The Worst Filing System Known To Humans

-Punk (5) A Song of Ice and Fire (2) Affect (9) Alienating My Audience (31) Animation (27) Anime (17) Anonymous (3) Anything Salvaged (15) Art Crit (41) Avatar the Last Airbender (2) Black Lives Matter (1) Bonus Article (1) Children's Media (6) Close Reading (90) Collaboration (1) comics (29) Cyborg Feminism (3) Deconstruction (10) Devin Townsend (2) Discworld (1) Evo Psych (1) Fandom Failstates (7) Fanfiction (28) Feminism (23) Fiction Experiments (13) Food (1) Fragments (11) Games (29) Geek Culture (28) Gender Shit (1) Getting Kicked Off Of TV Tropes For This One (11) Gnostic (6) Guest Posts (5) Guest: Ian McDevitt (2) Guest: Jon Grasseschi (3) Guest: Leslie the Sleepless Film Producer (1) Guest: Sara the Hot Librarian (2) Guest: Timebaum (1) Harry Potter (8) Harry Potter and the Methods of Rationality (3) Has DC Done Something Stupid Today (5) Hauntology (6) Homestuck (18) How Very Queer (35) hyperallthethings (10) hyperanimation (1) Hypercomics (10) I Didn't Ask For Your Life Story Sheesh (24) Illustrated (37) In The Shadow Of No Towers (1) It Just Keeps Tumblring Down Tumblring Down Tumblring Down (9) It's D&D (2) Judeo-Christian (9) Lady Gaga (5) Let's Read Theory (3) Lit Crit (19) Living In The Future Problems (11) Lord of the Rings (4) Mad Max (1) Madoka Magica (1) Magic The Gathering (4) Manos (2) Marvel Cinematic Universe (17) Marx My Words (15) Medium Specificity (15) Meme Hell (1) Metal (2) Movies (33) Music (26) Music Videos (21) NFTs (10) Object Oriented Ontology (4) Occupy Wall Street (3) Pacific Rim (2) Paradise Lost (2) Parafiction (6) Patreon Announcements (15) Phenomenology (4) Poetry (6) Pokemon (3) Politics and Taxes and People Grinding Axes (13) PONIES (9) Pop Art (6) Raising My Pageranks Through Porn (4) Reload The Canons! (7) Remixes (8) Review Compilations (6) Room For You Inside (2) Science Fiction Double Feature (30) Self-Referential Bullshit (23) Semiotics (2) Sense8 (4) Sociology (12) Spooky Stuff (41) Sports (1) Star Wars (6) Steven Universe (3) Surrealism (11) The Net Is Vast (36) Time (1) To Make An Apple Pie (4) Transhumanism (9) Twilight (4) Using This Thing To Explain That Thing (120) Video Response (2) Watchmen (3) Webcomics (2) Who Killed The World? (9)

Reload the Canons!

This series of articles is an attempt to play through The Canon of videogames: your Metroids, your Marios, your Zeldas, your Pokemons, that kind of thing.

Except I'm not playing the original games. Instead, I'm playing only remakes, remixes, and weird fan projects. This is the canon of games as seen through the eyes of fans, and I'm going to treat fan games as what they are: legitimate works of art in their own right that deserve our analysis and respect.

Sunday, July 24, 2022

AI Dreams of Genies: AI Art and the Fantasy of a World Without Workers

As AI advances, the hype machine advances faster. Its boosters and detractors agree: AI Art can grant all your wishes. This hyperbolic discourse hides the truth about our tech-mediated culture: we're being sold cheap knockoffs of our desires.



In the twitter replies to one of my earlier AI and Art articles, someone enthusiastically touted the ability of DALL-E to produce imaginary product designs. The product in question, a lumpy stein that sort of looked like R2 D2, underwhelmed me, and I pointed out it looked a lot worse than an artist could do. In an ironic reversal of the old complaint about modern art, the enthusiast responded, "yeah, but *I* can't do that."

A large part of the hype around AI art partakes of a dream of instant commodity satisfaction. Sure, your R2 D2 stein might be lumpy and weird looking but here it is, as if by magic! You might be banned from doing anything too political or sexual with these tools, but you can, at least for now, live the dream of commanding your favorite licensed characters to perform whatever stunts you want from the comfort of your home. AI Art popular on both social media and crypto marketplaces tends towards the same kind of off-model cheapness of knockoff toys, plastic crap or frozen treats shaped vaguely like "sanic" or "pokechu" or "red robotman". They come from nowhere (the "Third World") to fill the Dollar Tree in your neighborhood that just put all the bodegas in a 2 mile radius out of business.

