The AI Art Apocalypse
alexanderwales.com
alexanderwales.com
I read the papers. I see the pace of progress. I understand how these models work on a technical level and I am blown away by how quickly they are being iterated on. I expect little to no digital art (images, videos, films, music, writing) will be sold for profit by human artists 10 years from now, and the only reason I haven't extended that to physical media like sculpture or street art is that I don't know whether we'll have dextrous enough robots to make those yet (though dextrous robots will indubitably come soon as well).
People love to bring out the painting-and-photography example as a defense for why AI is not actually going to bring an end to the art market, but I just don't see it as a valid analogy. Photography and painting both survived because they're fundamentally different things and can be easily distinguished in cases where their respective artists choose to differentiate themselves. AI art is different, because its entire purpose is to replicate, and no matter what human artists do with the medium of digital images, the AI will always be right there to gobble up the new wave changes and learn to replicate them.
Whereas the advent of photography was never meant to kill the painting industry, these AI algorithms are very much meant to kill the image industry, whether that was the intention of the researchers or not.
I would 100% take you up on this bet.
> People love to bring out the painting-and-photography example as a defense for why AI is not actually going to bring an end to the art market, but I just don't see it as a valid analogy.
I think that painting-and-photography is the wrong analogy; stuff like DALLE is a continuation of Photoshop. Compilers didn't put programmers out of business, it just moved them one layer up.
I don't know. As Wait But Why likes to say, cars sure did put horses out of business, pretty much thoroughly and permanently.
Similarly, perhaps artists will wish to participate in the economy by making art, but they're going to find scanty employment when compared with machines which can operate 100x faster than they can, and arguably at a higher quality of output.
The human beings previously called artists will still be around as humans and will still have jobs, even if it’s some other job.
“Artist” is a vague term though, and I think we’d still have a lot of them due to Jevons’ paradox, the same way ATM machines have actually caused the number of bank tellers to increase. Especially because it’s high-status for society to say they support artists, which isn’t the case for other service workers like bank tellers.
There is a niche for horse rides, though - they’re romantic and I’ve seen them as tourist gimmicks in different cities. Artists have a bigger niche, as AI art (so far) produces digital images and not, say, physical sculpture or paintings.
This whole debate reminds me of the early aughts, when digital music was said to kill all music. I predict a similar development here too.
Human graphic artists will adapt and the next generation will incorporate AI generated images into their tool kit.
And some of them are already doing that. In Peter Mohrbacher's latest newsletter he shared a digital painting which was started by Disco Diffusion which he then touched up to give it his characteristic look. Given the details and feel of the painting were unmistakably his, I wouldn't have guessed that the base of the image was AI-generated.
When sampled digital audio first appeared in music, it went from experimental to pop hook(e.g. Paul Hardcastle "Nineteen") and then to the subject of court cases that clarified IP law for samples over the span of maybe 20 years. In the middle of that period one can find scare pieces about pit orchestras being replaced by synthesizers and so forth.
The articles tend to be correct in that the same jobs don't exist when they get automated, but you get new jobs instead. And so the panic, where it exists, is tied to whether you specifically are on the chopping block. For digital illustrators who are focused on a high degree of finish and polish, AI art threatens because it's so good at replication of a high-effort stylistic mode of production. Cartoonists are much less impressed since they are already in a streamlined, symbolic mode, so the net savings of AI is harder to come by. With a few years of traditional training, most of it just focused on the core "you can draw what you see" skill, you have what you need to build up a library of reusable visual language for your work: an arm seen from various angles is still an arm, you can add more anatomy to it or stylize it, but it has a symbolic function tied to particular shapes. And an AI can bring out more detail from the symbol, just as when an artist follows a reference more closely and renders it more carefully, but it isn't creating novel representation. That's still coming from the illustrator operating symbolically.
And when you lean into that, it's like, OK: presumably the AI will be able to do some style transfer and turn a crude cartoon into a fancy painting. That levels the playing field a lot but it doesn't change the nature of what makes an illustration good: having a sense of composition, storytelling and so forth. A lot of cultural context things are in there that will evade automation.
- Potentially superior quality to algorithmically destroyed dynamics, but this is truly on a per-case scenario.
- Exclusive content not on streaming platforms, either intentionally, or because labels or artists are dead and gone, or the IP is in limbo.
- Directly supporting artists because streaming platforms pay a pittance.
- Elitism/snobbery. Whether professionally, within a fanbase, as an influencer, in an audiophile community, whichever.
- The community/culture. People love to belong, never underestimate the importance of that.
- USP.
- Distribution to actual brick-and-mortar stores where you'll get in front of tastemakers and influencers.
- DJ culture is huge, particularly vinyl-only nights where they want to show their skills without hitting the sync button.
- It makes a great gift and people like to put them on display as adults in place of posters.
- People take comfort in disconnecting and doing a physical ritual, just like tea ceremonies in Japan.
- Listening bars are important in Japan. To anybody reading, if you're ever in the position to do so, I implore you to check one out.
- One of the few ways to own some music in the long-term.
I could go on for another 100+ points before reaching "digital abundance".
Anecdotal. It doesn't matter if you yourself would never buy a vinyl, what matters is that a lot of other people are. Likewise, it doesn't matter if you would never buy a supercar/helicopter/Breitling/Rolex/swimming pool/jetski/yacht/vegan food/meat/gluten-free/keto-friendly/whatever, because others are.
I never said "all".
> Anecdotal
Yes, I know. Just like your comment. But you listed reasons for people to buy vinyls, when many were simply "reasons for people not to use streaming services". That was my point.
So here's the misunderstanding. The job of an artist is not to paint, or take photos, that's just the medium. Just like a software engineers job isn't to write code.
The artist's job is to elicit emotion. The means by which they do that, whether it's painting, photoshop, or running 10,000 prompts through DALLE till they find the thing they're looking for is irrelevant.
StableDiffusion is a 10GB model made from 100TB of images, so there probably isn't even room to have memorized inputs - there's also some built-in defenses against that.
It's possible to make an AI that does explicitly make collages out of inputs (that goes under "fine tuning") and such a thing would be useful. OpenAI has a similar paper about a text AI that cites its sources called WebGPT.
It moves artists so far "up the stack" that being an artist becomes redundant.
If C-level execs could just ask their computer to generate the exact website they wanted, there would be a lot less web devs out there.
Tools allow _skilled professionals_ to do their job better. The existence of GCC doesn't make the average Joe a programmer; it makes programmers more productive.
DALLE allows any schmuck with zero artistic talent to generate their own art. It obviates the need for artists.
DALLE is to artists what self-driving cars are to truckers.
The same way Wordpress obviates the need for web devs.
> DALLE is to artists what self-driving cars are to truckers.
Always 10 years away from replacing them?
Wix and Squarespace (the spiritual successors to Wordpress) however have undoubtedly put lots (most?) contract WebDevs out of business. There are still plenty around - go check Fiverr or UpWork. However it's impossible to argue that these tools haven't made it easier for non-web devs to build a fully-featured and beautiful site for a fraction of the cost in a fraction of the time. It used to be that these no-code site builders all looked crappy and cookie cutter but in the past few years they've really upped the quality of their templates such that they often are indistinguishable from a "handmade" landing page.
I think the path is likely to be similar for these AI image and text generators.
Devs had to step up to create something better than Wordpress to stay on the market.
The same way, artists have now to create something better than DALL-E results.
That’s a terrible analogy. Compilers didn’t put devs out of business because it still would takes months of learning before someone that needs a program made could leverage a compiler to do basic things. So they hire devs.
It takes 0 hours of learning before someone that want a nice illustration made for them can do it with dalle/midjourney. Why would they hire anyone to do it ? Some people are going to say prompt engineering is a new skill, but I bet that prompt engineering is going to disappear as fast as it arrived as we build more practical ways of exploring the image space.
I think there's a good argument to be made that photography killed photorealism in painting (or at least, it's an ever-shrinking niche). Photorealism was cheapened, and in response, Modern Art was created.
What's getting cheapened now is transforming works into digitally-defineable idea-spaces. Create a new idea in art, give a machine enough representations, and it can endlessly generate new works within that space.
What's not getting cheapened yet is 1) New idea-spaces -> I've yet to see an AI generate something that could be defined as a new art "movement" 2) New mediums -> I'm sure that something like Dall-E will exist for 3D CAD files sooner or later, but there are a lot of mediums that won't be physically reproduce-able by a computer (think James Turrell). Works in these spaces will remain valuable or even increase in value. And while AI might be able to generate new ideas in these spaces, there will need to be people to decide to put in the effort to execute them. 3) Curation and "found art" -> Deciding which ideas (generated or not) deserve attention.
An interesting aspect of this was a (partial) severing of the connection between a lot of the population and the art world.
If AI means artists need to get weirder to survive, it widens this disconnect.
That might be the biggest tragedy in all of this.
Films have a ton of artistic value, and even today, and a regular joe can appreciate many brand new, critically acclaimed films.
AI will absolutely kill some forms of art, and force others to change in a way that alienates most people. (Pop music and background muzak seem like low hanging fruit, because music can be distilled to numbers and signals much easier than other forms of art.)
