AI content warning label
Edit: my post has been deemed malformed, and I’d like to apologize and clarify to the community. The concept of a digital watermark signifying that the artist didn’t use any image generation, LLM, GPT, etc is the proposition. I do understand it’s tough to identify the term AI in use, since most of our tech uses some form of code to modify our work without our knowledge. More-so, I mean to identify work, art, or content that did not specifically use tools to create. Again apologies!
Post: I’m wondering the world of Tildenisian thoughts on this. Say I make a piece of art, no matter the content, and it’s completely of my own hand. Should there be some kind of digital watermark to signify that accomplishment? Maybe accomplishment isn’t the right word.
I must be looking for validation, because I’ve made art recently where folks have asked the question, “What tool did you use?” and immediately felt dread and disappointment.
Perhaps it’s not even feasible to signify since “AI” is eventually impossible to circumvent when sharing your art over these series of tubes. Oh well.
What do you fine folks think?
This is a malformed question. I don't mean to sound like Stack Overflow here, but you have too many undefined terms/phrases.
If you use Photoshop, did you use AI? Probably? You probably didn't intend to, but it has a lot of AI for some definition of AI.
Did you take a picture with your phone's default camera app? It probably did some AI cleanup.
Your best option is probably honestly to just explain the tools you used and leave the judgment or lack thereof to the person seeing it.
I don't think any kind of technical answer is the solution here, just the artist's reputation and word.
I think in the context of general understanding, it's pretty clear what people mean when they say AI in relation to art. They're not talking about fuzzy selection or automatic color correction or lens distortion correction or anything like that. They're talking about wholesale diffusion type image generators.
On a technical basis it's pretty muddy what constitutes AI, because it's not well defined even within the field. In the general publics understanding though, it's pretty clear.
Right, the technical muddiness is what causes the issue with any kind of "certified non-AI" or whatever technical solution is suggested, and it is why a list of tools is more useful for a reader.
I saw an artist's Instagram profile include the phrase "No AI" and I think it's easy to understand in the context of the controversy currently surrounding generative AI. It's shorthand for "I don't use generative AI to make my art".
As an artist, I'm considering using this kind of phrasing for my social media profiles too.
I don't think there's any need to quibble on the technicalities of "what is AI" within the context of the art industry-- obviously we are referring to generative AI tools such as Midjourney.
Thank you for understanding. I hate to have to put such a stamp on such things (I think back to Parental Advisory on physical music in the 90s) but a universal watermark could be useful for the future. I’m just not savvy enough to figure that out.
I think it should be the burden of those using AI to disclose that.
I realize how naive that sounds, and it's unfortunate that we find ourselves here.
I see what you are proposing as less "parental advisory" and more "certified organic."
Not trying to undermine your point at all, but perhaps a slightly more apt comparison instead of Organic certification may be "Non-GMO Project Certified" since it speaks directly to the nature of the inputs rather than just how they were handled or what they were exposed to during growth and processing. But it's perhaps an unnecessary comment.
I think you're better off including that as metadata around your art versus on the art itself. Personally I find watermarks other than maybe a signature distracting in art, and it really takes me out of it. It would be doubly so with some "NO AI" mark. If you put it as a note sort of like how paintings have the medium they were created with alongside the title of the piece and artists name on a placard next to it in a gallery, it might be more appropriate.
"Digital art without generative AI" or something like that.
Ah I see the confusion. My suggestion for watermark is only in context of signifier, not physical or apparent on the art itself. Say you write a song and release it through CD Baby, it goes through whatever test (this is where my question gets too technical) to signify it’s human made, and it’s digitally imprinted with some code that can only be significantly detected. Like when you bought something on iTunes two decades ago and it was coded with piracy software… I’m still in the weeds thinking about this so I’m trying to make sense of it too.
Edit: in rereading your comment, I think we’re talking about the same thing. My bad!
Ah yeah, something not part of the image would be great. It would have to be entirely based on the honor system though.
The thing is that there's no reliable way to automatically detect AI art. There fundementally never can be either. That makes sense if you think about it. If a computer program can detect if a piece of art is AI or not with some amount of reliability, you could just include that code in your image generation engine, and when you wanted to generate art that passes as human made, you just keep generating images until the detection system no longer flags it as AI art. Then you train your image generation model on those images being acceptable, and suddenly you have a model that rarely generates art that gets flagged as AI by that particular system.
It becomes a never ending chicken or egg situation that can never be fully resolved, so there's no really good technical way to do this.
