#taken from a deviantart ai “creator” (ha!) | Explore Tumblr Posts and Blogs

bananonbinary · 11 months

Note

the main argument I've heard made wrt the IP thing and AI art is that the AI can only ever combine works from other artists. it has no creative input other than combining works that already exist; which, personally, i would be fine with if it was just being trained on work in the public domain and work from people who consented to be included, but these bots are scraping art from literally everywhere on the web with *very* few opportunities to opt out. even a human artist that's heavily inspired by other creators adds something of their own to the mix, it's kind of impossible *not* to unless you're literally tracing someone else's design, but AI is using art taken from other creators And Nothing Else, which when combined with the fact you can ask it to produce art "in the style of X" gets a lot of artists very mad

that feels like a very esoteric complaint though. how does one measure "creative input," and why does the actual method it uses to create not count? if you can't actually point to a bit of AI art like "thats my OC it clearly lifted that from this painting i did" then it's clearly done SOMETHING transformative. what has it actually stolen? the knowledge that your work exists, somewhere?

as far as i understand it, the main practical reason for "intellectual property" to exist as a legal concept, is to prevent someone from harming someone else's livelihood or reputation by claiming the work as their own. parody and free use are allowed only so far as they are sufficiently different from the source work. but these AIs...don't claim credit for the source works. they don't reproduce the source works anywhere at all. and what they create IS pretty transformative, even if it's in a soulless robot kinda way. it can't really be considered to be "taking away" from sales etc of the original works that exist, because it doesn't actually look like any one of them at all. I think even if I, a human, DID create "here's a collage of tiny pieces of shit i found on deviantart" it would still be considered fair use, but the connection between anything in the data sets and the actual art the AIs produce is much, much more tenuous than that.

I think any way that it could hurt an artist that you could define would also just cover competing artists in general. Even if an AI created a picture of, say, a tree In The Style Of Artist (let's call them Phil), it still wouldn't ACTUALLY be the same as the tree Phil would draw, it would just kinda look similar. and anyone who wanted a tree by Phil would know that this wasn't one, and wasn't worth their time. Phil could draw their own tree, and still sell it, and the AI would only really compete with that among Phil fans if it actually tried to claim that its version WAS by Phil.

This is sort of my main hesitation for the IP argument. we here in fandom spaces exist in a grey area of intellectual property, where EVERYTHING is "in the style of" or "based on" existing properties. if we try to legislate those very conceptual qualities, instead of actual concrete "look this was literally my image it made i published it two years ago," then our spaces will cease to exist.

again, that's not to say there's zero valid complaints about AI art. i definitely see the problem with corporations trying to replace human labor with them because the humans wanted better working conditions. (this is a larger problem imo where any progress in the workplace is stagnated because it will inevitably be used as an excuse to fire people instead of give them less work). and a friend explained to me why it's fairly offensive at the very least to try and bring AI art into what was clearly a space to celebrate the process of making art (eg, contests, writers groups and websites, etc). i just don't think this specific argument is a good path to go down.

#this is a bit disjointed #tldr: im an extremely practical minded person #and i just dont see the practical things we're arguing about here #and im very very concerned it will do more harm than good to transformative artists #copyright laws never benefit artists they only benefit corporations anyway

34 notes · View notes

feline-evil · 1 year

Text

So! A lot of stuff happened yesterday regarding Deviantart and a lot of information that was being spread was inaccurate and fear-mongering, and while i do not think the site is wholly perfect nor that they didn’t make mistakes i do want to try and put a post out there talking about what was present there and where people were mistaken! The journal has now been changed as has Deviantarts NOAI system (for the better i must add!!) but lets look at what was present for people to see when the discussion kicked off! I’m putting it below a cut because its long but please do read, i do want to try and combat misinformation and fearmongering, and do my best to reassure people while also pointing out what ARE missteps and mistakes on Deviantarts part!

First of all let me be clear, do not agree with AI art, do not think DA should have their own AI, do not think they should allow AI submissions at all. But. Lets get on to what was written on this journal BEFORE they updated it to what it is now:

IMPORTANT:

DeviantArt did not, does not, and will not use deviations submitted to DeviantArt to add to third-party AI models or training sets, or on DreamUp itself.