But AI will undoubtedly lead to the creation of entirely new forms of art as well.
AI is coming for our creative part, which we imagined would be last in line. It already is doing a reasonable job at it. If it can do that, and will next exponentially increase in capability, what really remains of us?
Humans will probably be stuck with the dirty jobs, though.
you think AI auto-generated movies will kill the human film industry? AI novels? these are both fundamentally different from the kind of recombination that DALL-E is doing.
>the only reason I haven't extended that to physical media like sculpture or street art is that I don't know whether we'll have dextrous enough robots to make those yet (though dextrous robots will indubitably come soon as well).
3d printing
>these AI algorithms are very much meant to kill the image industry, whether that was the intention of the researchers or not.
hmm
Every time people start freaking out about the AI apocalypse, i tend to point out that to my knowledge, there's no learning style AI that's beaten a mario game, especially in a real "human" style (play through each level and learn them). I've seen an AI that can beat mario 1 this way, but it doesn't "learn", it just "brute forces" by trying random inputs, and then save state reloading if it gets a bad output (dies/gets stuck), and then does thousands of those games in parallel. And this ONLY works with mario 1 in part because "go right" is ALWAYS correct. Take any further mario game, where more vertical/backwards exploration is a thing, and its not even possible (mario 3 for example).
It would be trivial of course to just program it to beat every game and get everything, by literally programming the correct inputs (literally a tas) but a task that is expected to be doable by a young human child is, again to my minimal knowledge of amateur research, not even remotely within the realm of possible for modern AI.
As others have pointed out this might become a tool in the toolkit that's kinda like clipart++ meets photoshop, and it will have uses there, but I very much doubt it's going to get any use in those mediums.
Comics/mangas are generally either A. pretty damn easy to draw (in the scheme of stuff that's out there), so i'm not sure fiddling with an AI to get the output you want/training a team to work with the AI's initial output is worth it or B. horrifically hard to draw/stylized. Things like berserk or the currently relevant sandman are probably not something you can get an AI to reliably output.
My first thought with all of this art is it's almost always surreal, and reminds me of custom MTG card art, so while I could see it working better there, I'm still very uncertain this is suddenly going to be THE TECH to take over.
And that's even before my jaded view that things like this ALWAYS have caveats that they don't tell everyone right away. They haven't open sourced the code claiming vague worries about the implications, and hell maybe they really have made the golden goose and are right to worry, but my experience having studied AI casually over the years is more often than not it's marketing bullshit to keep the funds coming in and stop everyone from pointing out how HYPER specific it is and what a fucking phenomenal pain in the ass it must be to get it to work at all.
I'm extra annoyed by this particular team because I followed their exploits in Dota 2 very closely (having been a long time fan of the game), and while you can find all sorts of overhyped headlines about the "amazing" feats it performed, put under a much more critical light with a better understanding of what exactly happened (and the the stuff that wasn't mass reported like players figuring out all sorts of ways to abuse it), it's much less impressive. Don't get me wrong, still amazing tech, but it rubs me the wrong way how they damn well knew they were generating hype that didn't reflect reality.
It is different for movies actually tellong a story, those are about emotions. Once AI can come up with convincing story telling for us humans, us humans clearly are redundant.
My work is in this area and I absolutely foresee some industries/arts getting decimated by the upcoming research in these topics - however, it's pretty hard to put the genie back in the bottle at this stage.
The real question is, why would we want to, if the genie is truly that good?
It's this work that creates the opportunity for masterpieces to happen. If that market goes away, the top end will be affected.
I'm not sure things like fellowships and residency programs can recreate this by just throwing money at the problem. There is something unique about the pressures of having to be bold enough to stand out from the crowd but accessible enough for broad appeal.
I'm not sure there's anything to be done, but I think we should recognize something is going to be lost here.
I imagine at its height, AI would make "commodity art" less valuable. But stuff like Rhythm 0, Black Square, The Lighthouse; that sort of art, by definition, can't be generated. Maybe we'll start valuing the more obscure stuff, not dissimilar to how I value the Google result that actually shows the Julia error I'm getting that only 3 people have seen more than the docs page that matches a quarter of the string I pasted directly from the terminal
In general meaning is the parts of an experience that are salient to its representation as a sign. It's a sign's signified.
Robots lack an embodied experience that would be necessary for them to understand human meanings. They don't know what it feels like to step on a Lego, so a whole class of memes and meanings are unavailable to them as a result.
There are probably things that we, as humans, don't know that we feel but still impart meaning. It would be very hard to get a robot to say something like "as the blood rushed to my ears" in a truly meaningful way. Most who end up using that phrase in a sentence don't use its full meaning when thinking about it (so there's higher than normal pressure in the head, etc) but if pressed can expound on it in a way that would be obviously different than a robot's (GPT3 would say, the meaning of this sentence is [0.5435, 24.2352, 1.3245], nearby sentences I've seen have embeddings [1.5532, 0.3552, 4.2442])
I wouldn't bet on this
AI could have an impact on digital art (besides using it optimize digital pictures in Lightroom), social media (why go to places for likes if that can be had by creating a deep fake, it does link back to the point about "originals" so) and the like.
That being said, forgery, staging and other shenenigans have a long history in photography. As wr talk art, as long as people are ooen about it it's fine. It becomes a problem if art work is sold as something it is not.
It's use online will become as ubiquitous as stock photography, and you'll see the same images used everywhere and for everything.
Digital art as a whole will probably be most negatively impacted, because we will view things much more negatively knowing the tools involved. I already can't enjoy anything Midjourney makes because it all looks the same. I just immediately turn my brain off when I see that signature style. Was certainly cool for a few minutes though.
ShutterStock should be concerned, but artists and graphic designers don't need to worry. Just another tool.
For example, here's some art from a modified Stable Diffusion model: https://www.reddit.com/gallery/wkd77a
I don't know how you can look at this and think, "ah yes, just some more generic junk."
> here's some art from a modified Stable Diffusion model: https://www.reddit.com/gallery/wkd77a
The "fantasy girl" style of paintings has always looked generic to me (even back when it was painted by human artists). They are cute to look at, for sure. And so is looking at photographs of models, unless it is your friend and you miss her. Otherwise the story just ends there. I think in the end images are just story-telling devices. They are a means to an end. The nice thing of AI art is that it'll inspire more people to realize that. And the same goes with words and representations around us. What's important are not the words and representations, but the story they are telling, which in terms has physiological and emotional ramifications, and may provoke thoughts and feelings and actions, or the change of a habit, a new sentiment, a grain of salt, a breath of fresh air, etc. Those are what the human experience is about. And yet we always tend to focus too much at the finger pointing to the moon. I really hope that's the change AI would bring into this world. More awareness to the moon. And then there will be aspirations and new movements as we restructure soceties.
That said, yes, it's still very impressive.
I think it's totally understandable if you don't care about this kind of art and it losing value, but what you're really saying then is "The art apocalypse will happen and that's good". I share this opinion, but I can also definitely see why working artists don't like it!
Most commercial illustrations (esp commisioned by companies/institutions) are mostly created to evoke a longing for consummerism (eg lifestyle advertisments), and it is often mindless. People's interests are in the feeling they are promised the thing would give them, and often not the thing itself. Or, at worst, as a stress-coping mechanism that has direct and grim consequences. eg America's sugar addiction (which are implanted into the population at young age when they are most vulnerable.)
These are not arts. They are Machiavellian traps for the mind.
Arts, in the simplest defintions, are just things one makes to express a certain feeling/thought/sentiment/etc. A piece of art is good when it can connect/resonate with you. It's a human connection thing. Tech would be an enabler for more novelties. But the story has always been there. A story about mankind.
You seem to have completely missed the point of this discussion:
> ShutterStock should be concerned, but artists and graphic designers don't need to worry. Just another tool.
That's the thing I was disagreeing with, and they're obviously talking about artists in general, including commercial artists, not just ones who are engaged with the platonic ideal of art.
And so are urban consumerism, artistic demands from brands, the arts industries, the defintions of galleries (eg Can an amber wine bar also be an actual gallery? What about a BOBBLEHAUS clothing store in Manhattan? What defines a space for "fine" arts really)... people's thinking will change. I can assure you things gon be very different in the next 10 years.
Some people may get angry about it. But that is how society progresses.
There are two types of people: those who keep still and those who move.
More Stable Diffusion examples: https://news.ycombinator.com/item?id=32493516
Like, of the millions of working artists out there employed by various companies, what percentage of their output do you think meets this bar of "I wanna hang it up in my living room"?
Maybe AI art won't be able to hit the bar of seriously moving people to feel deep emotions consistently, but what percentage of human-generated art manages to hit that?
There's a huge market of artwork that I don't like, that I would never hang up anywhere.... That's what this technology will compete with.
Criticism is fine and certainly warranted. Discussion and disagreement is encouraged - that's what makes us democrats after all. I don't disagree with images of Midjourney having a distinct look either, or things getting repetitive, for example. But what does
> Was certainly cool for a few minutes though.
even mean in the face of Midjourney running a bigger Discord server than Minecraft (!) and heaps of articles appearing how low hanging fruit in digital art are obviously subject to be replaced by these apps in the foreseeable future?