In my very first crypto art article I described many contemporary artists, particularly crypto-curious ones, as producing art designed for social media, art that functions best when encountered briefly as part of a general experience of scrolling down a dash. I don't think that's necessarily a bad thing, and you might remember I talked glowingly of how someone like Simon Stålenhag employs that quality to provide a jolt of unfamiliarity with his moody rural cyberpunk landscapes. I wonder, though, can even good artists compete with the flood of knockoff toy AI art?

How often do we look at an image online and not see it, merely register it? As in: process it automatically, bureaucratically, stamping it with a "seen" label and putting it in the outbox? I click on an image on twitter to make it full screen, and I'm already thinking about something else, I'm a million miles away. I mentally nod my head. Check a box. Close the image. Scroll further. Then I realize I haven't seen the image at all, had a single thought about it. Is this just me? Maybe I really do have ADHD. I certainly shouldn't universalize! But also: isn't this the behavior that social media has cultivated through its UX regime? There's always something a few posts down that you could be missing, and nothing ever sits on its own but is accompanied by a whole stream of comments. I'll warrant a few seconds of "engagement" with 50 things is worth more than 50 seconds of "engagement" with one. I just made a tweet about the horny new anarch faction netrunner cards and it's got 111 "impressions", whatever that means! Golly wow! I don't think it's just my brain being baked by this shit, gang.

Because I am the kind of woman who spends much of her time languishing in despair about this and that, I'm not immune to fretting about AI art making my job obsolete. To some extent this is because of the actual instances of DALL-E 2, Looking Glass, and Midjourney pieces that actually prompted a response of some sort from me. When the work is good, it's shockingly engaging in a way that the glorified photoshop filters of Deepdream just never were for me. Good enough to prompt two articles already about the most interesting possibilities! In my darker moments, struggling to figure out what the hell is actually going on with the external oblique and tensor fascea... fasciea... with the fucking hip muscles, I look at the best of these works and think, oh why even bother honestly.

But what stresses me out the most is not the great stuff but the mid stuff. Not the Mid*journey* stuff, the mid-*tier* stuff. It's sort of bleak looking at a bunch of basically dodgy photobashes, the kind of thing you'd have seen in a throwaway contest or meme in 2009, treated like an amazing achievement not just for technology but for *art*. Does this sound familiar? In that first NFT article a year ago I had the same critique of cryptoart and its most prominent and successful examples. Sometimes I struggle to talk about AI Art because a lot of the stuff I could say--that arts writers and galleries have become tech boosters and stopped talking about the actual art, that the economic framework inevitably forces a race to the bottom of the content barrel, that wildly different usages with wildly different artistic validity are lumped together in an effort to legitimize the worst of the field riding the coattails of the best--well, I already said it about crypto art as a whole.

I get a weird deja vu comparing works by crypto art superstars like Beeple and some of the outputs of AI art. Any number of Beeple's images could be DALL-E products, and vice versa, from the often deeply ugly textural filters on down to the conceptual level. What is his picture of those two boxer guys as giants kissing in front of a crowd of lazily copied and pasted prefab assets if not a perfect prompt for DALL-E? The means are slightly different, but it shares the knockoff toy aesthetic--just close enough to the real thing for the end consumer to fill in the blanks, without having enough detail or care that it would interfere with the race to the bottom towards ever cheaper production.

I suppose the artist formerly known as DALL-E Mini does have this going for it, though: its take on the boxer kiss nonsense has at least a dynamism that the stiff, flat Beeple painting lacks!


Can artists who, bluntly, give a shit about their craft compete with the mass deskilling operation represented by this technology, and the algorithmically shortened attention span of mass culture online? Eh. I'm not all that optimistic. But AI assisted artists might just be able to at least turn some of the logic of the genie on its head in their work.

I'm fascinated by DALL-E's ability to produce slightly uncanny photographs of products that have never existed. The way it feeds into the fantasy of just having stuff appear magically without having to ever confront the process of its production or the people involved in its creation is totally rancid, don't get me wrong. But I've always been fascinated by alternate material history, and the production of forgery art. This kind of work, art that exists to play in the tension between the real and the fake, parafiction and parafanfiction, has the potential to provide that jolt of defamiliarization.

It's difficult for me to talk about this because I haven't seen anyone actually *do* this yet. I'm just looking at the kind of product photos people are generating and extrapolating possibilities based on other parafictional projects I've seen. Imagine being confronted with what seems like an anomalous object--say, I don't know, a Polly Pocket produced in the People's Soviet Republic of the Cascades, or a VHS of the full and finished film of The Thief and the Cobbler, the details aren't really important for the thought experiment. You notice the specific tells of the AI and are confronted with the object's unreality. I think it's possible to make the leap then, as a viewer, to a question of why this has to be a hoax rather than reality.

