use of it to gatekeep valuable goodies was never widespread. Most sites want to be indexed, most URLs excluded by the robots file are not of interest to the search engine anyway, and us…

>Entities training models have no incentive to follow such metadata. If we accept the premise that "more input -> better models" then there's every reason to ignore non-legally-binding metadata requests.

Name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.

>Robots.txt survived because the use of it to gatekeep valuable goodies was never widespread. Most sites want to be indexed, most URLs excluded by the robots file are not of interest to the search engine anyway, and use of robots to prevent crawling actually interesting pages is marginal.

Robots.txt survived because it was a "digital signpost" a "digital sign" -- sort of like the way you might put a "Private Property -- No Trespassing" sign in your yard.

Most moral/ethical/lawful people -- will obey that sign.

Some might not.

But the some that might not -- probably constitute about a 0.000001% minority of the population, whereas the majority that do -- probably constitute about 99.99999% of the population.

"Robots.txt" is a sign -- much like a road sign is.

People can obey them -- or they can ignore them -- but they can ignore them only at their own peril!

It's a sign which provides a hint for what the right thing to do in a certain set of circumstances -- which is what the Law is; which is what the majority of Laws are.

People can obey them -- or they can choose to ignore them -- but only at their own peril!

Most will choose to obey them. Most will choose to "take the hint", proverbially speaking!

A few might not -- but that doesn't mean the majority won't!

>If there was ever genuine uptake in using robots to gatekeep the really good stuff search engines would've stopped respecting it pretty much immediately - it isn't legally binding after all.

Again, name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.

Re: Nightshade: An offensive tool for artists against AI art generators

#144

Setting aside the efficacy of this tool, I would be very interested in the legal implications of putting designs in your art that could corrupt ML models. For instance, if I set traps in my home which hurt an intruder we are both guilty of crimes (traps are illegal and are never considered self defense, B&E is illegal). Would I be responsible for corrupting the AI operator's data if I intentionally include adversaria…

How would that situation be remotely related?

Re: Nightshade: An offensive tool for artists against AI art generators

#145

[flagged]

And? Even if neural networks learn the same way humans do, this is not an argument against taking measures against one's art being used as training data, since there are different implications if a human learns to paint the same way as another human vs. if an AI learns to paint the same way as a human. If the two were exactly indistinguishable in their effects no one would care about AIs, not even researchers.

And yet, some people don't even want their artwork studied in schools. Even if you argue that an AI is "human enough" the artists should still have the right to refuse their art being studies.

Re: Nightshade: An offensive tool for artists against AI art generators

#146
post #97

With this "solution" it looks like the world of art enters the cat-and-mouse game the ad blockers were playing for the last decade or two.

I might be missing something because I don't know much about the architecture of either Nightshade or AI art generators, but I wonder if you could try to have a GAN-like architecture (an extra model trying to trick the model) for the part of the generator that labels images to build resistance to Nightshade-like filters.

It doesn't even have to be a full GAN, you only need to train the discriminator side to filter out the data. Clean reference images + Nightshade would be the generator side.

Re: Nightshade: An offensive tool for artists against AI art generators

#147
post #117

Earlier quoted context omitted.

Why do you have to prove that? There is no replication (except in very rare cases), how someone draws a line should not be copyrightable.

Therein lies the crux of the issue: AI is not “someone”. We need to approach this without anthropomorphizing the AI.

Companies aren't someone, yet in the US we seem to give them rights of someone.

Re: Nightshade: An offensive tool for artists against AI art generators

#148
post #62

[flagged]

Human beings and LLMs are essentially equivalent, and their processes of "learning" are essentially equivalent, yet human artists are not affected by tools like Nightshade. Odd.

Humans don't fall for optical illusions? News to me.

Re: Nightshade: An offensive tool for artists against AI art generators

#149
post #80
post #73

Earlier quoted context omitted.

Humans and AI are not the same thing, legally or physically. The law does not currently grant AI rights of any kind.

If a human isn't violating the law when doing that thing, then how is the machine violating the law when it cannot even hold copyright itself?

In some locales sitting on the street writing down a list of people coming and going is legal, but leaving a camera pointed at the street isn't. Legislation like that makes a distinction between an action by a person (which has bounds on scalability) and mechanized actions (that do not).

Re: Nightshade: An offensive tool for artists against AI art generators

#150

Earlier quoted context omitted.

And? Even if neural networks learn the same way humans do, this is not an argument against taking measures against one's art being used as training data, since there are different implications if a human learns to paint the same way as another human vs. if an AI learns to paint the same way as a human. If the two were exactly indistinguishable in their effects no one would care about AIs, not even researchers.

And yet, some people don't even want their artwork studied in schools. Even if you argue that an AI is "human enough" the artists should still have the right to refuse their art being studies.

> the artists should still have the right to refuse their art being studies.

Why? That certainly isn't a right spelled out in either patents or copyrights, both of which are supposed to support the development of arts and technology, not hinder it.

If I discover a new mathematical formula, musical scale, or whatnot, should I be able to prevent others from learning about it?

Post reply on HN