Aren't they? Aluminium cars are still quite rare right?
That said, I still agree that we probably don't need to fear AI art. In fact, we might - as artists - want to thank it.
For a long time now a lot of artists have felt a kind of ennui because of the lack of a clear path to innovation or creating something "new" and inspiring. Things have become stale and overly commodified, with musicians, photographers, painters, etc. often complaining that anything they create or see is unintentionally retreading well-worn tracks. You talk to any striving and ambitious artist and they'll tell you there's not much inspiring art anymore.
The way this AI art works is that it essentially re-arranges old motifs in new ways. It's a permutation device that shows you the current state of art (if it has a large enough training set) applied to arbitrary situations. Artists might be able to use it to find the negative space where new exploration is possible and finally start creating inspiring art again.
Imagine coming up with an idea then feeding it to an AI and seeing if it's already been done? If the AI can't express it, then there might be something interesting there.
I think, creatively speaking, this is bad for Bob Ross and good for Dali. I'm sure Bob Ross himself wouldn't mind though.
I think this is projection on your part - likely from the point of being a jaded consumer and/or HN contrarian who sees all modern media as formulaic and uninspired, unless you can demonstrate how you know the opinions of enough artists in enough genres (apparently all of them) to be able to diagnose all of it with "a kind of ennui." Maybe you're a creative who feels this about your own work, but from what I've seen following a number of creatives on Twitter, a lot of them are angry that this technology sources their work without consent, and threatens to take away their livelihood.
Some are embracing it as a creative outlet, but most of them were doing just fine without it.
>Artists might be able to use it to find the negative space where new exploration is possible and finally start creating inspiring art again.
>Imagine coming up with an idea then feeding it to an AI and seeing if it's already been done? If the AI can't express it, then there might be something interesting there.
You seem to be making the common mistake of equating novelty with quality. I suspect this has a lot to do with the differences in how innovation and "hype" work in tech culture, and many creative fields, where in the latter cases it's more accepted to work within established genres and draw upon prior elements and motifs rather than try to find the next "empty, legally distinct slot" in the market to fill and become the next big thing. But art rarely tends to be about doing something completely unique.
Not that it would work anyway - these systems have such a vast corpus to draw from that they would generate something based on an input, no matter what.
>I think, creatively speaking, this is bad for Bob Ross and good for Dali. I'm sure Bob Ross himself wouldn't mind though.
If you think it's any more difficult for an AI to crank out something in the style of Dali than Bob Ross, you aren't paying attention.
I don't think you understood what I said. Of course it can copy Dali now. Dali is old hat. Stale. I'm talking about someone with the artistic ambitions of Dali - except today. Someone who hasn't found their unique artistic voice yet.
As for the rest of your comment, it just seems like you haven't been paying attention to any artistic communities because your claim is basically like saying "there is no inflation" just because you don't really notice it.
I may be focusing too much on "originality", but the overall assessment is accurate: modern art is not nearly as inspiring as people would like it to be. Whatever inspiration means to you, I can't exactly say.
For some glimpses into the status quo of Western art:
- https://www.e-flux.com/journal/28/68012/the-time-that-remain...
- https://scholarscompass.vcu.edu/cgi/viewcontent.cgi?article=...
- https://old.reddit.com/r/ArtistLounge/comments/kgy8j4/how_do...
- https://www.goodreads.com/en/book/show/41784198-the-decline-...
Jaded bastard is a property of the person, not a property of art.
I agree, that's why I called it fancy clip art.
Also, no human has ever experienced, much less can be informed by, the "totality" of the human condition. Whatever that's even supposed to mean, besides the effect of a human being having "cobbled together a collection of data points" through life experience.
You're trying very hard to frame creativity as transcendent in scope and almost supernatural in scope, as if art were a divine spark no machine could ever be capable of reproducing. But history is littered with examples to the contrary, and so far only examples to the contrary.
Fine arts, contemporary artists showing in galleries, mediums that can't be printed on photo paper (sculpture, canvas paintings, etc) will be more resistant to displacement. But the massive majority of commodity art & visuals - adverts, billboards, book covers, article images and so on - will definitely be displaced in the very near future.
And then software will "learn," and the arms race will continue.
That being said, I'm quite enjoying generating pixel art to stand in for my programmer art before I commission artists.
Newer art programs like CSP already have a lot of tools Photoshop didn't, even if it's just 3D models for poses.
Earlier today I saw a video (sorry, lost the link) of an AI editing program (still a prototype). It generates layers of a scene in place, say sky, mountains, rivers, foreground, etc. each of which would be individually manipulated. Next the program seamlessly blends the layers. The host then continued to add objects/subjects and through "in-painting" made all of it blend into a single coherent scene.
This kind of capability is close to being released. This year.
This is literally how you do prompts to AI programs.
How will the software learn? By having their work feed into the algo. The creative space is massive and fashions and currents change. In order for the algo to catch up it would need a massive dataset and by the time this is satisfied the art would have moved on to something else..
This tool will still remain and will most likely be used by artists too, though just as a tool in the toolbox, not as a complete process. Fotography didn’t destroy painting, it changed it in unforseen ways.
Who exactly is being defrauded? Is the dev defrauding themselves? Their future customers? How would anyone even know? It's not like concept art is used for anything external?
I can imagine the standard shifts from quality to scale. I.e a company doesn't give a shit that they have nice design, but rather they make something so ugly that noone could mistake it as being something someone commissioned from an AI, and spread it in necessarily low-supply spaces. (i.e. a new social media company's logo is a green poop emoji that they pay every west coast MLB player to get a tattoo of on their forehead).
I need to write about this experience (mostly so I can feel “done” about it), but my show was titled “Is This Art?” and consisted of around a dozen images generated by VQGAN/VQGAN+CLIP. I sold almost every single piece! Not bad for a non-artist, frankly.
Anyways, I think the key to my “success” was two fold: One, most images took an input image, all were photographs I took in the local area. Second, I was brutal in my curation of what I actually decided to print and include in the show. The keystone piece was actually a warped image of the coffee shop itself! In an art medium without clear constraints, my challenge was to define those constrains for myself. Not exactly a knew problem for artists.
I also chose to overlay the output image and original photograph in some cases, and unmask the photograph in certain locations in the image. I used this opportunity to touch up the output in photoshop, and add/remove artificers and details. I was limited by my lack of digital art skills here, but this was the fun part.
Final thought, a good eye for color and composition helps any artistic endeavor, and the same is true for this medium.
I don’t yet have the curated pieces available online, but that’s on my TODO list. I really appreciate you asking, though, especially now that I’ve got the physical part figured out, over, and one with. :)
In his novel, since the algorithm draws from the human, and the human is changed by the algorithm, it causes weird "dreams" where groups of people suddenly all get really violent for a while, then regress into stupor. Having sat on the midjourney discord and gotten almost hypnotized by watching the feed, I can definitely attest it already is very pleasurable, and that's watching _other_ people's streams that they have to type in prompts to generate. I can't imagine what it will be like when mixed with GPL-4 to generate political memes.
So he basically predicted Tik-tok, wow
Okay, it may be novel to prompt ‘A cat by Van Gogh’ and admire the result, but it’s not using the tools to their fullest. Once the novelty of swirly cats wears off, people will realise it’s just another tool.
Where AI really shines is through combining styles, materials and mediums together in ways that are not possible otherwise.
It’ll take time to find out how to use them to their full creative potential, and that’s something I’ve been trying to do over the last year. I’ve got a paintbrush possessed by a drunk genie that knows the entirety of art history and the possibilities are endless.
The people who say AI art is not art because it’s all done by AI are the same who said 30 years ago that digital art is not art because it’s all done by computer, or who 200 years ago said that photography isn’t real art because…
I do think we’re going to see a similar movement to the Arts and Crafts movement in the 1800s. They reacted to the industrialisation of the decorative arts by deliberately focusing on traditional craftsmanship and rejecting mechanisation.
If art no longer needs to be created by a human, then only the things a human can do will be valuable - ideas and craft.
AI tools need ideas to drive them. Left to its own devices, VQGAN + Clip, for example, is perhaps interesting but not particularly artistic. Grey and smudgy. Throw in a few ideas however, really push what it can do, and the results are phenomenal.
Handmade furniture is more valuable than an IKEA flatpack because it’s unique, someone put time and effort and love into it.
Human-made art will become more valued than before, not less.
These models are rudimentary for now, but will improve rapidly.
Look at the coins with the dragon and his horde for example. The newness will wear off, people will become sensitized to the flaws, and AI "art" will find a niche which is quite a lot smaller than people impressed with technology imagine.
Art and artists aren't dead, AI is not at all "creative" just derivative. It will be one influence among many for actual artists, become a tool for some, and for the rest of us, take the place of "art mill" art for the kinds of places that want to look fancy without putting any effort into it. (I'm looking at you, "luxury" apartment building with the exact same bad, inoffensive abstract print on every floor's elevator lobby)
The inverse of the legal sense of this word is "transformative". If you are making something new, then that debt goes away to the extent that such a debt would prohibit the creation of that work.