DeviantArt did not consent to third-party technology usage of images on our site, which were used to train AI models. In an effort to combat future unauthorized usage, we have enacted “noai” — an industry-first directive alerting AI models of deviants' desires to opt out.

DeviantArt is the only platform giving creators the ability to tell third-party AI datasets and models whether or not their content can be used for training. This is a protection for creators to help them safeguard their content across the web.

We encourage other platforms who enable creators to publish their work to implement similar protections. We are making our “noai” directive available for use by all.

You will notice in this section the first two bulletpoints. Bulletpoint one explicitly states they are not using your art to upload and add to third party AI, AI training, or even their own AI. This is stated several times throughout this post! The opting out of ai usage does not mean if you do not have that checkmarked ticked they use your art for this, no, it is in fact about checking on a code that prevents what is ALREADY HAPPENING AND HAS BEEN SINCE THE INFANCY OF AI ART THEFT. Which is that sites trawl deviantart for art to use and usernames to acquire for prompts people can shove into their horrid ai programs. Deviantart has nothing to do with this, and states here they do not have any desire to submit your deviations to this. Right now, before this update, your art already was subject to AI and they were not doing that and had no control over it! Sorry to be the bearer of bad news but most the fear-mongering is about something that. Is already happening. Not having the checkmark that says 'i do not wish to have my art used for ai' does not make it any worse, thats just status quo, same as it ever was. So no. Not checking that on does not mean you are now allowing your art to be taken by AI and that it wasn't being before. It means things stay as they are. Which again, your art was already subject to AI, this was not going to make it any worse.

Bullet point two! Deviantart clarifies that, since they do know and are aware of how many highly popular artists currently on the site have been targeted by ai in the past, they have had no sayin this! Which is why they have come out with this update, to try and combat this! This update was not about allowing AI to access your art, it was the complete opposite.

There are several points within this journal that clarify this over again, when they talk about that checkbox stopping ai from using your art they are NOT saying it aill be used if you dont, only that it is the status quo of them having no control over this and that they are not selling nor giving out your art to AI because they do not consent to that and neither do you. There was never a plan listed here to allow that to happen. The username opt out is also aiming at software that is ALREADY AND HAS BEEN FOR AGES using peoples Deviantart usernames to generate art in their style! Not opting out does not mean deviantart automatically goes 'oh yes you can use this person now here you go art theft machines!' It means the code they have made and invented now isnt applied to your account, so you have the same level of protection that you have had for years before this journal! Which is that, yes, AI of any kind could and had for many high profile artists scalped their work and style! Which, yes, sucks so very much!! But i do not think representing this as not opting out makes this any worse is accurate, it will just stay the same; bad. This doesn’t mean not having it be opt out as default was a great move, of course, i’m not saying that.

Now. Do i think everything they did was good? No. They should not be involved in making an AI system, even one that is more ethical as they seem to be trying to make theirs; Not rolling out their AI defence system as automatically you are given their new defence toggled on as default was also a misstep, but people i believe were fear-mongering far more than was helpful for artists trying to understand what was happening to a platform they use; same as the need for a mass toggle opt-out for pre existing deviations. Not because without that defence on and that toggled not off your art would be more susceptible to AI than it had been before! Again, cannot stress enough that was clearly never stated and that they state the opposite! It is still bad that the username out was not set automatically, it should've been opted out by default, again i am not defending that i think it was a misstep that undercuts the vastly positive changes they were bringing about, as was dipping their toes into AI in the first place. But, within hours they changed that. It is now opt out by default, and there is a mass system for previous deviations, they have listened and updated their system! This is good, this is a good thing, these protections they are implementing are good; while im sure third party software may still wriggle around, there's truly not much we or deviantart can do if that happens; this is a step forward for the site! This is good progress! They are still making a mistake allowing AI art on the site at all and having their own AI system, but truly a lot of what was mentioned in this journal was not what people were saying it was at all, there was an intense level of misinformation being spread that clearly went against several important bulletpoints mentioned several times in this journal. I am of the belief they should drop their AI software entirely, and that AI art should be not permitted on site, but i do commend them for actually attempting to give artists more power over an already prevalent issue. I'm not dick sucking here, i fully believe in holding the site accountable for their mistakes, but i do not think fear-mongering and misrepresenting information is helpful for anybody, hence why i am rambling in this post.

