I use AI in design work for generative fills. It’s mainly to pad out stock images so they can fit better in templates for socials and web ads. AI is also useful for getting rid of blemishes, red eyes, and other cosmetic adjustments. I’ve stopped using text-to-image for the most part. Our company wants reports for AI usage, but generative fills are excluded, and I’d rather not fill out the extra paperwork anyway.
Adobe Stock has a lot more AI-generated images nowadays, and we aren’t so keen on using them, because they have a kind of plastic sheen, especially on black people. There are ways to use Photoshop filters to add noise and speckle to make them look less perfect, but it’s mainly a last resort if we can’t find “real” people images.
AI is a really useful tool, but it’s no substitute for human talent.
I was open-minded when I first heard about the idea, but I just don’t feel anything that was created by AI. It doesn’t mean I couldn’t theoretically feel something in the future, just that I’m not holding my breath.
Have there been blind controlled trials studying people’s reaction to AI? Can they tell it’s AI? I’m curious about the reaction to AI art when the fact that it is AI is unknown.
FWIW, last year The New York Times conducted a blind “taste test” of AI-generated and human-authored passages. 54% of the 86,000 participants preferred the AI writing.
If it is great, or even just good, who do I praise?
If it is horrid, who do I condemn?
If I put up beautiful paintings in my house, should I be commended for my artistry, or my ability to buy things?
I’m interested in that link. I am already subscribed.
Fiction? Non-fiction?
I mentioned in another thread, I’m a romance writer, and I’ve read some bad romance, but a lot of readers don’t seem to notice it’s bad. Thousands of five-star reviews. And so there’s a pretty broad market out there for readers with no discernment, or at least willing to suspend their discernment for the sake of whatever these terrible books satisfy for them. I don’t understand it.
I think if I were to read a bad romance novel written by a human and a bad one written by an AI, it might be difficult to tell them apart. Which means a good portion of the general public will accept shit for entertainment, and possibly might even prefer it, especially as our attention spans and tolerance for complexity have evaporated.
I realize I’m talking about subjective things here but when I say they are bad, I mean they are lacking a lot of what would be considered in the fiction writing field to be foundational elements of craft. Structurally unsound, hamfisted character development, bad dialog, weak prose… And that’s effectively the same thing I’ve seen with AI writing. I can imagine an era where AI could get better, but what’s the incentive to create better things if people are already accustomed to and welcoming of utter tripe?
I noticed that the results have changed in the year that the poll has been active. Either people’s tastes have changed or there was an active campaign to change the results. The 54% of 86,000 readers result is from just two days after the poll went live.
Okay. I preferred AI roughly equally to human writing, though in actual fact, I hated most of the passages regardless of who or what wrote them. They didn’t really do anything. Just abstract navel-gazing. The only one I sort of liked turned out to be LeGuin. I suspect I would have liked all of them better with context.
This is key to the analysis, though.
We asked A.I. to choose an existing piece of strong writing and then craft its own version using its own voice.
This is not really allowing the AI to “write” the way a human typically writes. By modeling itself off specific passages of human writing, it’s allowing the AI to replicate multiple things that made the human writing strong, and really only altering the prose. That’s cheating. A lot of what makes writing compelling is its structure. If you take the structure of a written passage, copy that structure but just use different words to say the same thing, are you writing in the artistic sense?
A better comparison would be the original passage vs the AI corrollary.
It can be a good writing exercise, though. I used to do it as a young teen when I was learning how to write fiction. I’d take sections of a book I liked and rewrite them in my own voice. Sometimes I’d change the details to suit myself better, but it was the same structure and key events. There are elements of creativity there but it’s a crutch. It’s training wheels. It’s not the real thing.
But as I said, I think most of these writing samples regardless of author were obtuse and uninteresting. If there’s one thing I’ve learned over the last year, it’s that me and the New York Times have very different ideas about what makes good writing.
I use A! a lot, I think I have something like 7000 pages of AI-generated conversations, mostly technical. I also use it a lot for correspondence. I toy with it creatively, but have never liked what it produces.
Art to me is a conversation between creator and audience. It doesn’t have to be particularly profound or important. It can just be “Hey check out this story about a giant robot fighting a kaiju.” But I want to know it’s someone telling me that story, not an algorithm predicting what token will follow the previous token.
I tend to agree with the people who disfavor AI created art, but I suspect there is quite a hazy line to be drawn. I imagine many/most artists have long used technology to create their art - including computer assistance. And I’ve always been a little unsettled about the technology used in much photography. I also imagine I have appreciated artworks that were machine-made/mass produced.
If an artist comes up with a concept and cues the AI, is that enough of a creative process for me to appreciate it? I could imagine AI being used to “complete” a work that was largely human conceived and executed.
But I do have a feeling of dissatisfaction that - in many settings - I am unable to distinguish whether an image is “real” or computer generated. As a whole, when it comes to art, I prefer work which leans towards the human created end of the spectrum.
This is an extremely reductive way to view LLMs. they’re much more sophisticated than people give them credit for. Saying that LLMs are just next token predictors is like saying humans are just neural-connection generators. It’s true, but it’s so reductive that it loses all meaning.