Generalizing from these concepts... we don't care so much that literally every bit of art is "based on" or "influenced by" something. We care if the end result is novel or not. This is very much an "I know it when I see it" kind of standard, so it might sound like I'm just talking in circles, but there is a difference.
Currently, it's difficult to get an AI system[1] to generate more novel images. There are certain patterns it knows how to do very well, and if you ask it for something that violates those patterns it gets confused because nobody's drawn that before. This puts us closer to "derivative" rather than "transformative", because it's harder to use the system for the latter purpose.
[0] Existing precedent and law specifically names a few categories of work that are assumed to be derivative; such as sequels and translations.
[1] I have only tried Craiyon / DALL-E Mini, and a handful of very primitive systems, both of which are not the most powerful image generators available.
This assumes that these sorts of flaws will remain just as bad for the foreseeable future. I doubt that's true, just looking at the progress in recent years of other machine learning domains.
Sure, eventually things will plateau, but it doesn't look like we're there yet, not even close.
An analogy that TFA curiously doesn't touch is the advent of photography. What did it do to painting?
Painting became less and less figurative as photography became more and more accurate. The point of painting was to represent reality; but as photography became so much more accurate, that point became moot.
And so, painting went into a different direction. Represent not reality, but feelings, what the artist feels when looking at reality; something that's out of reach of photography.
AI is trying to conquer this as well. Human art needs to go further. Something weird that AI cannot touch.
Real Art, that which moves one's soul, is beyond the capabilities of artificial intelligence for a large number of reasons, chief of which because it is a communication between beings aware of finite existence.
Note, I am not saying the consumer markets will not be flooded with cheap mimicry of art as pooped out by ignorant noncomprehending AIs. That will most definitely happen. And a generation of would be fine artists will never pursue their innate vocation thanks to the misconception that AI is capable of creating Art (which it cannot.)
However, you do not have to fall for the stupidity most are caught. If you have an artistic vocation, realize that indirection and metaphor are your human creative super powers that AI cannot touch. Metaphor requires comprehension and often complex interwoven comprehensions, which is so far out of AI’s reach it is laughable the general pubic believes otherwise. AIs are idiots when you know them well. They have zero capacity to create Real Art composed of complex metaphors embodied in a form other than how they appear.
It’s easy to test. Show people “real” art and AI generated art and see how often they can tell which is which.
Maybe a GAN generating images from random seeds without conditioning would make uncopyrightable content.
Each person is individually.
True, as is seen in the "banana with duct tape".
Wait 5 years until the first android AI is being raised as a child and develops first hand embodied and social knowledge.
> From Middle English artificial (“man-made”) via Old French (modern French artificiel), from Latin artificiālis from artificium (“skill”), from artifex, from ars (“skill”), and -fex, from facere (“to make”).
Regarding the artificial baby - I was thinking along the lines of putting the Gato [1] model in an android body. Nothing wild technologically, just give it the experience of having a family, friends, things to do, like we do for kids. This will induce the agent to form the same kind of value judgements as people make. It won't be just trained on the whole internet, but have access to the real world + human society, the best training dataset.
I believe the lack of access to the real world, the human society and a humanoid body is what keeps AI models from being more human-like. We can't become a developed human without society, for example. Why would AI be able to? The next step in AI is not about the neural net architecture but the curriculum, especially real world exposure.
[1] https://www.deepmind.com/publications/a-generalist-agent
in contrast, the whole point of paiting or a sculping is the final completed artifact.
What I'm trying to get at, is how recording technology "defeats" the purpose of music as its performace; and so, irremediably splits it from it's source (its origin, the composition of the music). I suppose improvisation remains; but highly constrained within prearranged structures (a la "AB[improvisatino goes HERE]C") like in jazz.
I do not understand how hyperpop nor DnB are supposed to be a response to what I'm trying to get at (then again, I'm just now barely conveying my point... I hope.)
That's not what people mean when they talk about AI art though. AI art is Commercial Art. What Horkheimer and Adorno famously called "The Culture Industry".
I'd love to see this technology aimed at concept art for games. I don't think this would hurt employed concept artists, I do think it will be a tool concept artists can use to quickly thumbnail significantly more versions of a vision.
That artist will then refine the results into something supremely engaging to humans.
Really I see this as an incredible tool for artists to help refine truly innovative ideas.
It freed artists of having to focus on realism, if they wanted to do so.
I can see how it's semi-easy to have a database of images up to a certain date (before AI generating algo existed), but what about if you want to train the AI with modern data?
For example, say if I want to query, in 2035: "Flying hippos in the style of McGrundsBulle" where McGrundsBulle is an artist who only started producing art in 2025... How is the AI then not going to trained with "flying hippos" generated between now and 2035 by AI like DALL-E 2 / midjourney / etc.?
I'd argue it wouldn't make sense to train better AIs using current AI's generated pictures for many of them are the stuff of nightmares and it doesn't seem to be because they've been trained by nightmarish datasets. They're just creepy due to how the current AI works and many details are obviously off, which makes them very creepy to humans.
[0]: https://github.com/LAION-AI/laion-datasets/blob/main/laion-a...
The reason human painting will survive is very simple: because 1) people value human-made objects independent of those objects' objective qualities, and because 2) paintings exist outside the digital world, and there is no robot sufficiently dexterous to match the skill of a painter in applying paint to a canvas. That being said, as this technology matures, lots of currently employed digital artists will no longer be needed.
ok but that's not what these models are doing? they're mimicking training data, and sure it's really impressive how far that gets - but ask it to do something outside of that raining data or where the # of samples is low and it's "creativity" falls apart
I think these models exhibit a form of creativity, but not creativity in the vein of say - an art style not seen in the training data
There are very few artists who you can call “original”, the majority are riffing and remixing based on previous predecessors. Making find tweaks here and there.
This is something you can achieve with an AI model just by increasing randomness.
That's actually the most interesting cases. And I strongly disagree that it falls apart
'The Edge of the Universe'
https://www.reddit.com/r/midjourney/comments/wmu10p/the_edge...
Citation needed. We know how all ML starts with an RNG, since we built it. We don't know how human mind works.
It's debatable--it goes back to Barthes's "death of the author", whether art's value and meaning resides solely in viewer perception or also in author intention--whether AI art will be "true" art, but it will very soon be good enough to replace artists who work in subordinate contexts (which is, unfortunately, generally a necessity if one wants to survive within the reigning corporate-capitalist totalitarianism).
AI is not going to force us to "become artists and poets", because it will make it even harder to earn money (i.e., the stuff most people have to earn through humiliating subordination) at these things. It will force us to become dependent, in the way artists (in practice) already are. This isn't as horrible as it sounds. The rich are already dependent on the state, which enforces their so-called "property rights"--this fact being something so-called "libertarians" are too stupid to realize. The middle classes are already dependent on employers, which turns out to suck a lot more than being a rich person dependent on the state. We already depend on complicated supply chains and national defense practices; it is simply the case that we'll soon (a) be dependent on AI, as it increasingly excels at the menial subordinate work we once relied on other people to do, and (b) be dependent on intelligent management (first coming from nation-states and UBI; later, one hopes, coming from a global entity that can eradicate poverty worldwide) of this weird quantity called "money", because we will no longer be able to reliably earn it through work, machines having taken that over (and thank God).
Without a complete overthrow of corporate capitalism, we're going to regress to the state of medieval Europe and many of the poorest, most corrupt countries today, in which there are people who have money and there are people who work but it is impossible to get meaningful money by working.
----
[1] Please note that there's a distinction between personal property and private property. You're not an asshole for wanting people not to take stuff you need for personal use, such as your car and your books and your toiletries. That's personal property, which is a different thing. If we take this reasonable idea, though, and run it up without attention to how it might behave at scale, though, we run the risk of enabling the kind of nightmarish private property that afflicts countries like the US, inevitably leading to extreme inequality, intractable political corruption, and metastatic governmental dysfunction.
Assuming that current models' problems with, say, details and specificity, will continue on is equivalent to looking at 19th century automobiles and scoffing that they'll ever largely replace horses. Or looking at "smartphones" from the 90's and determining that they'll never see much uptake from the average consumer.
AI image generation models are going to continue getting better and better and better, and the flaws we see now will get less common and less severe.
It's equivalent to looking at 19th Century automobiles and asserting that they'll be able to travel to other stars by the end of the decade.
That's not actually equivalent at all. Looking at improvements to automobiles would mean them going faster, being more reliable, etc. Space is obviously a different domain.
Your analogy is like if people were suggesting that AI image generation models were going to become advanced enough to start driving cars or folding your laundry.
The fact that a lot of money has been poured into unsuccessful emerging technologies by very intelligent people in the past should be a warning that it may be impossible to predict the development of our current emerging complex technologies like AI beyond a few years.
AI might as well soon reach a plateau -- or it might take off beyond most of us can imagine -- or something in between.
Now, if the original moon landings had already been reasonably priced, and there was a ton of super rare resources on the moon, different story.
