This was meant to be a much bigger article, but after even more research I don't feel like it hits like I want. Consider this a draft to what I wanted to make, which may not necessarily use the six month narrative from this.
I previously made an article on AI generation around the cusp of DALLE 2. At that time, AI image generation was just beginning to peek put from mostly memetic generation from prompts like "Shrek Bloodborne" or "Ralsei smoking weed", which were mostly incoherent blobs that occasionally resembled the thing.
Needless to say, a lot has changed and it's actually worse than I could have imagined. A lot of what I feared might happen has and the precipice of that looks bad. You have much more division in the space- maybe more so than NFTs which were towards it's waning point by the time I originally wrote my article. In the effort and interest of creating a more cogent argument against AI image generation, I'm revisiting the topic. There is so much more to say now especially with the rise of a more cogent community around AI.
AI Art Generation is an inherently soulless enterprise where artists' work is routinely stolen with no credit. What it produces shouldn't even be thought of as art. What is the endgame? The replacement of art in general probably can't happen for a variety of reasons- legally speaking already the enterprise is dubious and you still need an actual hand to curate the results even if you do decide to replace the whole art team. Curation might be a bit difficult given how utterly varied the art world is, making consistent results really implausible. This is usually how you can rat out an artist who uses AI or copies other artist work, because it becomes abundantly clear that they are not doing their own work.
Still, the very real fact is that AI art is already in use for news illustrations for editorial articles or even larger stories. They look like trash and garbled representations of what they want to depict. But ultimately, there is use cases already in action- and I feel like it is not too far off from being used more widespread. AI is great for creating utterly generic versions of whatever you feed it, with no thought put into purpose or meaning. And there is a market for that- I'm thinking WikiHow articles and people who use stock photography.
The fucking gall to put this behind a subscribers only newsletter is hilarious. Really funny.
Still- the benefits are few and far between. I haven't been able to find many intended use cases that actually saved me any time or were actually pratical. I'll outline three I attempted and how utterly useless AI was. These are not practical, every day functions but still ones that you'd think AI would understand and be able to do.
I truly thought this would work okay.
This one seemed easy for the AI to handle but I could have not been more wrong. I liked the Halo Wars design of Cortana enough but much to my surprise there existed no actual fullbody art. So I thought perhaps if I fed the AI half of the inage, it'd generate the legs I needed.
GAHHHH!!!
No dice. I tried fruitlessly for a while before finding another image of Cortana to graft legs onto.. Again, no satisfying results. Frustrated, the final image looks something like this. There is some minimal AI usage but it's mostly used to try and flow the diseperate elements together- which doesn't actually work and I had to do a lot of manual work anyway. So strike 1. Maybe it can give me an image for a character that is broadly popular but doesn't have a lot of "triumphant" images associated with them.
This was actually a joint effort between me and one of my girlfriends- the AI material maybe only accounts for 13% of the whole image itself.
This turned out to be sort of a bust too. I'm not sure if the terms used to describe this are just too cloudy for it to be of any use, but it gave rather disappointing results. Even trying to fudge results with direct images or other terms I had seen in other prompts turned out to be rather minimal in terms of giving me pleasing results.
Lmao, not even in the ballpark.
What "AI Artists" do not tell you is just how utterly useless their precious tool is for very specific creations. If I wanted something as simple as a Huttslayer Leia picture, this wouldn't be too hard in the grand scheme for any artist that you can pay for. For most companies with pre-existing IP I can't imagine this really does much to sway them to actually use this. It's way easier to just get someone to draw it.
Like where did any of this get sourced from? This is a action figure. This is bad.
Totally normal to kiss a cojoined clone of yourself back to front.
I wanted something seemingly uncomplicated and in the results, I found that AI has particular weaknesses with- multi character scenes. It doesn't get it. It's borderline incoherent, often fusing traits together, tearing bodies limb from limb, all with the trademark soulless art eyes.
She's trying with tongue but her awful nightmare body keeps her away!
I think perhaps it's my fault for expecting copyrighted characters to work within a public AI system, but even then, it's fascinating how it falls apart. Even if I had a more successful run at any of these, it would still be frustrating with the amount of imprecise details I'd probably have to clean up.
