How artists are sabotaging AI to take revenge on image generators

https://theconversation.com/data-poisoning-how-artists-are-sabotaging-ai-to-take-revenge-on-image-generators-219335

by taimurkazmi • 3 years ago

49 102 3 years ago

there is going to be a "pre-GPT" internet training set from 2022 and earlier that will basically be the last 'complete' not poisoned data set.

As AI increases as an overall % of all online posts and activity it will death spiral on model quality.

pupileater• 3 years ago

Haha, like https://en.wikipedia.org/wiki/Low-background_steel

tomrod• 3 years ago

Not too far off!

AI retraining on its own output is not a great thing.

sjfjsjdjwvwvc7• 3 years ago

I disagree. There is still plenty of genuine data being produced. Also in the case of images, most of them have digital watermarks.

I see no reason to believe there will be a death spiral on model quality - in the last year or so the opposite has been happening.

Of course the jury is still out on the future.

DaiPlusPlus• 3 years ago

Also in the case of images, most of them have digital watermarks.

Do they now?

amelius• 3 years ago

Only if we're too lazy to pre-filter the content manually.

dragonwriter7• 3 years ago

there is going to be a "pre-GPT" internet training set from 2022

Well, yeah, there are several here, and I think all the major image generators are using some combination of them as their starting points: https://laion.ai/

As AI increases as an overall % of all online posts and activity it will death spiral on model quality.

Nope, it will just mean that it will be more expensive to source additional training data on top of the massive trove of existing "clean" (from intentional poisoning) data (much of which isn't perfectly captioned and human work on improving captioning can improve its utility in model training, as can more advanced models with more advanced text encoders, etc.)

If poisoning was widespread, it wouldn't impact "big model" quality much -- they aren't grabbing new random data on the internet for continuous training. It might drive up the expense of community fine tuning, which often does depend on sourcing representative imagery for target styles or concepts from, among other places, the internet.

mercurialsolo• 3 years ago

This is just a race to the bottom and creating artificial work - folks creating poisoned data sets and then armies of data engineers who look at creating valid data sets for building models that can be grounded.

sjfjsjdjwvwvc7• 3 years ago

I feel for artists to an extent but no way are they going to succeed „ruining AI“ by poisoning image data sets. And even if that was feasible, what the hell do they gain?

This is the whole „photography is not art“ or „digital image manipulation is not art“ or „oil painting is not art“ discussion all over again and I can believe serious artists actually take pet in this luddite theatre.

chii4• 3 years ago

And even if that was feasible, what the hell do they gain?

a few more years of employment, at best. It sucks, but it's reality. There's no stopping the onslaught of AI.

Better start learning how to take advantage of the AI tools to make yourself not irrelevant.

camillomiller3• 3 years ago

The lack of humanism in comments like this one are worrying. The fact that you don’t understand art or the artistic process doesn’t mean they are not relevant to a healthy society.

dragonwriter7• 3 years ago

a few more years of employment

Many less years than if they learned the new tools that magnify productivity.

somewhereoutth• 3 years ago

Photography/manipulation/oils are/were done by humans with intent.

Putting some prompts into an image generator and choosing which one you like is hardly the same thing.

However, the idea of AI generated images could be considered art (a bit like Duchamp's toilet), but it is pretty much done now.

Of course, 'paintings for hotel rooms' is something different, and indeed may be affected by AI

wholinator2• 3 years ago

Well, i think it's strange to assume a serious artist would just take the first crop given to them and not alter at all! After all, there are numerous ai tools for altering ai images to fit your choices better. You simply pick the best template and then direct the AI, with intent, at exactly which parts of the image you'd like changed and into what. It's a delicate balance of prompt engineering (knowing exactly what to say, sometimes terms not know by the lay public become necessary for the best creation of your idea), and actually having an artistic idea in mind. It takes inspiration, tact, composition, knowledge, etc.

Sure, putting in "throwing up technicolor on the burger king riding a giant snail" is incredibly easy. Being able to fix everything the AI cannot interpret about your ideal, let alone having an ideal strong enough to differ from what your presented is artistic. I'm well aware there are people vehemently opposed to this idea, but there should be an advocate for the opposition

Philpax2• 3 years ago

Putting some prompts into an image generator and choosing which one you like is hardly the same thing.

