> a building that looks like Holly Herndon

May 5, 2022 5:52 PM

Infinite Images and the latent camera Holly Herndon and Mat Dryhurst talk about their experiences with DALL-E (versions 1 and 2) and their thoughts on a future where anyone can make conceptual art from a few words. Holly and AI previously and previouslier.
posted by snortasprocket (19 comments total)

I'm just here to say Holly Herndon is great. Off to read the article.
posted by abulafa at 6:38 PM on May 5, 2022 [1 favorite]

John Henry was a steel drivin' man.
posted by surlyben at 6:40 PM on May 5, 2022 [1 favorite]

TIL that Holly Herndon is all about the cryptocurrencies, so thanks but noping out
posted by scruss at 7:31 PM on May 5, 2022 [6 favorites]

i used to think stuff is pretty neat but now i find it extremely off-putting

be interesting to see some sort of pepsi challenge for human vs AI but i doubt think the outcome would be anything other than depressing
posted by glonous keming at 8:12 PM on May 5, 2022 [1 favorite]

I'm about the farthest thing from a crypto apologist, but reading this PR from Coinbase interview with these folks sounds more like one of the few artistically-interesting applications I've heard of yet.

Maybe she's quoted elsewhere as "all about the crypto" but that reads more like "let's make sampling and remixing more attributable but permissionless" which - though certainly a claim made by web3shits, isn't itself an inherently exploitative use the way most web3 things come off.

This seems several steps removed from cynical bored ape NFT cash grabs. In the end I enjoyed her music first, so I guess I'll give more leeway.
posted by abulafa at 9:05 PM on May 5, 2022 [4 favorites]

I despise crypto/NFT/web3 bullshit, but this is a thoughtful probing of a deeply fascinating and soon-to-be-world-changing topic and derailing it with "btw they have bad opinions on something unrelated" does the post a disservice I think.

I've been absolutely stupefied at DALL-E 2 for the last month -- we've basically taught a computer to imagine and can also see its thoughts. I check the unofficial subreddit on a dall-e daily basis just to marvel over the beauty, creativity, and surreal humor of the latest images; my favorites are going in the DALL-E 2 thread every few days as long as it's open, but here's the latest batch for your perusal:

"watercolor painting of an off-white cowboy hat on a wooden table in front of a window" [source]
"photo of a koala swimming through thousands of tomatoes" [source]
"heart made of water" [source]
"Supervillain drawn in the style of Akira Toriyama using colored pencils" [source]
"An airbrush caricature of an old man" [source]
"Screenshot from 2020 Star trek the next generation reboot" [source]
"lion playing violin with blue moon in the background" [source; also available with an N95 face mask!]
"Stranger, by Caspar David Friedrich" [source]
"An orange cat staring at a drawer filled with socks on fire, high-resolution photo" [source; apparently inspired by this weird shitpost]
"A bunny eating a bagel!" [source]
"mosaic of bad ancient roman cat being chased by angry ancient roman woman" [source]
"a painting by Grant Wood of an astronaut couple, american gothic style" [source]
"a toulouse lautrec painting of a giraffe eating a crepe with the tour eiffel in the background" [source]
"A pre raphaelite painting of a person waiting for their iPhone to power on after plugging it back in" [source]
"A photograph of an apple that is a disco ball, 85 mm lens, studio lighting" [source]
"a kid dressed up as a Storm Trooper celebrating May the Fourth by performing a TikTok dance, unreal engine" [source]
"Mozart drinking a smoothy" [source]
"A pondering philosophical grizzly bear, digital art" [source]
posted by Rhaomi at 11:38 PM on May 5, 2022 [11 favorites]

I despise crypto/NFT/web3 bullshit, but this is a thoughtful probing of a deeply fascinating and soon-to-be-world-changing topic and derailing it with "btw they have bad opinions on something unrelated" does the post a disservice I think.

I agree on both points (NFT/Web3 et. al are garbage and it's possible that someone in that space might have something important to say).

One of the last times Mefi thrashed around the AI-art subject, I tried to make the connection between the beginnings of photography and where we're at with AI image generation, and in this article the authors did this much more eloquently by pointing out the evolution of pictorialism. It's encouraging to me that they have a firm understanding of the history of image creation.

I get it though, the first AI images that emerged were slightly creepy, used garish colors, and felt like a one trick pony. I.e. very easy to write off aesthetically.

But it's clear that things are evolving rapidly, and that shaky, neon-colored "Early AI" style is getting left behind quickly and for me, that brings this odd sense of quease and intrigue. I'm keeping an open mind, personally.
posted by jeremias at 3:17 AM on May 6, 2022 [3 favorites]

This and GPT-3 are ... freaky.
posted by MacD at 6:11 AM on May 6, 2022

I don't think I understand how this changes whatever kind of game it's changing; I don't share her excitement.

> While this development is but the latest advancement in a legacy of machine learning in art too long to do justice to in a blog post, this act of conjuring artworks from language feels very very new. Feeling is an important dimension to the act of creating an artwork, as while we have for some years had the capacity to generate art from the laborious process of training GANs, often waiting overnight for results that invite the observer to squint and imagine a future of abundant possibility, a perfect storm in the past 18 months has led to a present in which the promise of co-creation with a machine is realized. It feels like jamming, giving and receiving feedback while refining an idea with an inhuman collaborator, seamlessly art-ing . It intuitively feels like an art making tool.

It's play and exploratory; like sketching and studies but the artesanal thing has to come from graft. The thing on being read in 200 years' time made the point about being humbled within the many small steps in the act of creating so as to bring forth enduring art. 'If it were easy to do it wouldn't be worth achieving the outcome', so to speak.