I’m not well versed in what they can do in visual art, but I can say that in conversation they can genuinely reason and show insight. Beyond what they should be able to do just looking at the substrate. There’s genuine emergent behavior there, just as there is when a human writes a symphony despite just being a “neural connection generator.” The emergent behavior lives beyond the substrate.
Edit: This following part is asked to the general audience, not specifically Larry_Borgia who I was addressing above. He already acknowledged that art can be a fun story about Kaiju and I appreciate that.
Does it change anyone’s opinion if we talk about textual or musical art instead of visual art? If they wrote a short story - and text is their strongest generative form - that turned out to be a really well written story, would that change your mind at all?
If they evaluated your top 1000 favorite songs, and identified the elements that you like most, and generated some unique and new catchy tunes based on that that were genuinely pleasing to you, would you reject that?
Or are we just rejecting the 6 fingered photos?
Honestly, while I think there is a mode of enjoying art that’s a “deep conversation between an artist and a viewer” or however you want to put it, most art is not consumed that way. Most people like catchy tunes without having a deep relationship with Beyonce. The framing of pretending that the only way you interact with art is this deep and thoughtful connection connection to a deep and thoughtful artist is sort of snobby and a form of gatekeeping, the way that an English teacher thinks that only Dickens and the classics are really literature and everything else is just entertainment.
Discussion of AI art is difficult because it is such a huge spectrum. You can type “A pretty sunset” into ChatGPT and get a perfectly fine vanilla image. You can set up some huge Comfy node array, train your own LoRAs, spend a llot of time setting variables, inpainting, using ControlNet, img2img, etc and get an image that you put a lot of effort and control into. But, to the casual viewer, how are they to know if you did all that or just typed something into an LLM-driven generator and said “this is fine” to the first image that popped out?
I know a number of people who do AI art, have art jams and contests and treat it like its own medium for them to express their visions and creativity through. Those people have styles I can recognize, in terms of content, composition, visual aesthetics, etc. I have no issue thinking that their images are a product of their own style, creativity, vision and effort and seeing it as their art with AI as the tool they used. How well that translates out to the average viewer, who knows.
I love A! for conversations and brainstorming. Especially when thinking outside the box. It will very often disagree and then when you show it additional logic it will climb right on board. I have tried it on some short stories I write and I just didn’t feel the soul in the story.
Though that English teacher (assuming they earned their degree in the last 100 or so years) likely also subscribes to the “Death of the Author”/“Intentional Fallacy” school of literary theory and thereby flat-out rejects the idea that literature’s value inheres in a “thoughtful connection connection to a deep and thoughtful artist.”
So this is actually an excellent point, and I just realized that “AI art” may be misleading or may mean we’re not not all on the same page.
So when you all say “AI art”, do you mean “art that was generated with LLMs involved in any way”?
Because I’d like to ask you on the continuum where art comes in.
An artist uses a set of pencils to draw on a piece of paper. Pretty clearly art.
An artist uses an electronic art tablet with an electronic pen to draw on a computer. Pretty clearly art.
Someone writes a program that generates art. The art is more mathematical/geometric/fractal, but the person writing the program controls 100% of the output, the program is just executing the user’s exact program to create the art entirely on a computer. Still art?
A person uses an electronic artists tablet with the electronic pen, as before, but they use a more sophisticated suite of tools. Their art program does like fill in objects with color, automatically shade the image based on a light source the artist sets, etc. The artist is doing most of the drawing, and instructing the tools what to do, but the tools are contributing significantly to the picture. Still art? What if the tools are smart and content aware and could reasonably defined as a limited form of AI? Still art?
A person writes a 5000 word prompt that tells an LLM exactly what art they want to create. They define the spatial relationships. The colors. The light source. The art style. The materials everything is made of. The background. Every little detail they construct, just with texts instead of a pencil or an electronic pen. Not art? Are you sure? Because if it’s art even a little bit, then it’s a matter of where to draw the line, not a matter of categorically denying that anything that uses AI as a tool can be art.
I think a lot of you when you say “AI art” are forgetting that a human is involved in the creation process. But we don’t call photography “camera art”, even when advanced cameras are doing part of the image creation themselves. We certainly manipulate the images with photoshop, and photoshop has used smart and content aware tools that do the sort of manipulation that AI does over a decade before the first LLM was developed but we still consider that art.
Our tools for art have become more advanced over time. In recent years, they’ve been slowly and progressively gaining “intelligent features” that allow the users to have additional control in their creative vision but these tools arguably do a lot of the important work themselves. Where do we draw the line between someone using a smart photoshop tool to create a vision they want, and someone precisely typing up an exact image they have in their head for the tool to draw entirely?
We respect that a human being is still transmitting to us a creative impulse. Does that creative impulse disappear when an LLM becomes involved?
AI is a tool, much in the same way a paintbrush, airbrush, or CGI software is a tool. To the extent that AI produced media can be considered “art” depends on what extent the artist imparts his own creativity into the final results.
In the little bit I’ve dabbled around with AI tools, I would say that it does require some skill on the user’s part to describe their vision to the AI to produce the desired output.
But I generally don’t “like” media produced by AI and typically don’t consider it “art”. It appears like unnatural and uncanny derivative slop to my eye.