At a base level, "putting some prompts [..] and choosing" sounds like intent to me. A user is choosing what they want to capture and making a choice, much like I, as a photographer, look for a particular scene (or happen across one!) and choose how I want to capture it.

Of course, you can do much more in terms of control of image synthesis these days. ControlNet and friends make it feasible to control every aspect of the resulting image.

baobabKoodaa• 3 years ago

There's a lot more to AI art generation than "putting some prompts into an image generator". Sure, putting some prompts into a generator _also_ counts as "art" in the same manner as "child scribbles doodles on paper" counts as art. But you're dismissing an entire field of aspiring artists without having even a basic understanding of what AI artists do to achieve the images they publish. Go watch some videos of decent AI art workflows and then come back.

sjfjsjdjwvwvc7• 3 years ago

If you believe you can’t use Stable Diffusion for example with intent, I don’t believe you ever seriously used it. There is a ton of work going into it, from training your own loras, building workflows that achieve what you are looking for, using img2img and literally dozens of other related tools and methods.

With the same argument you can say photography doesn’t have intent - you just point the camera somewhere and press the button.

Edit: for what it’s worth, Duchamp just stole the toilet from another artist and passed it off as his own work - I would argue that is much worse than using something like SD.

kazinator5• 3 years ago

There is a whole lot of art that is not "high art". Just commerical art. Graphic design. Not all of it is "paintings in hotel rooms"; it's all over the place. People have jobs in this area.

beepbooptheory2• 3 years ago

Serious artists are going to be around long after the last GPU gets clogged with sand.

plastic_enjoyer• 3 years ago

I can believe serious artists actually take pet in this luddite theatre.

I think more people should participate in the luddite theatre. Luddites weren't against automation and technological advances per se but that the profits of increased productivity weren't shared. The people that are profiting off AI generated art today don't want to participate in distributing their profits more widely, despite generative models being based on the works of other people, some of whose art was used for the training without permission.

namlem• 3 years ago

Is anyone reaping great profits off AI art? After all, it's not even copyrightable.

keybored2• 3 years ago

I’m tired of the “they’re Luddites” middlebrow dismissal, ignoring all of the history and lessons of that time.

Yeah, Industrialization was a completely painless and smooth process. No one was ever inconvenienced. That’s why we can make fun of the then-naysayers now.

indigo00862• 3 years ago

Hardly comparable. The fact that people can generate work doesn't mean they're owed labor because their work is being automated away. The physics of work is not a creative endeavor and even though I think ownership of creative works is grey matter the idea of having to share profits after generating more work with machines is just blindingly illogical

glimshe• 3 years ago

But why do they have to participate in the profits? How musicians participate the profits of digital audio workstations that allow people to make music without real musicians? How did older artists profit from current artists that learned, and oftentimes virtually copied their style? The solution for artists is to embrace these tools to make even better art.

polski-g• 3 years ago

Don't "want to"? Nobody wants to. If you want to make them, then lobby your representatives to pass a tax law targeting AI-derived revenue.

michaelcampbell• 3 years ago

I remember when Waze and Google Maps came out and angry neighborhood residents would try to "poison" the traffic data set by reporting non-existant police in their neighborhood streets.

That also didn't work.

indigo00862• 3 years ago

Yeah, this is really just artists poisoning themselves hoping AI dies. It'll take away from them using their creativity to actually capitalize and they'll just spend their time trying to go against relatively intelligent people on their own turf.

taylorius3• 3 years ago

Why couldn't highlighting the fragility of such machine vision systems be an artistic endeavour?

RugnirViking• 3 years ago

it would potentially be, if it had any hope of working.

davexunit• 3 years ago

AI art is not art. AI art is copyright laundering.

SirMaster• 3 years ago

And even if that was feasible, what the hell do they gain?

More visibility by the world through news articles exactly like this?

More visibility perhaps clues more average joes of the world into what exactly is going on. Maybe it gains them more people who might be on their side and have sympathy for them, etc.