In contrast, training machine learning models like this, while not exactly cheap, is clearly cheap-er by a few orders of magnitude. And there's obvious commercial application: near-instant art in an absurd variety of styles, that can be created by anyone, no actual art skill required. That means there's the incentive and necessary resources to fund further improvements, and at least at the moment, there doesn't seem to be an obvious technological 'wall' that we won't be able to get past (though of course the better it gets and the fewer flaws there are, the less room there is for further improvement).
I've been "successfully" creating AI art for a few months now, and although there are some staunch opponents (mostly on Reddit), the medium has been generally very well received by both the audiences and many artists in the established community.
Fantasy girls (modified SD model from NovelAI): https://www.reddit.com/gallery/wkd77a
The Rock in GTA: https://i.redd.it/9nq3xq2xpie91.png
Fake anime: https://www.reddit.com/gallery/wk53cf
Benedict Cumberbatch in different video games: https://i.redd.it/gg23dah9bih91.png
Four stages of Elmo (NSFW: somewhat horror-y): https://i.redd.it/2qgugkwww1h91.png
Design agencies should be trying to build / buy these tools now as fast as they can. Otherwise the AI agencies will eat them up.
Someone could build a machine that splatters paint like a Jackson Pollock, but that hasn’t devalued his art at all. Keyboard and synthesizers haven’t stopped people from playing real instruments.
I'm not so sure about that. Trumpet playing appears to be largely dead. There's nobody like Herb Alpert today.
There's been an explosion in virtuoso instrumentalists on YouTube from what I've observed. Jazz seems be having a big renaissance especially.
I've heard some modern trumpeters, but they weren't very good. The trumpet is a very difficult instrument to play well.
Guitars and keyboards are popular because of the wide range of sounds they can easily produce. Trumpets have a much more restricted set of sounds it can make, so I understand the reluctance to specialize on it.
But when done well, a trumpet sounds heavenly. Nothing can match its beauty.
Oddly, the main instigator turned out to be a "Pokemon in real life" fanartist who didn't notice he's already a professional plagiarist.
> So in that context, saying “horses stuck around when the automobile came” is true, but if you went up to a painter and said “hey, within your lifetime painting will see a 90% decline, stop being taught formally, disappear from daily life or awareness”.
The issue with this claim isn't automation replacing artists (though I don't think that will happen either due to Jevons' paradox) - it's just that AI generated images don't replace paintings because they aren't paintings! Print shops already exist and may have replaced you though.
> I’ve had a lot of struggles with this. I have a specific image in my head, I’m trying to prompt for it, and the AI just does not want to do it. The most trouble that I’ve had so far has been with trying to get a tavern running across the plain with chicken legs.
There's a general unfixable problem here, which is it's hard to be aligned with silly prompts without giving you silly output for "normal" prompts. That's also why they're complaining the model output has too safe composition - the developers are lucky they even got it to do that, it's better than random blobs of color and body parts like older models would generate.
But the picture they want probably is hiding somewhere in Midjourney's latent space; it's just a matter of finding a prompt that recreates it.
One way to do that could be to sketch the picture you want some other way and run it through a reverse image-to-text notebook like https://github.com/pharmapsychotic/clip-interrogator.
Another would be prompting it with your sketch so that you can get an image in its "house style" - which doesn't seem very appealing for most models, but Midjourney has a pretty strong one.
>The issue with this claim isn't automation replacing artists (though I don't think that will happen either due to Jevons' paradox) - it's just that AI generated images don't replace paintings because they aren't paintings!
It is conceivable, that a popular artist will become replicated using an automated method or a manual one. We see that happening all the time with music and art. The best artists are the ones who influenced others the most.
There were some deeply moving pieces that got generated from some of my input prompts around neurodivergence. They weren't quite right, but still profound to me. After seeing them, I've started looking for somebody I could commission to use them as a starting point to create a real art piece, something I've never even considered before. I would approach said artist and say here is this low res image, here are some things I do and do not like about it, here's what it means to me, and here's some money. With this, I'm now engaged as a creator, curator ,collaborator, and patron in a world I'd never considered participating in.
Not only do I think this creates economic opportunity for artists who would like to work in such a capacity, it also gives way to new ideas for those who would react and reject this paradigm. Technology always has had a way of breathing new life into human endeavors. Sure, horses are no longer the primary drivers of transportation and agriculture, but equestrianism is a $100 billion industry in the US, growing 5% a year with an increasingly diverse pool of participation. The cruel and commodified low end of the sector is long gone, but it remains as a passionate and vibrant field. I expect the same from AI in all the fields it touches.
AI has a place as a tool to produce content in a fast and cheap way. And yes, as a result of that certain jobs will likely disappear. But art will continue to exist and great artists will still be followed and admired by the public.
Not that art should necessarily be a clear depiction of real things, far from that. Good art (read: art I feel a connection with and inspired by) is the presentation of a SUBJECTIVE POINT OF VIEW. In these images, all I see is the averaging of many points of view (the actual artworks the model was trained on, I presume), resulting in no point of view at all.
This technology will take jobs like designing 3-star hotel lobby wallpaper, not the places of human artists. Without a human point of view, "art" like this is flavorless jello.
In short, it's a mistake to judge AI art by its current capability. It's about to go on an exponential tour.
While I'm still rambling on about this – the author states: “In a strictly mechanical sense, yes, an artist can still create, knowing that the art could be done faster and better by AI.” The art cannot necessarily be done better by an AI, and the speed with which art is created does not matter, artistically.
It's been going on one for the last five years. Two years ago it would have been difficult to generate a picture that's recognizably anything at all, using any non-specific engine.
Now we're down to complaints that mostly don't matter that much. Two years from now..?
The Singularity is ever near because exponential growth!
I absolutely dont see Jukebox as "taking over/replacing" music... like all ML/AI models... it is yet another tool. "Learning to use Jukebox" (or DALL-E or StableDiffusion) is like "learning to play the guitar/learning to paint"... no need to be afraid.
Still, OpenAI has plenty of funding, so it's either they don't see much value in developing it further, or they can't improve it. I'm surprised that no one else does either. For example, with GPT models every big AI lab started to compete trying to outdo each other. Same is happening now with DALL-E. But no one has tried to train a Jukebox successor.
I guess this, in art terms, is the post-scarcity society. Now that's arrived, it's a big problem? Hmm.
These AI algorithms only regurgitate new permutations of what they are trained on.
There will always be the need for the pipeline of new material, but they're definitely won't be as large of a need for commodity artists who were already just regurgitating existing material in the first place.
These art AIs are very good at driving the point home that our DNN AIs are seriously capable. But those capabilities need to be nursed by ever-growing armies of people. True, maybe there is a limit to the amount of people needed to profit from coloring pixels in images, and maybe, just maybe, that ceiling is going to go down the more these artist AIs are used. But AI has the potential to break open areas we can't even imagine now. I for example have trouble imagining that in the time of the horses, there was an industry for traffic signals, another for transporting horse fuel between countries (but maybe there was?), another for teaching teenagers to drive the carriages and get their horse-driving licenses, and so on and so on. Although I don't know much about those days, maybe it was popular to be young and to horse around on top of a good steed.
Maybe we will be teaching our kids about how to survive in the jungle of wild AIs we have created for them, and teachers will be needed. Maybe we will hemorrhage AI-produced content, and in twenty years, there will be an industry for certifying that something is human-made (maybe a lesson to learn about how chess survived in the age of incredibly capable chess-playing computer programs?). Or maybe something else we can't quite see yet.
Just because there isn't a perfect 1:1 copy of the data in their released software, doesn't mean we should ignore the fact that this data--which they do not have rights to--is critical for the development of their software.
Many people are conflating this issue with the copyright of the software output (i.e., "you can't copyright a style"). But many artists are rightfully angry that their creative products, which they own the rights to, are being used as a critical component in the development of someone else's software and they have no say in it. They also have no recourse to get their data removed from the development chain.
This is not a new problem, of course. But with the image generators it's become more explicit about how the output relates to the input: "Give me a painting of a clock by Salvador Dali".
I think in terms of data usage rights, we need to consider the inputs of these models and not just the outputs. Data laundering through AI is only going to get worse.
I'm curious whether thats just the text descriptions people are putting into the generators to get clicks or inherent in the AI art generation techniques used.
I'll also throw out that for most of the history of art - "art" has always been cheap to produce. My pre-schooler created "art". Its really only the outliers that we remember. And really the ones who make it do it through verbal communication of their ideas or have some mechanism for conveying a consistent, overarching theme/narrative. [1]
A professional AI "artist" might need to figure out how to become more of a visual creator/curator - ushering the generated images into coherent story that can be shared via discussions at gallery showings, etc.
Perhaps real professional visual artists need to find the equivalent of a live show that has kept music viable as a career or a live book reading that makes an author relatable.
[1] https://freakonomics.com/podcast-tag/the-hidden-side-of-the-...
Currently most generated images seem to need significant touchups to be a good image.
For future generators to meet the specificity requirement they will need the language to describe it.