This is an object "from nowhere" by definition, but it implies an entire material history of production. In this sense it's the exact inverse of the knockoff toy: instead of a real object that obscures its source and the labor that produced it, it's a fake object that uncovers an impossible labor that might have been. AI art in fact offers unique potential for this kind of fabrication. You could produce everything from design documents to set photos to behind the scenes shots of animators... you name it. I suspect it would all need extensive editing but, well, that's sort of the point, isn't it? The AI isn't a genie, it's just another tool that can facilitate this sort of production.

I said it was probably unethical to obscure the actual production process going into AI images; here's one case where a *conditional* obscuring of that process might be valuable, for the way it can set up and facilitate the hoax, which is then revealed in order to call into question the assumptions we take for granted. Just like any speculative fiction, it can lazily reify the state of the world, or it can dare to imagine an alternative and confront the viewer with it. I am trying to look on the bright side here. I am squeezing these lemons I've been given so furiously you'd think I was a Juicero from a timeline where the Juicero actually fucking worked. (Potential DALL-E prompt??) I think what we're most likely to get out of AI art is: right wing propaganda, shit memes, and the continuing steady deskilling of visual artistic labor. But god can we just for one second, if we have to do this, at least pretend to care about making aesthetic and functional distinctions between the types of art made with the same general tools?



The shallow cheerleading for procedural art and its genie model of production, its imagined capacity to simply deliver, effortlessly, all our dreams to us exactly and unchallengingly as we conceive of them, is counterbalanced by equal and opposite shallow fearmongering about how it will lead to the end of civilization. None of it seems interested in asking questions about labor rights, deskilling as an economic phenomenon, the place of tool assisted art creation in the context of surrealist automatic drawing or dadaist readymades, a relationship to conceptual art, or the deep weird questions about the vibrant autonomy of the materials with which we compose art. No, instead you get a bunch of whinging on about the Human Soul, which always sends me straight to a mindset of please calm the hell down, you are literally talking meat. Get some perspective. Sheesh.

The two sides of the debate are perfect mirrors for each other, actually. Erik Hoel's bafflingly popular substack (🚩) post is a good and popular example of this. Hoel takes for granted the notion that all of the tech hype around AI is legitimate, not just a bunch of boosterism and marketing. He just happens to think it's terrible rather than super cool. Future generations, he bemoans, shaking and crying, will grow up in a world of only soulless mimickry of art. Walter Benjamin predicted it! (God these hacks love misquoting Walter Benjamin, it's like a compulsion.) His evidence? Well, he told a robot to paint a picture of "a robot with flowers growing out of the top", and it did.


Erm, sort of. By the top did he intend "the top of the head" or "the top of the torso"? If it's the former, I count 7/10 that generously fit the prompt, 5/10 that I'd say fit the prompt *well*. 2/10 seem to sort of be doing their own thing, flowers growing around and through the robot. One has a neck made of plants, which is pretty cool but again, not the prompt. If the claim is that it is "incredibly, gobsmackingly easy" to go from a prompt to a finished accurate result, I'm not convinced. That's sort of the trick, though, isn't it? Both the hypemen and the chicken littles have no compulsion to show you ALL the results they generate. I have no idea whether this sample of ten images is all he generated, or just the most representative. Both parties have a great incentive to, bluntly, lie about just how easy it is.

And they can basically get away with it because most people don't really pay much attention to images. The sloppy, half-hearted, mid results of most AI art programs can pass as the real thing primarily because no one's paying very much attention at all, to the fine details where the images tend to get runny and incoherent, or to the accurate fulfillment of the prompt, or (and this one's harder) to the negative possibility space AROUND the computer's image generation where other possible representations lurk. This isn't an indictment of AI art, it's an indictment of a visual culture in which most people are simply not visually literate.

You might think that a reasonable response to this would be promoting visual literacy. Erik follows--indeed, is a paid subscriber to!--some people on substack very, very worried about the state of our education. Is "Common Sense", one such subscription, working on this problem? Well, no, they're busy writing articles like "I Refuse To Stand By While My Students Are Indoctrinated" and "My University Sacrificed Ideas For Ideology. So Today I Quit." and "I Criticized BLM. Then I Was Fired." and "Elon Conquers The Twitterverse: Our chattering class claims Musk is a supervillain. The truth is simpler: He wants free speech. They don't."

Great stuff!

Look, comrades, I get that a critique of AI art that valorizes human creation feels tempting. Is this not similar to my own argument, that a lot of our visual culture is garbage? Well, no, not really actually. Because the difference is I actually care about art that I can talk about and analyze and that makes me feel something, not about some abstract criteria of whether whoever made it was enough of a Genius. That's fucking stupid. My criteria is ultimately simply this: I have something to say about a flaming piano, a fan art of a magical girl anime, a fox playing with a felt blanket, and Conway's Game of Life.