So TLDR: Your deviations were already subject to the things people were fearmongering about before the opt in/out system. Yes deviantart has made mistakes in getting involved in ai art and i am not defending their creation and allowing use of their own system. Yes they made a mistake not flagging everything as opt out as default but that has been rectified. No they were never planning to scalp all your previous deviations and give them to ai, that was already happening without their control or say, you had been opted in for years with deviantart having no say nor power over that, because AI art is art theft on a massive scale. Do i want to go on the line in saying Deviantart will never change their stance and might fuck this up worse and undo all the extremely good things they have implemented? No, i don’t work there and i’m not affiliated with them. My goal here is, to the best of my abilities (not claiming to be an expert i have just read through the post several times), purely to point out things that were misrepresentated in the immediate reaction to this update, so that people can make a better informed decision on their feelings and support of the site. This was a flawed introduction of a very good and helpful system, and it’s a shame that A) deviantart also tarred it with making their own AI, and that B) people misrepresented the facts of what was announced, because this NOAI is a really good thing that is being implemented. That is all.

(Also don’t come here to argue with me about wether AI art is good or bad, you are not going to sway me into being positive about it; i have very strong beliefs about it and i am not here to argue nor debate.)

#long post #Deviantart #Ik this isnt normally do on this site but i just did want to help here a little!#if this makes me catch heat so be it im genuinely #just trying to clear up some stuff that i think was causing a lot of distress and upset for artists #i also dont know what tags to use im just sticking it w a deviantart tag and??? Thats all i know to do

16 notes · View notes

lord-radish · 1 year

Text

I've been thinking about ChatGPT and AI a bit lately. There's a lot of angles to it that I think fall by the wayside, and I want to really get into how I feel.

Like as a technology, AI-generated text and art is really interesting! We fed a computer so much external stimuli that it can receive a text prompt and generate a brand new image with elements that it associates with the text prompt. Ten years ago, the most we could get was a fuzzy outline, like a dream. Now you can get noticeable, recognisable images. That's insane.

AI Dungeon is awesome. ChatGPT is one thing, it's a step above Cleverbot, but with its drawbacks (absorbing so much of Reddit that it has the capacity to be bigoted, and having to be "shackled" in a way not only to prevent bigotry, but to avoid lawsuits when people ask it how to make napalm). And then there's the question of how both programs got that information that they're trained on, which is where my support begins to slip.

Because technologically, I do think AI generated content is really fucking interesting and cool. There's actually something to it as opposed to some horseshit like NFTs, which was selling a problem for it to be a solution to. It was greed and misinformation with negative practical use - it complicated the process of tying purchases to an account by inventing a series of different platforms and accounts tied to different fake currencies. NFTs had no practical use and continue to have no practical use, it literally just exists for the sake of profiteering.

AI, as a technology, actually has a point. It is actually something of an achievement, and it actually produces something of interest, or even value - art as viewed by a computer.

But with the AI boom - which feels like the NFT boom all over again, being the hot new tech bandwagon every corporation is jumping onto - new questions arise such as surveillance and automation.

Surveillance is inherent to the nature of neural networks. To train an AI, you need to feed it an unfathomable amount of content for it to form context clues and define a "vision". ChatGPT was trained on Reddit, to my understanding. AI art needs a ridiculous amount of art. Things that you made are being taken, without permission, to inform these machines.

DeviantART - who are bad enough already, what with owning everything you submit to their website and selling original designs to companies like Hot Topic without crediting or paying the original artists - announced their own AI art program that would be trained on the entire DeviantART portfolio. If you as a content creator don't consent to your work being used for that, there's almost nothing you can do.