If I got a real artist to do this, I'd ask why their limbs are cojoined or why their faces seem to be pressing like clay.
It's interesting that later on I learned that AI already has a preference for women in these scenarios, but the fact it's still this bad raises a eye. I need to stress that I am showing some of the better results of this that are using other people's art as a basis or I'm specifically plugging in artist names. I didn't want to do either one of those, but to give this the fairest shot I had to, and it was only after doing so that I got anything even remotely approaching acceptable? That goes for ALL of these.
What I found particularly frustrating in the six months was the rise of the "AI Bro"- much like Crypto Bros of the past they don't understand the technology and even more frustatingly, any of the basics of art. AI Art is tailored to AI Bros, after all who could forget the dames with magnificantly huge busoms that made headway into the larger discussions of AI?
I have disappointing news about everyone's "unproblematic" internet favorite that either they know is behind stuff like Ultimate Battle of Destiny, Potter Puppet Pals, Two Trucks, Cabinent Man, Bustin', etc- Neil C is hopelessly naive about AI Generation. I don't know if this is just him being centered about the issue because he hasn't looked into it (although to be frank most of his responses seem to imply he's fairly informed) or if he is just trying to see if there's any application for his work.
It's really cute that he thinks that. Like he has no idea.
Here's the thing though- AI Art isn't art. The community mostly consists of people who think being a idea guy will be a job they can apply to in the future- and the fact is you need to be more than that. They tell each other about prompts that worked well and build off that in an amicable way. As tensions have flared between artists and algorithm mice, it becomes inherently clear so many of them wish they were artists and never put in the time to learn. Even the few artists who attempt to use AI in their art, it's stunningly obvious and makes it apparent just how it's not even worth the time to "fix" algorithmic slush. Any artist working in the field simply is not going yo use this because the process negates any of the positives coming out of this. The idea this could quicken workload is insane. Even with the lightyears done on advancing the technology, it still stumbles, producing obvious tells on closer examination if not outright copying things directly.
From all of this, it makes the threat often repeated in discussions of AI art just being a new form of art medium rather hollow. Photography is often brought up, and sure, there are superficial similarities. Photography made portrait painters less useful, it made realism less meaningful for paintings. Of course, art evolved and changed in response, going in more abstract directions.
AI Art isn't photography. It's emulating art that already exists, trying to copy it from a dataset that is often taken from stolen images. The data set does not include the full images it was trained on, but bits of pieces based on it that it then uses in a attempt to produce images from prompts.
Full explanation of how AI Art works here.
The key thing with AI is that is already actively replacing jobs. A lot of these were entry level or more accessible for starting artists, and now those jobs are gone. Automating art through theft-ridden databases is one thing, but supposing there was an AI imageset made entirely out of "clean" images, it would still not be worth it at all. It removes what makes art so great, the element only an artist could bring.
If you're not an artist, this part may be abstract to you but the journey of creating a piece of art is one of life's simple pleasures. It's one thing to have a finished piece of art, but you remove the fun of figuring it out, improving as an artist, getting to be indulgent with talent or pleasing someone that you're drawing for.
AI skips the journey and any self-respecting artist would realize this. There's nothing here to be gained from having a finished piece you don't have any possible ownership over. AI evangelicals have had to resort to being protective over their prompts. Given that they aren't artists, it may not occur to them that ideas are a dime a dozen- even the really good ones! Nobody is going to open a job spot for "prompts". Idea guy is never going to be a job.
I think the larger narrative is that ai can't replace us. That's true to the extent that we value artists and art in general. The truth is we don't- so why would a larger corporation figure? We'll be employed as long as we're allowed to, as long as the clout is there, and it makes the industry smaller and staler.
AI evangelists desire a future for AI to be respected to the same level as the "real" thing- even AI writing is starting to become something I'm feeling a little colder on. Six months ago it was part of my workflow for writing fiction quickly, but newer models are being subject to outrageous claims like being able to replace google or replacing the need for essays. AI cannot be reliable to hinge the entirity of a industry like teaching, art, and more on it.
These are nightmare scenarios that some desire. In replacing the process, we ultimately lose track of the quiet skills and morals we gain along the path to the journey. Humans are too unpredictable to be rigid.