Even if they don't have a clear end game, doing something, anything they can probably feels better to them than doing nothing, and just going quietly into the night without putting up any resistance.

spinach• 3 years ago

This is not the same as those things at all. The poisoning of the data sets only works because the AI is scraping the internet for images and stealing the artist's works. They are mad because their art is being stolen and they didn't consent to it.

It's easy to call other people names when it's not your life's work at stake.

kossTKR• 3 years ago

A nanoscopic amount of artists sabotaging a few images will not do anything to anyone, there'll just be a small version bump in vision models and the 0.0000001 contaminated images will also be classified correctly.

While i really feel for the artist like everyone else potentially losing their livelihood to AI, this is an annoying fantasy story.

Exendroinient23• 3 years ago

Just wait for luddite political candidates becoming popular when white collar workers will start be inevitable replaced. It will be a total disaster for any sensible discourse.

hospitalJail5• 3 years ago

I think the US already went through their populist cycle between 2008-2020. If that cycle theory is true, we shouldn't have to worry about it until our old days.

ghusto2• 3 years ago

I dislike the unquestioned assumption that "Luddite bad, machine good".

If you love progress for progress' sake, think about where you're progressing _to_. So far we've "progressed":

- My furniture no longer being made by skilled carpenters: Everything is falls apart after a couple of years, and great, I have to put it together myself to begin with

- Lowering of the bar to _literally anyone with a computer_ for music creation: I know all old people say this, so I won't break tradition: Your music is shite

- Hi-Fi not being particularly hi-fi: A Bluetooth receiver playing lossy format crap out of terrible speakers

- Cashiers to automatic checkouts: Cool, now I have to do their job without getting paid, then wait for the one staff member they kept on payroll to come and check my shopping and rescan everything because of course they don't trust me

I could go on all night. Point is; progress just means forward motion, not always in a good direction.

deafpolygon• 3 years ago

Weird to phrase it as if the artists are in the wrong.

masklinn• 3 years ago

From the boot’s point of view, the skull is in the wrong.

hospitalJail5• 3 years ago

Who is right?

The people who are happy to have higher quality artwork at lower costs, allowing humanity to progress further?

Or the establishment who want their own desires fulfilled at the expense of the rest of humanity?

sebastianconcpt2• 3 years ago

Can you explain how extinguishing the artists or, its practical equivalent, destroying the economic incentives for them to exist is being in the right?

bertil• 3 years ago

Poison works by being able to scale its impact dramatically: attacking chemical cornerstone processes, disabling a receptor, etc.

ML is, to simplify, a statistical process that tries to identify probability distributions. That implies they’d have to purposefully degrade large amounts of images that the system has access to. Unless those images are impossible to distinguish from the others (because, say, they come from the same sites) I’m not sure how they’d do it consistently.

They could mark sources, by using an unused combination of keywords and having a unique image attached to it, but poisoning—at scale? I’m not seeing a stable equilibrium where that’s possible.

bejd• 3 years ago

Previous discussions:

https://news.ycombinator.com/item?id=38013151

https://news.ycombinator.com/item?id=37990750

beepbooptheory2• 3 years ago

I agree with people this is a long shot, but how can you not admire the spunk? Whether we realize it or not yet, this is the kind of resolve and energy we need to bring to the future.

Antagonism is not automatically luddite-ism! This is capitalism right? Dont we need the competition?

gcanyon• 3 years ago

We are on a roller coast ride of technological advancement. Gravity is pulling us inexorably forward, and we can't see where the track eventually goes. We have brakes, but not enough to stop forward movement entirely. We have some input on the direction we take, but our limited vision down the track makes it hard for us to anticipate and correct for any upcoming curves we don't like.

It's going to be a bumpy ride, but hopefully we have fun along the way.

taylodl2• 3 years ago

I think you misunderstand the violence that lies in our near future. Before, when we began automating manual labor we proclaimed you should get educated and design the machines and improve the processes. Now, we're developing tools to displace large swaths of those workers. There simply aren't other jobs for them to do that we haven't already automated. History shows this is not sustainable and will lead to protest and violence. I would suggest we avoid that, after all, that would be the smart thing to do, but our species has proved time and time again that it's not smart in the aggregate.