Automation, on the other hand, isn't inherent to AI-generated art. That is a human issue - or rather, a corporate issue. Because anyone who's even remotely clued into this shit knows that AI art, either visual art or written content, has noticeable tells and flaws. It's fun, but you can't replace the entire spectrum of human expression with an algorithm.

The issue comes from companies like Ubisoft, Netflix and other massive media conglomerates who immediately jumped to AI as a cheap "baseline" writing tool, choosing to lower their employed writers to the level of revising AI-generated text. Not only are they lowering pay, but they're lowering/eliminating residuals on top of what is already an ongoing attack on art in the name of greed and profiteering.

We're watching TV and movies - easily accessible content - disappearing from digital services which operate at a fraction of the cost as, say, DVD manufacturing. And while licensing rights and all that is one thing, the reason stuff is disappearing is because companies are using them as tax write-offs. You straight up cannot access shows on digital platforms because a company like WB/Discovery wanted to avoid losing money. That is literally the only reason.

And I think the push towards AI-generated scripts stems from that greed, as well as these big network dickheads wanting their NFT moment, or their Bitcoin moment. They want to use the latest fad to get attention and fuck over their employees for profit, and then when AI as a serious proposition goes belly-up - which is absolutely going to happen, because AI-generated content isn't financially viable - they hire the writers back full-time for the same reduced wage they implemented during the AI bubble.

I don't think it's ChatGPT's fault. I blame Ubisoft for spearheading the movement to shift writing staff to revising AI-generated content. I blame every media company who followed suit, and for fucking writers in the ass for decades regarding residuals. We're seeing industries like video games and movies making billions of dollars a year in profit - not revenue, profit - only to condemn years worth of entertainment as a tax write-off and jump on a fad as an excuse to further degrade the working conditions of an essential work force.

And yet with that being said, I do have more specific issues with AI generated art that I'm going to get into.

AI prompts can get flagged by whatever software is being run, and those flagged prompts are being managed by human teams. There was a news story the other day about a team of African content moderators who were being denied counselling and other healthcare options despite being inundated with sexually explicit, violent, illegal and morally upsetting content for the entirety of their work day.

Companies that run AI art programs hire cheap labor, expose them to the most uncomfortable and harrowing content that the programs can produce, and refuse to pay for counselling when those workers are traumatized.

AI-generated art is a boundless medium, for all that can entail. As it stands now, AI art is being misused by corporations - in an effort to make it more consumer friendly, some of the poorest people in the world are being exposed to the worst possible creations of these programs. That's a decision that someone made for AI art to be commercially viable, and it's fucked up. Companies are trying to fuck over writers and artists so they don't have to pay them - again, that's a decision that a whole lot of rich jerkoffs decided on.

AI-generated content can produce just about anything, given that it's trained with enough material to form context. But "anything" means ANYTHING. It can roleplay as your grandma or pretend to be another person ala Cleverbot, or it can generate hate speech. It can show you a picture of Mesapotamian Spider-Man, or it can show you a depiction of a sex crime. Those are huge risks, and in trying to shackle the less savory aspects of AI-generated art, more people are explicitly and purposefully exposed to it, and subsequently exploited when it becomes "too expensive" to take care of them.

I think that AI-generated art is a really interesting topic. I think it's an incredible technology. I think it's being used incorrectly, and I think it has some deep flaws that call into question whether AI generation - whether its output is moderated or not - is ethical or conscionable. And it bothers me to see it reduced to "AI art is soulless and devoid of meaning, only people can make art".

Like the only reason AI art is being championed by rich dickheads is because it saves them money. Everyone knows that AI tech isn't good enough to replace human creativity. There's only a push to replace human content creators because it's cheap.

I also think seeing the possibilities of a computer generating art is incredibly fascinating! Like no, I'm not gonna buy it as a print and hang it on my wall. But the fact that a computer program was able to pull so many elements together and create an image that's even remotely as coherent as it is? That's fucking insane and cool.