But if you want precisely a thing. Then using a generator is unlikely to give it to you. As a tool for inspiration or for supplying foundational material to tweak from there, it could be brilliant.
But I don't think it will fully automatically supply end game content for those with specificity in mind.
The author gives the example of needing dozens of images to accompany scenes in his book. If you feed DALL-E something like: "create a panoramic view of a grey ocean with four mountains and an ominous dark sky", then sure, it can handle that prompt fine. It'll do a great job at broadly interpreting adjectives and phrases that are unambiguous. Mass produced art for advertising/media is going to change drastically because it doesn't need precision.
However, once you introduce specificity and detail the tools are going to suck. For example, telling it to generate a portrait of a specific person with subtlety in expression, body language, pose, and demeanor. It'll create something that looks aesthetically sane but not anywhere near what you had in mind.
And just for some extra counter-predictions of my own:
- The novelty of seeing thousands of "good enough" images in mass media is going to wear off
- AI generated images will become obvious to the average person
- Society will largely get tired of AI art due to the lack of novelty
- The utter collapse of the art profession some are predicting is never going to come
But that said, the economics are both obvious and uninteresting. It’s the other stuff that I’ve seen most artists talking about, and what I want to at least touch on here.
I may be biased by the fact that my fucking job is the one this shit is coming for but I think the economics of this shit is very much worth talking about.
If I am a musician and I sample someone’s music, I have to pay them for the privilege. But this shit is evolving prompt books chock full of the names of working artists. Why don’t they have to pay (for example) James Gurney something every time someone types “fine art painting by James Gurney” into their prompt? Why don’t they have to work out licensing deals where Gurney can say “actually I do not want your black box to spit out anything based on my work at all, none of my work has ever been Creative Commons and I never gave you any kind of license to put any of my art into your database, you’ll be hearing from my lawyer soon”?
When videocassette recorders came out, the movie industry threw a fit over the prospect of home taping eating all their profits, and every blank tape you bought had a few pennies of its price go into their pockets to cover this. Maybe all of the money going into these AI tools should have a percentage diverted towards artists. How to spread that out is a problem, given that we are generally a bunch of freelance entities - maybe it’d be a good reason for me to finally join the Graphic Artist Guild if that was how that money got spread out. But if this shit is gonna eat all the cheap work then maybe a lot of its profit needs to go to subsidizing the people still trying to get to the point where they’re good enough for art to pay their bills.
All this shit is in a legal grey area, where the art it’s resampling and mashing up is assumed to be free to use and quite possibly may not be. This is something people have not understood for the entire life of the internet: just because an image is posted online does not mean it’s free for you to take and make money from. This shit may technically be “sufficiently transformative” under current copyright laws but I think there are some new copyright laws that need to be made here.
Also, if you are about to post a reply to me about the inevitability of progress, know that I hate you for it, and that I hope that we see a sequel to GitHub’s Copilot very very soon that makes you just as fearful for your job’s future as this one does mine.
That said by a certain point the kinks will be ironed out and likely skirt around such issues by only incorporating/manipulating just enough to be considered fair use and creative transformation.
[1] "The model can generate known entities including trademarked logos and copyrighted characters." https://github.com/openai/dalle-2-preview/blob/main/system-c...
If there's a source that says otherwise, I'd love to read about it.
As I said though those things will likely be ironed out to avoid this more in time.
So the nonlinear diffusion systems out there are told by prompt to make what in the art world is considered to be very rude.
Let's have a Butlerian Jihad.
You might be able to force OpenAI to hand over a part of dall-e's revenue, but how would that work with an open source model like stable diffusion that runs on consumer hardware? There's really no way to control how much/in what way its used. Even if you make it illegal to use James Gurney's work to train model without his explicit permission, people could just finetune it in private with essentially no way to determine whether they used works they didn't have permission for.
The closest thing to the cassette recorder solution would be to put a tax on every computing sold, but that seems unlikely to me to be palatable to the general public.
We definitely have very interesting times ahead of us.
Which attempts to address some of the stuff wasn't addressed in the original article.
(Also totally unrelated: as a software dev I would be scared to lose my job but also unbelievably excited for "an automatic programmer")
We've seen it in manufacturing of goods now we will see it in the manufacturing of art. Those in the lower end of the pay scale need to start looking at ways to adapt to the new reality.
The market will start looking a lot like the actor's market where a few get paid very well, others get paid a little and most, by far, do it for free since they love doing it.
I think it's possible for people to adapt by creating projects that use AI. They can use their creativity on the project rather than the art itself. One of the interesting side affects of AI art is that you can generate ideas with it.
Here someone requested images of cups in the style of an architect. It would be great to have some of these in my house.
https://www.reddit.com/r/dalle2/comments/w7buyx/a_coffee_cup...
AI art will have an impact on society as a whole.
I agree in relation to both non-digital visual arts (kind of) and digital visual arts (especially). This may also apply to CAD, game design, ui/ux and 3D animation too. So it is a big deal. However, there's no real AI replacement metric if your art is outside of these, very digitally orientated categories. A big shark in formaldehyde, psychedelic peeing reindeer, silver bromide developing on a metal canvas, titanium pellets that roll through a game machine etc (e.g. all recent artworks). They're all art, mostly good (at least to many people in the art market or critics), and AI is in not really relevant in terms of labour replacement.
Namely, if you've shown this stuff off to "normal" people, you've probably noticed that their reactions are kind of mixed, most being fairly dismissive. Some might even be confused because they just sort of thought that this is how art was made on a computer in the first place. Personally, I don't think a hypothetical DALL-E 3 or even DALL-E 4 will change their minds. At a certain point a sophisticated AI art tool becomes indistinguishable from Google Search.
Indeed, part of me wonders if where this is all heading. Google Search for things that don't actually exist. Infinite Pinterest. Indeed, I think this is the most serious scenario.
I mean, think about it. Imagine what it would be like if Google search rolled out a new search engine where you paid to search for images, but you could use whatever images you found for whatever. Search for a dragon sitting on a pile of gold and you'll find a heap of dragons sitting on piles of gold in any style you could imagine which you. Download it and use it in whatever you want. Its yours now (or so they claim, you can't help but wonder if its too good to be true). That really is essentially what AI art is.
And when you put it that way, it sounds pretty underwhelming. If you're a creative director /marketer on a budget, its infinite stock photos at a price just slightly cheaper than what you were paying anyway. If you're a creative on a budget who can't draw good, it's a good-enough source of creative imagery (so many damn NFTs). If you're a small business owner, you might be able to find a logo you like after enough digging, though you might feel that you're limited by your ability to ask for what you want.
At the end of the day though, it amazes me how the smartest people in the room consistently fail to comprehend how unreliable spoken and written language are as a form of communication. It's a very slow way of communicating versus pictures and imagery. Honestly, this whole infatuation with AI makes me think that these people are trying to invent yesterday's future, the future of the 60s, one where you could talk to an "artificial brain" but the idea of a computer as a creative canvas was absolutely absurd. It feels like a strange step back. It feels very backwards to suggest that "prompt engineer" will be a future career.
I think that's why the real future of AI art probably isn't machines that make the art for you, but "magic paintbrushes".
Technology both displaces jobs and enhances jobs. It's impossible to tell a priori the economic effect.
It's just as possible that artists leveraging these tools are paid more because they're incredibly more productive than the overall demand for art somehow being sated and reducing artist employment
The industrial revolution had a huge impact on society, discussing that and how it will displace or transform some occupations isn't the same thing as actually fighting back against the new technology.
> The AI can make art. I’ve seen people dispute this, since what the AI makes by definition may not be art, under at least certain definitions. I think those definitions are stupid though, and will elect to ignore them.
is in the article itself:
> Art, at its core, is an attempt at expression and evocation. It’s a way of talking about things that are hard to put into words, a way of making arguments that don’t have their own formal logic, a way of connecting with others, of sharing. Some people paint or write because it’s a way of letting something out of themselves, of working through thoughts or trauma, or spending some time feeling some particular emotion or working on a thought...
> Art is a communicative act. It’s a conversation. You see a picture and it makes you feel a certain way, and yes, sometimes you silently process that art, but most of my favorite aspects of art as discussing it with other people, wrestling with the art in public, teasing out what it’s trying to do, or what it’s doing without trying. I generally think that this is one of the best parts of being an author or an artist, this very public back and forth, sometimes with the art having to defend itself as the critic shadow-boxes...
Sooner or later, a general artificial intelligence will create art and it will be a crisis for human artists. That's not today. DALL-E is not trying to make an argument, work through some trauma, or share a particular emotion, when it turns your text query into a feature vector into a JPG...
I recently attended a talk by Aaron Hertzmann who compared the rise of AI-assisted art to the invention of the camera. Yes, some types of art will stop being commercially viable---stock images of D&D characters, to take the article's example---just as photorealistic portraits stopped being viable in the age of photography. But art as a social act of communication is alive and well...
So that's new demand for art created by it being so accessible. Most of the art has some very obvious flaws and deficits; much of which could be cleaned up quickly by a human. I certainly imagine new jobs for artists that use this tool to assist in very quick jobs, for applications higher value than my use case but which still wouldn't exist if a human had to do all of the work.