I have nothing to say about "a robot with flowers growing out of the top," and neither does Hoel.

He can't.

Because if he did say too much about it, if he drew too much attention to the actual art in a way that prompted his audience to really look at the images presented and think about them instead of scrolling past them on the way to his next bit of insipid Great Man worship, they might start to ask questions like, "wait a minute, can you ACTUALLY look at this selection of AI art and reconstruct what the prompt was?" 

I actually would hazard a guess that you can't. Oh, not the robot part, you could probably get something relatively close to that. But... I didn't actually tell you the WHOLE prompt that Erik found it so "incredibly, gobsmackingly easy" to write. No, this prompt specifically said the image should be in the style of a particular *artist*. Can you guess who it was? Perhaps the relatively homogenous treatment of color and the flat gradient backgrounds suggest someone like Edward Hopper. Some of the flowers are almost post-impressionistic, and there's an awful lot of yellow... could it be Van Gogh? Ah but then you look at image 9 and that seems almost more like the primitivism and fascination with masks of modernists like Fernand Leger... oh but the flowers are all wrong for that!

Did you guess Grant Wood?

Would you *ever* have guessed oh, yeah this is obviously the American Gothic guy?


I wouldn't have.

So is the genie really handing Erik his wished for image? Because as far as I can tell, Erik has not in fact easily generated "a painting by Grant Wood of a robot head with flowers growing out of the top", he has generated some images of robots with flowers growing on and around the upper parts of their body in what looks like a mid century illustrative style, but he has FIRST TOLD YOU that he generated "a painting by Grant Wood &c &c" and you, and I, and everyone else, has sorta gone, yeah alright I guess that looks kinda like that prompt, and kept scrolling. This stuff works primarily because we're grading AI art, and boosters and doomsayers alike, on a curve--something I'd expect a guy who is just so sincerely worried about "intellectual standards" to be a bit more mindful of. 

That's the thing, though. These "intellectual standards" never really involved letting the plebs in on the game. Art was supposed to be produced by an elite, necessarily limited number of geniuses. For the people who sincerely believed that belonging to the circle of geniuses was their birthright, I'm sure it does feel, as one of the stupid tweets he quotes bemoans, that the internet "laid waste" to music. What this really means is that all forms of art were rendered increasingly *proletarian* by digital technology, increasingly subject to the specific logic of labor. The genie--the Genius--AI is just the latest in a long line of practical and social developments, youtube on one hand and "postmodernism" on the other, that threaten the genie--the Genius--that is the soul of the artist.

Erik and his ilk buy into the hype because they are the other side of the coin to the tech boosters. They're in the same damn class, or at least *aspire* to be: the class that gets to ultimately control and dictate taste, art, legitimacy, and what is available on the market to us. But there is a legitimate conflict and contradiction inside the class between the hype men who seek to maintain their market position through mass deskilling and the commoditization of art, and those who seek to maintain their position by simply being the people who dictate what small pool of art is "legitimate". Hence the different stance on whether the Genie AI is a benevolent spirit or razing demon. They can agree on one thing, though: an educated working class that can discern for themselves what has content and what just *is* "content" is the enemy. They'd be so much harder to sell to!

It's time to dispel the smoke and mirrors they're using to conjure this spirit and talk about the real work of art--and that's what I'll be concluding this series with next time.

This Has Been

AI Dreams of Genies

105 people supported this article on Patreon. Will you join them?

If you liked this article, consider tipping me on Ko-Fi!


In The End We All Do What We Must: Universal Paperclips, Clicker Games, AI, and Agency

Remember Universal Paperclips, that clicker game? Remember turning the human race into paperclips? Ok, so, what if you just... didn't? What would that choice tell us about game design, agency, artificial intelligence, and people?

Let's Pop Together

What is it that makes today's Pop Team Epic's so different, so appealing? And can it explain why people hate the marketing for Ready Player One so damn much?

The NFT's Aura, or, Why Is NFT Art So Ugly?

NFT art is bad for the environment, and bad for artists, but critics and supporters of NFT art are both missing a key fact: it's also just bad art. Whether Beeple or Bugmeyer, it's time the stars of the NFT revolution experienced some real art criticism.

1 comment:

  1. One thing that helps temper my feelings on AI art is that my favorite artist is Imperial Boy, and all of my attempts to get an AI to imitate their fantastically complex and lively cityscapes come nowhere remotely close. The way AI art muddles the details becomes a crippling problem when you want something that's all ABOUT the details.

    ReplyDelete

Support on Patreon
Store
Reader's Guide
Tag Index
Homestuck Articles
Solarpunk Articles
Twitter
RSS Feed