It'll never replace human art, and I kind of take issue with the idea that AI art is bad because it's "soulless" or somehow "lesser", as if the point of AI art is the same as a person drawing a picture. I think companies who focus more on profit and who don't know/care enough about the craft of making art or writing are the ones devaluing the work of artists. Those companies using AI art as if it's even remotely comparable is an insult to actual artists, I can agree with that. But the AI itself isn't the problem, it's how it's being used.

Rather, AI art has a whole other set of ethical quandaries relating entirely to itself - such as consent to use pre-existing artwork to train AI programs, the existence of hate speech within the program and how to deal with it, the surveillance capabilities of a machine that absorbs everything you feed it, and the exploitation of people who work on the backend. It's an incredible technology, but to make it work - and especially to make it palatable to the public and to corporations - there are so many ethical concerns to AI art that get swept away by surface-level "can a robot write a symphony" takes.

In short, the topic of artificial intelligence and the ethics of using it to create art is a broad and messy topic, and I think there are arguments that are applied broadly to AI that are mis-attributing broader problems to AI itself. There's a whole world of ethical concerns and discussions relating to AI content generation itself, and I think that those concerns aren't as prevalent when the focus is on attacking AI for decisions that boardrooms chose to make. There's so much more to talk about.

#long post #artificial intelligence #ai art #I'm open to different perspectives so don't drop a nuclear bomb on my ask box please #this is just an opinion that I have at the moment. as things change my viewpoint will change accordingly

0 notes

bananonbinary · 7 months

Note

Honestly, your post just goes to show that you really don't understand the widespread non-consensual use of copyrighted images in AI art and AI art-tool training.

The AI isn't "just storing" your artwork. What the tool has done is mined and scanned thousands if not hundreds of thousands of pieces of art, and uses them in order to "create" another image.

Essentially, this is like me buying a load of magazines, cutting out little bits of them or tracing over certain parts and compiling all these different aspects into one image, occasionally filling in the blanks with my own original components. The end result might not wind up looking anything like one particular original source piece, but its still been done using them.

(Although in the case of AI art tools, there has already been instances where the AI has directly replicated an image and simply changed the tonal presence, style or minute components within the work. The Lion's Head instance being one of the more wide-spread.)

This means that as an artist, you might be scrolling DeviantArt one day, see your photo with what looks like a shitty filter on it, and come to learn that its someone's "original AI artwork" that they're claiming to be the original creator of. You've then got to go through the ordeal of filing a DMCA or a similar copyright claim to get them to take it down.

But. That doesn't actually stop the root of the problem. All you've done is taken down one person's recreation. The AI tool still has your artwork and is still churning out pieces using your artwork. So if you want to weed the problem out at the root, you've got to find which AI tool they used and go through the very time consuming task of forcing the operators to remove your artwork from the AI tool's "memory" and block them from re-entering your artwork into their databanks.

Now; literally all of that could've been avoided if the AI tool hadn't been trained by mass-mining opyrighted works in the first place. There are literally hundreds of websites that host open-access artworks, images and data that AI could've been trained off of. But because people wanted to profit from the AI tools and wanted them to be able to replicate and utilize popular art pieces and styles, they simply said they didn't care and did it anyway.

(And are still openly admitting to it, why they did it, and how bad it is for them that people are now creating tools that scramble the AI tool from being able to scan and replicate their art.)

That's what is bad. That's why we don't like AI art tools.

i'm confused, the scenarios you describe here:

and here:

are extremely different. ARE people finding their own artwork with a shitty filter on it? because yes, i'd agree that's not okay. and i will reevaluate my understanding of the topic if there's evidence that's actually what the bots are doing.

but the first thing you describe is just the art of collage, which is a) not morally repugnant and i actually think the laws surrounding it could stand to be more lax, and b) still not really what ai actually does. as far as i understand it, it's not transplanting anything directly, it's just saying "based on how 300000 drawings of trees look, there's statistically usually a leaf here" or w/e. which is why i say it's "storing" the images, because it's just running statistical analysis on them, not actually copypasting anything directly. again, if people could say "hey that's my tree it stole my tree" i'd accept it, but as it stands i just don't see how there's any recognizable reproduction happening.

#disk horse /

12 notes · View notes