That was my fear, and I suspect my proposition will never be reality. I'll just have to know it in my heart I made something myself, and those providing my content are trustworthy. Soon it will be unnoticeable though.
I hope you find the right method for you! I haven't decided yet for my art. I might just write "No AI" in my profile too. But yeah I also hate that I even feel the need to declare that, just because of what's happening with generative AI these days.
I think AI disclosure is kind of unnecessary. Think about why people tend to not like AI generated content; It can feel generic or have some undefinable quality that people don't like. If someone uses AI tools to help them write or make some art, it's not automatically upsetting to me. I can evaluate any art individually and decide if I like it or not, but having a guarantee that it is "authentic" doesn't add any value for me.
Even if we judge art on how much effort we percieve the artist putting into it's creation, AI art isn't automatically low effort. We can already see that a lot of AI generated content is unappealing to humans, so if someone is able to make art with AI that is actually appealing, they must have put a lot of thought and effort into using AI tools in a unique, creative way.
Recently I discovered a musical artist on YouTube whose music works very well for fuelling my work process. Discoveries like this are rare and thus, very valuable, for me. Some days later I clicked on a video that came up on the side feed that looked like it's going to be from that channel. It started off nicely and I started working, and... soon something felt off. I was surprised at the gimmicks the composer would resort to. They grabbed my attention and my work process became hindered while I was semi-consciously considering whether to keep the track on or not. After a while I caught myself feeling uneasy again. This time I put my full focus on what the track was, realised it consisted of very short clips that didn't form any sort of coherent whole, and at that point I knew it wasn't going to be from the same channel at all. And sure enough, it was an AI copycat, obviously designed to lure fans of the original channel, down to having AI make similar looking thumbnails for the videos. Many more similar channels have started to pop up, churning out low quality content in high volume, covering every possible area of interest.
The problem is that I am a human being. I don't have endless hours in my day to spend consuming this and that piece of music just to determine if it's high enough quality to put on while I work. When generative AI facilitates insane amounts of too-low-quality content to flood my feeds, it becomes a time consuming process, almost like another job, trying to find the things I actually enjoy.
It's like trying to cook something that requires milk and having a hundred cartons without expiry dates in front of you. You'll have to taste them one by one. Some of them will taste only very slightly off so that you're left wondering if it's good to use, others will be very unpleasant, and only a few will be perfectly good. Suddenly it takes ridiculous amounts of time for you to perform one simple task that used to be frictionless when expiry dates were still being used.
I used to think labels were unnecessary too, but I don't anymore. I'd go even further than that: I need a platform that completely bans AI generated content. Generating slop is so much faster and easier than composing and recording music is, that the former will drown out the latter if they're allowed to co-exist.
This sounds like a frustrating experience, but I don't know how an AI disclosure would help in this situation. If someone is going to be dishonest enough to copycat another musician with generated songs and thumbnails, why would they not just lie about if they used AI or not? It kinda relies on the creator of the art to be honest, which doesn't seem to be a strong suit for anyone abusing AI to make an easy buck.
Sure, I was just saying that disclosure isn't unnecessary. How platforms/society can get people to actually disclose is a different matter.
If I was a tech bro in Silicon Valley right now (attention, tech bros of Tildes!), I would try to build a tool that analyses publicly available content and puts it on a scale of expressive uniqueness. AI generated content is generic and would receive low scores using this metric, if done correctly. This could be used to filter out / remove / ban such content.
The downside is that some human generated content would also receive a low score. In my personal opinion this would actually be a good thing. Very generic stuff doesn't really deserve to exist next to pieces created by a highly developed professional artist with rigorous discernment and expressive focus - not even if it was created by a human. Platforms could differentiate and serve different user bases and purposes.
The real downside, I guess, is that the appropriate definition of expressive uniqueness is hard to achieve. What makes something expressive and unique, and not simply.. random and unique? As well, if such a tool existed, generative AI could be trained on it to make content that fits the definition and I would probably kill myself because there would still be something wrong with it, sort of like there's something wrong with having potato chips and ice cream as your main diet, but your body doesn't necessarily give you clear enough signals to realise that before your health is ruined.
There's actually been a lot of effort put into creating tools that can detect AI generated content but it seems to not be going very well. Everything I've seen so far has a huge problem with false positives, I doubt it's really ever going to be a completely solvable problem.
My suggestion is actually a blandness detector rather than an AI detector, but I'm not saying that would be any easier to create.
But, actually, it would be interesting to see some of those false positives. What if they're exactly the kind of human generated slop I'd love to remove from sight?