I don't have much of a prediction except that it seems a long long ways from there not being value a human can by working on prompts and cleaning up images.
For sure there's already applications where you don't need a human involved per image. But for applications where it makes sense to spend 5, 10, 60 minutes of a human's time to improve the output, I think the baseline output could improve drastically and as long as humans can still make improvements within a timeframe that makes sense for the application, it will still make sense to do so.
I can almost hear the "would be a shame if something happened to it."
Imagine you built a house, and someone came a long and said "I bet you feel like you own that house."
Insanity.
Please, go ahead an generate an AI image to advertise your videogame or book. Take out as much of the human creative process as you can until we're left with a world of algorithmically formulated 'art'. I've got a moron friend that keeps showing me images he's made using an AI, at first glance they look interesting but after a couple of seconds you realise it's one dimensional and hasn't impacted you in the same way actual artwork does. There's a nuance to creating breathtaking art, and a lot of that comes from ideas and emotions far more complex than algorithmically smashing pixels together.
That's probably just your knee-jerk reaction. The same that people express when they are told they are eating a vegan food. The right question to ask yourself is: how confident am I that I would be able to distinguish between AI and human art if I didn't know how it was created? You seem to be pretty confident that you can tell them apart, but I call that bs.
Are you for real? I can confidently say that you could. If I put a generated image next to an actual piece of artwork I'm confident even you could very easily distinguish between the two. I mean, you making it out like it's in some way difficult which is confusing. I can pick up whether an image is generated using Dall-E within a few seconds, much like most people I know.
Not sure what you mean with that vegan nonsense, but okay haha.
On one hand the business has allowed artists to improve and grow by allowing them to spend their time on their art (instead of working on something else for at least 8 hours per day so they could eat, producing less art and of less quality)
On the other hand, the fact that artists depends on their art doing well finantially in order to sustain themselves must have conditioned their output. The link between "art that sells well" and "art that is actually good" is ... unproven, at best.
This could be an apocalypse, but it could be a good thing as well. Perhaps we do need art that isn't "businessy".
All of the classical works of art people consider "actually good" were done under patronage. The Sistene Chapel ceiling wasn't painted for love or passion or charity.
Art and commerce have always been linked, for the simple reason that art takes time, effort and resources which need to be compensated for, so the artist can eat, afford housing or put food on the table. The "starving artist" archetype (and related belief that artists need to suffer to keep their art pure) is a romantic, exploitative fantasy, like that of the "mad genius" linking mental and emotional illness to creativity and insight, and unfortunately common. When AI puts all of the amateur and working artists out of work, they're not going to decide en masse to just stick it out for free, they're just going to quit. You're not going to get more, better art, you're going to get less of any kind of art that isn't machine-generated, because that's how the markets will optimize.
After everybody gets bored and numb from all these fantasy pictures, artists will discover how to make art from latent spaces. What's encoded there is an input of human history. Many new things will arise by tweaking and distorting that history.
Many new effects are discovered by playing with the machine. Things we cannot yet imagine, but one day a new style will pop out.
As artists we work with the past and we work with our tools. With AI we take that to a new level.
Copycat art is boring, but I'm sure we will see a lot more of it for a while.
I really enjoy generating images, and I really enjoy drawing even though I'm not really good at it. I like to draw with my five-year-old, but I've also exposed him to Dalle-Mini. I think showing him dalle-mini was a mistake, and I'm going to avoid repeating it. (We are a low-media household.)
My concern about AI art has to do with children learning art skills. Will it be hard to encourage developing art skills? If a person can just ask for an image rather than learning how to make one, will they be motivated to do the work to build the skill? Or will it suddenly become a waste of time to the young mind?
I could see a difference in the case of wanting to see a custom image and being able to summon it via description vs. needing to learn to produce it. I’m not particularly artistic and never developed any such skills… so I don’t know: how crucial is that particular motivation to those who acquire art (production) skills?
If you just want to look at an image of a thing, finding one is good. You might even find a real picture of something. ( In the future, this isn't a given. ) Plus you'll have to filter out images that are irrelevant, or even harmful.
If you need to use the picture for something, there's a lot of advantage to crafting your own picture. (licensing, privacy, attribution, ownership, et al)
When I wrote my comment I was trying to think of a parallel from the past, and I kept thinking of calculators. An even better example would be the decline of hand writing as a skill.
The same is basically true for drawing, but you're doing more of the early parts of the process more often. To draw a figure from imagination you have to learn the shapes. To do that you have to find a real figure in pose, and carefully copy it to get the proportions of the shapes as accurate as you can(which is assisted by various measuring techniques to correct your work). Then, after a few thousand of those, you try to draw an imagined figure and voila, your hand makes recognizable shapes - maybe a bit simplified, out of proportion, or with poor anatomy, but something others can parse correctly.
And so doing a lot of drawing is like building up a vocabulary, becoming more observant and aware of subtle visual differences and therefore more able to communicate them back. I can't see AI replacing this: it can be a good servant and give a lot of options, but it can't mind-read that you want a hand posed slightly differently.
Neither of them create paintings though. So your kids are safe to learn that.
Taking something previously rare and valuable and making it cheap and available and showering it upon people like rolls of toilet paper doesn't bring happiness imo. It reminds me of this story by Ryunosuke Akutagawa I once read, titled 'Yam Gruel'. Basically a peasant looks forward to a yearly bowl of yam gruel which to him is precious and rare. But as a cruel joke he is invited to a dinner by a noble expressly to be served unlimited supply of yam gruel, and from then on he can never enjoy it again.
From what I've seen so far, AI "art" sucks. And I'm not talking about fake/real. It's just crap, as far as my opinion is concerned.
You sit together with people and talk about a design, you can quickly draw something and talk about that. It's an incredibly fast tool in an artistic process, you will not replicate that with text prompts.
[1] https://roalddahl.fandom.com/wiki/The_Great_Automatic_Gramma...
This is the singularity. The typical definition of the singularity involves "unpredictability," but that is just a symptom. The cause of the singularity is AIs generating new information faster than we can consume or act on it.
Of course in art there is no point in exceeding the rate of human consumption; as others have pointed out here we face a future in which AI continuously generates art tailored to each individual. As a result, this doesn't quite rise to the level of "unpredictability" envisioned by singulatarians.
All fields will eventually face the same reckoning as this author has spelled out for art, and some (such as circuit or AI design) will introduce positive feedback loops that will not only put the relevant engineers out of work, but also cause the field to advance so rapidly that we humans will have no chance of catching up, let alone predicting what the AIs will develop next.
Largely agree with AW the OP, but think the question of "what art are we talking about" needs more air.
A lot of the turmoil around these tools appears to be about the threat they pose to the careers of artists and designers (and soon, most related domains). But some of the simple versions of this story (not this one) overlook the nuances IMO.
Some of those being,
- the distinction between "fine" art in the commercial "art world" gallerist sense, and graphic design/illustration/commercial art - that professional commercial artists are going to adopt and exploit these tools with the same intrinsic advantages they have using with other tools - that in a very short time there is already an explosion of work that uses these tools as one (large) element to produce synthetic work across media that would have been prohibitively expensive/time consuming without them
Even as a bystander, but with a background and interest as a serious and modestly successful artist in a different medium,
it's been remarkable to see just how quickly people with fewer kids, better ideas, and more time than me, seized upon these tools not as simple image-making widgets, but as sophisticated tools to exploit to generate a stream of imagery to be applied to create Other Things, both as proofs of concept, commercially, and as yes "real art" in the gallerist sense.
These things are quite obviously disruptive in a way most VC only dream of.
Idle other comments:
The "eerie" uncanny sense that scaling up the visual cortex abstraction stack toward "grandmother neurons" that these systems has, is IMO considerably more unsettling than the impact on specific industries. What we can see (literally) in these systems is a visualization (in the data viz sense) of how much further along towards AGI these things may soon be.
Their failure modes are more interesting to me than their successes. The ramifications for how they fail and how they are of course unaware of their own failure, is a true cautionary tale.
IMO those failures offer an excellent basis for "cautionary tales" and illustration to the general public of why we need oversight and governance for the deployment of these systems. It's one thing for them to produce "nightmare fuel" when asked to create kittens in the kitchen; it's another for a comparable (smaller less tuned closed-source closed-training-dataset...) system to be put in service approving mortgages, assigning credit risks, or (as made ProPublica famous) assessing whether someone is a candidate for bail.
(Highly recommended: The Alignment Problem)
Anyway. The singularity is already here, it's just unevenly distributed.
If AI were suddenly beginning to write and produce whole songs I think that would be a closer comparison to what's occurring here.
Very interesting comments thread (as usual here).
I'll hijack this post to promote this event in which I'm involved:
https://www.eventbrite.com.au/e/creative-ai-sydney-tickets-3...
We will be discussing this 'issue' and address many of the questions raised with academic researchers and leading industry creatives from Canva (who have been disrupting the graphic design industry for a while already), NVidia AI (who are the n#1 hardware engine behind this revolution), as well as Spotify, Warner music, Google Creative labs and others.