GaggiX2• 3 years ago

It's never going to matter, ignoring that they are a extremely small percentage of images on internet, if they have enough signal to almost not be noticed by a human than there is more than enough signal to train an AI.

mysterypie2• 3 years ago

altering an image’s pixels in a way that wreaks havoc to computer vision but leaves the image unaltered to a human’s eyes

Can someone who understands the original paper give an ELI5 on how that's possible?

I understand how labeling an image of a cat as "dog" could poison a dataset.

I also understand how adding images of a toaster (which sort of looks like a handbag) to a dataset of handbags could poison a dataset.

But I don't understand what's happening in Figure 6 in the original paper. The pairs of pictures of the dogs, cars, etc. look absolutely identical. What exactly is happening there?

taylorius3• 3 years ago

Computer vision systems are not as robust as human vision systems. There are subtle changes to pixel values that can utterly flummox a computer vision system, but seem undetectable to a human looking at the image.

gooob• 3 years ago

what is the nature of these pixel changes?

mysterypie2• 3 years ago

How do subtle pixel changes cause the image of a dog to classified as a cat rather than as gibberish?

hackernewds• 3 years ago

I've read about this

https://www.nytimes.com/2023/02/13/technology/ai-art-generat...

ensocode• 3 years ago

Artists are happy to use the global working plattforms (Instagram, ...) to show and monetize their art to a world-wide audience but are unhappy if these content is used to train AI. If you show your art in just your own local gallery nobody will exploit it. Its the same in software engineering with public repos. So I guess we all have to adjust and use AI for the better instead of complaining and fighting it. Or am I missing the point here?

dig1• 3 years ago

This reminds me of an attempt when spammers tried (and still trying, but not as often as before) to poison antispam bayes filters with all sorts of garbage in email. The result is that filters became even more powerful over time.

sonicggg• 3 years ago

If they are fighting back, they are failing miserably, because the AI generators look better than ever. Midjourney, for example, is crazy good.

hospitalJail5• 3 years ago

I talked to a luddite on this:

"What if I take my own pictures, and train the data on my own pictures?"

Their answer:

"No"

Luddites, nothing more. They will become obscure while everyone else advances. She uses her colored pencils, I use digital art. Guess who works at a coffee shop.

camillomiller3• 3 years ago

These comments are so disheartening and disgusting. Are you enjoying your revenge of the nerds moment? Go fellate a robot, will ya

ghusto2• 3 years ago

You say Luddite like it's a bad thing.

Guess who works at a coffee shop

Her? Happy and fulfilled as a human being no doubt.

hospitalJail5• 3 years ago

Buddy she is depressed, in debt, and on a ton of pills.

hurrdurr56• 3 years ago

Solution:

1. Run a large number of verified unpoisoned images through the AI poisoning algorithm

2. Create a large high-quality data set consisting of the before and after poisoning image pairs

3. Train an AI using this data set so it can detect or even reverse the poisoning

4. AI train stops for no one

Lockal2• 3 years ago

It probably can be simplified to writing an algorithm which detects that image was generated by AI (like https://deepmind.google/discover/blog/identifying-ai-generat...), and consider all AI-generated images as bad for training set by default.

dragonwriter7• 3 years ago

Gotta give the team behind Nightshade and its earlier, rapidly debunked predecessor (Glaze) credit for a really strong media-relations operation.

But that's about it, and this isn't meaningfully going to impact AI image generation at all.

Lockal2• 3 years ago

It reminds me of a wave of comments "Deadline tomorrow !!! Everything you’ve ever posted becomes public from tomorrow. I give notice to Facebook it is strictly forbidden to disclose, copy, distribute, or take any other action against me. If you do not publish a statement at least once it will be tactically allowing the use of your photos, as well as the information contained in the profile status updates. DO NOT SHARE. Copy and paste."

But now people are posting images instead of text. And it all looks so stupid that it only deals reputational blow to some subgroup of artists who now look like a group of technically illiterate luddites.

How artists are sabotaging AI to take revenge on image generators

Related Stories