I think the issue here is that human art is also largely derivative. That's not a bad thing, it means that the body of art can grow in new ways, and we're not stuck doing cave paintings still.
The main difference is that humans have lived experiences which are the secret sauce which makes new art. If everyone has the same lived experiences, we'd just be making the same cave paintings for tens of thousands of years..
I don't know how you can measure what's a new lived experience which fundementally changes art in some new way versus just another small remix though.
I agree completely. I can recognise some earlier composers that the composer in my above link has been inspired by, but they have a unique way of drawing from those influences - a lense or filter they used when creating, and that filter is their entire personality. A skilled artist learns, over time, how to be true to their personality when creating.
You've hit the nail on the head. It's such a growing issue, and this is simply just the infancy stage of said problem, that there will soon be indistinguishable content from a human to a robot making art, writing a story, composing a symphony, or writing a movie. I agree with others here that when AI generated content is actually good, that there's a human behind tweaking and using that tool in an artistic way. That being said, if it's to your point of trickery, mimicry, or like those AI musicians with 300,000 followers within a week, I find that problematic and it should be garnered for us, the consumer to decide if we like it because of its origins.
I think my advancing age may influence my view, but to me the point isn't really about the level of quality. I see viewing (or listening) to art as a kind of interactive communication, and it is important to me to know with who or what I am interacting. It's like back in the day sometimes there's be a story about a dog or horse or whatever that painted. Maybe it was good, maybe it was art, but I'd still want to know that it was a dog-- and also if it was a dog with a lot of human help. Even now, there are stories in the art world about some old master's painting where questions arise--was it the master or some underling painting under the master's direction? It isn't that it's not still a good painting, but the process and ownership do matter.
Art isn't about the product to me, and probably most people. It's about the labor, the emotions, and the ideas that the person poured into it.
Like, most people aren't looking at the Mona Lisa because they think it's a great painting or because the image quality is so spectacular. They're looking at it because of who painted it, the fame, noteriety, and cultural impact that specific object holds. If it wasn't about that, you could just save yourself a ticket to the Louvre and a trip to Paris and Google an image of it real quick.
When I look at a Zdzisław Beksiński painting, I'm creeped out because I imagine what emotions he was feeling and the thoughts he was thinking when he created those images. If I saw the exact same thing made by an image generation tool, I'd just think "huh. The tool must be kinda broken". It's not intentional, because there's no mind behind it with an intent. There are no experiences informing the artistic decisions it's making.
It's just cheap mass produced crap in my opinion.
When they asked "what tool did you use?" was it in the context of Ai?
Because "What tool did you use?" has been a very common question for decades whenever you show someone your art. Bevause they think if they use the same tool, their art will be better.
I would tell them the tool, and then tell them it doesn't matter. Your art foundation matters more.
Sorry if that's off from what you're asking.
But if the context they asked it was about Ai. I would tell them no. I drew it myself. I also think that the burden of proof shouldn't be on the artists NOT using Ai. AI artists should watermark their work as Ai. As it's a far newer tool and there's significantly more ways to create art in many mediums that isn't AI.
Unless you want to go down the traditional art style of labelling. Like created with Acrylic on Canvas. You could also just leave a description saying created in whatever software you use rather than a watermark.
I know the software isn't important, but if you think of it as a traditional painting, you're just writing what your medium was. Very normal. Very established way of labelling your art.
Tool was meant to signify what AI I had assumedly used (ChatGPT, Grok, Gemini, Dall-E, etc). I like the idea of your suggestion, labeling, as it's proven to work in the museum/gallery world for however long. Fraudulent art and recognition of craft is the more prescient task unfortunately, and the near future will show us that we won't be able to discern human vs robot content.
The best proof I've seen is when artists have a video/stream documenting their work process so the audience can watch the artwork evolve. Digital artists sometimes provide (often for sale) their work assets like layered Photoshop files, which as I understand it, is difficult for AI to produce, but I'm not sure if that's still true.
The following is tangential as it's a physical rather than a digital watermark: There is some research on light-based watermarks that can be applied in real life settings while recording a video, as one measure to protect against deepfakes. The light signatures are supposedly difficult to reproduce with generative AI available today. So if someone is recording a video on a camera they could potentially use some kind of physical watermark like that.
This is fascinating. I had no idea this existed!
A nutrition label-esque system might be interesting? It looks like Twilio attempted one a couple years ago, but that’s not gone anywhere. It might be interesting to see it as applied to artwork? I don’t particularly mind if works have some amount of AI generation in them, but others’ tolerance levels differ dramatically. Could be a nice way to strike a balance.