The event is free, online or F2F in Sydney!
#CreativeAI_Sydney
It's always been about the artist. And I don't see that changing.
This does have the potential to affect a graphic designer at a startup because they are, in some sense, more fungible. And even that's a big "maybe" to me at this point. At least, in the short team, I see them using the new wave for AI generators as a co-pilot rather than an auto-pilot.
IMO this really undermines your argument.
Creating novel assets for games is crying for automation. Capitalism had reached its limit in optimizing the efficiency in that front, that most 3A games nowadays feels like unfinished. They already hired an army of artists but still the production isn't up to speed.
Some automation like DallE-2 will surely turns up the speed. I think some shuffling of work force is inevitable, but art is about idea and expression, a model can realize it but it can't come up with cohesive one on its own.
So art will change, but I think it will be changed for the better
This is because access to music has become cheap. Discovery of music has become cheap. Signalling behaviours though rely on the costliness of signalling through a particular medium. Costly signalling is the theoretical framework you want here to understand what is going to happen to art if the tech is as good as it is currently being hyped.
We like to think of art as this inherently communicative act - as the author says. But the main psychological motivation is signalling. We would not waste so much energy as a species on such behaviour if it didn't have some kind of evolutionary benefit. So I expect much of the energy that goes into the artistic signalling medium will be redirected toward more costly mediums.
Nowadays there's more music being created and consumed than ever, but the musicians don't make nearly as much money as in the golden days of the CD. The mega-stars like Billy Eillish still make a good amount through brand deals / live shows, but the 'middle class' of musicians has fallen away.
I think lower barriers to entry through creating art via AI assistance will mean more art available, but it probably will make it harder to make a living as a professional artist. That said, I don't think anyone who chooses to be an artist today is primarily motivated to earn money.
I’m not sure where you got the impression that the bygone era of the CD was so lucrative, or that its largess was more evenly distributed in a meaningful way. If you’re open to challenging that conception, I’d highly recommend reading Steve Albini’s article, The Problem With Music (also sometimes titled Some Of Your Friends Are Already This Fucked; not linked because it doesn’t have a canonical link and I’m never sure which site reposting it can be relied on not to edit it). Anyway the tl;dr is this: to the extent there’s ever been a more even distribution of earnings for musicians, that certainly wasn’t attributable to CD sales. The article was written fairly early in the CD era, and those dynamics deepened through that period. I’m doubtful it was ever attributable to any particular sale of recordings on any kind of media. Then, as now (and probably since well before recording technology was available) musicians making a decent or better living by and large made the majority of it performing live.
For what it’s worth, the article I mentioned factored greatly into my decision not to pursue a music career, even as I embraced the early availability of cheap digital recording (and intermittently still do). I’m sure there’s an alternate universe where I’d be thrilled that my life’s work would be playing music, but I have little doubt that my entry into the middle class would have been far less likely than by taking up programming.
The prior technology was Boomers' universal counterculture, where if you wanted to do one of being a hippie or join the new age or do drugs or go to concerts or have weird sex you had to do them all at once - that's why they seemed convinced that going to an outdoor concert was literally the same thing as saving the planet.
Subcultures developed after people realized that didn't make any sense, and there was just enough communication technology to find people to get into your hobby, but not enough to mix and match them, so now your hobby life was often still defined by being a goth or punk or playing D&D, but it's less universal.
The kids are capable of joining more than one group at once because they can do it all online. And nobody is in a band anymore.
What difference does it make tho?
Not entirely sure of this. "What's your bag" was a popular phrase.
https://english.stackexchange.com/questions/294306/what-does...
I don’t think you know a lot of artists. Artists, like the actual ones, do Art to create meaning and participate in the meaning making process.
Assuming they do it for external signaling and the sort of implied “hey look at me” desire is off base.
is just
> for external signaling and the sort of implied “hey look at me” desire
in a different light. You cannot participate in the meaning making process without signalling to other meaning makers, attracting their attention, and influencing their subsequent signalling. If you don't signal, then you haven't participated because you just didn't try. And if you don't attract attention, then you haven't participated because you had no effect on the proceedings.
This is not to say that it's your interpretation versus their interpretation. It's to say it's both at the same time.
> is just
> > for external signaling and the sort of implied “hey look at me” desire
OP didn’t specify “create meaning” for anyone else, or otherwise conveying art to others. One of the weird things about art is that it’s a form of expression that doesn’t necessarily have an audience. Quite a lot of art is more akin to keeping a diary, or singing in your head, than “look at me”.
> You cannot participate in the meaning making process without signalling to other meaning makers
I have ample evidence to the contrary which I have no desire to share with you, or anyone, other than noting its value to me having created it.
Right, and they make the art, and others pay huge premiums to collect art to speculate and signal.
'High art' (the kind that hangs in galleries) has a meaning attached to it that escapes into the level of a meme and that often has little to do with what is represented to the senses, which is easily replicated. For example, the Mona Lisa is now famous because of an attempted theft in the early 20th Century. Other art is meaningful because it was at the vanguard of a particularly novel aesthetic, associated with a specific artist (very common), or because it encodes a commentary on a particular aspect of history (Picasso is a good example of all three). This kind of art will endure because AI cannot copy what makes it "valuable", though it may still be of assistance to the 'high artist'.
The same is true for music. Kendrick Lamar isn't famous because he's exceptional technically (some enthusiasts I've spoken with find him fairly mediocre in this regard). He's famous because of what his music communicates, because of its (apparent) authenticity, and how that resonates with listeners. That plus Kendrick's story is what gives his music signaling value, which is then amplified into a meme. Most people on the other side of the meme adoption curve probably don't even understand why they should care about Kendrick or his music. Rather, they're imagining they like it in order to follow the trajectory plotted by the tastemakers. I strongly suspect all 'high art' is like this.
A perhaps more germane example of the above phenomena arises in NFTs. NFTs of weird anthropomorphized monkeys are valuable because of a collective hallucination that they in fact have value. You can use AI to generate the monkey JPEGs, but you cannot so readily use it to craft the meme.
'Low art', by contrast, is much more utilitarian and often devised in response to fairly prosaic demands. Technical sophistication usually wins out. This is what AI will supplant. It's a problem insofar as many employed artists were relying on low art to sustain themselves. Perhaps art will now become even more so the province of the wealthy?
Personally I've always been more a 'low art' person since 'high art' seems to be about participating in mutual delusion and even snobbery, and its surface characteristics may not evoke anything profound when divorced from the subtext with which the art has been furnished. Incidentally, I think this is partly why 'nerds' gravitate towards things like fantasy + sci-fi literature, anime, and video games. These are artistic mediums that are very directly evocative and wherein the meaning and the superficial/sensory experience are rather closely aligned. And, of course, they prioritize technical sophistication.
Anyone who's a fan of 'low art' should probably be very excited because we're fast approaching a time where you'll be able to experience a near infinite abundance of 'JIT art' that caters precisely to your tastes.
Anyway it's an empirical question that we might get to see the answer to - so no need to argue about it really. If artistic creation is disrupted as some predict, we'll get to see the impact on artists.
Everyone else engaging with art will take what the algorithm(influenced by monetary deals/cuts/middlemen as ever in history) puts in front of them. Which will be the "hey look at me" people.
The friction that endows signalling with its cost is simply opportunity cost — you can’t be a vocal fan of everything or go deep on every TikTok trend-strand.
Also note that “discovery” here has been almost entirely mediated by TikTok or Spotify algorithms. If you’re tired of what those algorithms offer, in practice, it’s outlandishly difficult to do better than random walks to find something new you like….except through your friends’ signslling.
That all said, I think you might be right. But I submit this radically different reading is plausible. My own anecdata suggest this.
If you like niche genres there are plenty of catalogues though and the amount of wide known artists are not that big.
Is this the result of some study, or where is this observation coming from? I have almost certainly not noticed this, in fact I might almost say the opposite.
https://www.newyorker.com/magazine/2021/03/15/genre-is-disap...
Fandom/identification with specific artists remains as strong as ever though - you see a lot of "Swifties" or fans of a specific rapper, for instance, and they use it as a form of identity. You see fewer "metal heads" or "classical music erudites".
All anecdotal, of course. Would love to see studies on this.
But now anybody with a phone can dial up millions of tracks in a split second with Soundcloud, Spotify etc, including those KLF releases that were once unobtainium.
Ignore the whining of frontline managers about how "no one wants to work" (i.e., their own bosses are squeezing them by refusing to let them hire, forcing them to do more grunt work). The long-term trend of wages is downward. Labor markets do not find equilibrium. They diverge. The lizardfolk who own all the capital can wait. But laborers cannot; they need to eat. The whole goddamn system is rigged and 99+ percent of us are losers.
This isn't just about AI Art, or the mediocre but serviceable code generated by GPT-3 and the like. We are very likely headed for a broad-based, general wage collapse everywhere in the world, and this sort of situation is likely to result in a global violent conflict between people and capital, in which either an entire socioeconomic system is overthrown (a chaotic situation, but the good outcome) or humanity is thrown into slavery from which it is unlikely ever to recover.