I don't know how the idea of reading a book entirely written by a AI would not be utterly depressing. And well, a picture is worth a thousand words.
ChatGPT is limited in ways that the other OpenAI models weren't- and those models also become more limited over time. The problem with having wide arrays of training data is that you introduce so many elements into it that can be used to form illegal images like child porn or celebrity porn without consent. Given some people have found medical forms in the training data, a lot of these data sets still have the issues outlined in my previous article- and with more aggressive results this time. LENSA often depicted some people naked and a branch of Stable Diffusion called "Unstable Diffusion" has been under serious fire as of late.
It also becomes apparent how bad these newer models are becoming at being tools. Very few AI images are even edited to remove the problem spots and comparing an AI's take on a image it's fed to copy reveals how it little it understands about what it's been told to redevelop.
Not only that but ChatGPT and other OpenAI tools struggle with not being in control of the project from the jump. I used InterKit to aid in writing but even though it is a much more primitive model the function it serves is seemingly much more in line with being a tool than the newer stuff offered by OpenAI.
I think this is the main issue and my main gripe with these new writing models- they are often touted as replacements by their userbase but the premise is flawed from the jump. Trying to do public domain research through ChatGPT proved to be a rather nonsense endeavor- it would either refuse to give me information, would not be helpful, or straight up made up facts. A Google replacement this is not.
When it comes to AI tools my use is perhaps not broad enough for it to be much help but I am really trying to see if there are any benefits to this technology that its apostles tout for myself and I often come up completely empty. As both a writer and artist, neither side of recent AI development has been all that helpful, to be frank!
This is where the idea of AI being a tool, as soft supporters of the technology would like to claim, comes up against reality. A lot more recent developments in this space seems to be designed more as a means to replace as opposed to aid. A lot more of the hardcore AI evangelicals will outright say this and their work also demonstrates this. Books, comics, art pieces all basically written and visually composed by AI, with the only human hand involved being picking sample art and clicking it together into a pdf.
This is the greatest weakness with the AI community and effort as a whole- the lack of effort shows. As much as people would like to tout this "AI is a tool" argument, it often becomes evident that a lot of it's uses have been anything but.
Even as a replacement though, it still largely relies on the work of others to even approach what it's supposed to look like. It turns out even with as much context as possible, it can still churn mangled visual mush.
Still, the technology has undeniably become better with time, or at least it's cherrypicked examples have become a lot more impressive. What's the solution to this problem that frankly never needed to exist?
What we need to do to solve the AI problem is to make it as unattractive as possible. Much of the work is already being done by AI worshippers who describe insanely nightmarish scenarios as if they'll be the norm, but reducing profitability by making AI work open source and public domain will absolutely aid in this goal. No corporation likes the idea of public domain unless they can convince you that they either own it or if it's ubiqutouess enough to be ingrained as a thing people want to see takes on. I'll talk more about public domain later and how it went wrong, but AI can be seriously stifled if it's unattractive.
Lastly, while the notion of having clean training data is a fine idea, it is not really super possible to navigate into court of law. The current movement is a bit flawed.
Much of what more-or-less killed nfts and the crypto world was their barrier to entry as well as the fact it was full of scams. Given that the AI community has yet to clean up their act in any meaningful fashion currently, a lot of the work to make AI unattractive is already being done.
I think if anything still bothers me about this approach is that it still is adapting to a world that frankly doesn't need it. If AI activists got their nightmare world scenario where everything is developed by AI, what would even be left for humans to do? Revel in the tedium of office work, or what's left of that? Creative pursuits were supposed to be what was left behind, not the other way around.
I am not interested in the discussion of whether ai art qualifies as real art. Frankly it doesn't, but it's a completely pointless discussion. In the grander scheme of corporations and larger systems deciding a course of action, I just don't think it can be trusted in the slightest to do the right thing. Many AIs being developed right now still steal art for training. In retaliation to proposing that they don't do this, AI communities merely double down and paint themselves as victims who merely want the lofty title of "artist".
But if the solution is just to open source everything and make it unattractive for corporate interests, that seems relatively doable. It won't entirely fix the issue of this thing existing, but if it is going to exist it might as well exist entirely in public domain. To an extent, this is the current state of things- so make it stick.