But that's very much part of the mythos you create for the output you're looking to sell. I doubt that these outputs sell and the notion of monetising it lands us in the argument of who owned or inspired the machine to do the bulk of the work involved in declaring this to be art. Given the notes on remixing for the author of TFA, maybe they expect to be able to retain the kudos for creating the meme even if no cash payout.
posted by k3ninho at 6:40 AM on May 6, 2022

I read a Twitter thread recently on exactly this programme and the serious ethical problems it (like most AI/machine learning tools of this type) presents but which boosters like Herndon (whose AI-assisted album I should say I enjoyed!) do not speak about.

As the thread notes, there's a conversation missing here, perhaps.
posted by deeker at 6:46 AM on May 6, 2022

Not to abuse the edit window...

I found the thread nested within another thread on the dangers these machines pose and the distinct possibility they may be insurmountable...
posted by deeker at 6:51 AM on May 6, 2022 [1 favorite]

I don't think any technology is intrinsically good or bad, or dangerous or benign. The only relevant question, to my mind, is whether humanity as a whole can be trusted with it. In the case of DALL-E, GPT-3, and similar AIs, obviously humanity cannot. But it's equally obvious to me that humanity can't be trusted with technology such as firearms or motor vehicles either, and those proverbial ships sailed a long time ago. Maybe the only healthy response is pity for those who will have to live in the world we created.
posted by Faint of Butt at 7:07 AM on May 6, 2022 [2 favorites]

In the long arc of time, we have basically three choices.

All art and creation are the sole and intrinsic property of one original human creator employing whatever tools they choose. Intellectual property is an infinitely lucrative legal field determined to make the subjective objective and get paid while occasionally paying creators. We'll call this "infinite Disney"

Art as we understand it and creation itself is the important part and a snapshot of a product is nearly valueless in and of itself. The value is in the means to produce more or interpret new ideas and collaborate with any creative force machine or human, everything is a tool and nothing is final. Live performance and creators themselves are the only artistic endeavor of value and they all remix, even if it's based on embeddings stored in wetware we might have previously called "influences." We'll call this "happening singularity"

Or, maybe, some middle ground where we teach ourselves and our tools and our machines to understand some concepts of how art is created and they become tools just like brains and bodies that remix and mutate memories and knowledge and talent into new things but still certainly rely on and deserve to give credit to original sources. Although couched in some Web 3 terminology, I think this is what Herndon and company are shooting for.

So, I'm not sure what the value is in agonizing over how nothing will ever be real again. It already isn't, it's just not evenly distributed. Trying now to build some provision in tooling and creative monetization to make it worthwhile to retain credit and accountability of creative ideas. Seems like a decent middle ground. The alternative is giving up entirely and letting the trolls who have no compunctions about theft, reproduction, and disinformation run the show.
posted by abulafa at 8:49 AM on May 6, 2022 [1 favorite]

We may agree that creativity is a human activity, but I doubt that we can agree on what creativity is and how it works in the human mind. So far, we know that it happens. As to what art is and how it is done is also a still argued over topic with no final answers looming in the mists. So far, we know that there is stuff called art out there. But, if you think it’s a pretty picture and you don’t care if some starving artist in a loft made it, or that it was just computer output based on Markov Chainy churning through a zillion images found on the internet, then yes, you found yourself a pretty picture.
posted by njohnson23 at 9:28 AM on May 6, 2022

I share the concerns about the strange and potentially destructive copyright issues of AI, but my gut tells me it may be a bit of a red herring, it's an easy target to focus on (albeit an important one).

But again, I reach back to the late 19th and early 20th century where photography had been around for 30-60 years, and yet we still saw painting and art thrive and evolve with all the "isms" of the day ( Impressionism, Post-Impressionism, Cubism, Surrealism, etc.)

I don't see AI-assisted or generated art to be all that different. In the near future, some canny and talented artist will discover that AI can be trained on her own artwork (perhaps oil, watercolor, photos, what have you) and spawn new pieces using her own style, not biting other artist's work.

Or what about people with severe motor disabilities who find it difficult or impossible to physically hold brushes, pencils, etc. The option to create or contribute to art through the medium of language opens up the field of visual arts to new contributors, greater diversity, and who knows what else.

Unlike the smoke of mirrors of Nft/Web3, blah blah, which strike me as technological solutions in search of a problem, the language based AI image stuff actually seems to be solving problems that certain (not all) artists face.
posted by jeremias at 9:55 AM on May 6, 2022

I found this whole thread to be fascinating and thought-provoking. Thanks, all.
posted by rrrrrrrrrt at 10:23 AM on May 6, 2022 [1 favorite]

It wasn't just the Pictorialists. In a sense, the entire modernist project can be seen as a response to photography. Like, it's no coincidence that photography is invented, and suddenly we get realism, and various forms of impressionism, abstraction and minimalism that continue through the 1960s when modernism allegedly died an ignominious death. Though, really, go to any large law office or monthly art walk in a large city, and you can see undead copies of modernism shambling along to this day. Which is fine. I say it like it's a bad thing, but it's fine. (Also, I know that there were other factors. The point is that photography was a huge one.)

It is also true that photography absolutely gutted several commercial art industries. Illustration and portraiture are two that come to mind, but I'm sure there were others. Yes there are still illustrators and portrait painters, but fewer, maybe than there would otherwise be, and they don't have the monopoly on image creation that they once did. Arguably there's a benefit that outweighs the harm (I mean, I think there's a benefit that outweighs the harm). Everyone reading this thread will have had hundreds or thousands or more of photographic portraits made of them over the course of their life. I'd say that is a far better outcome than having one or two or zero portraits painted over the course of a lifetime, though a 19th century portrait painter might have a different take.

A friend of mine went to see Jurassic Park, and he came back raving about how it had "REAL dinosaurs!" But he also said he knew that a big reason for his reaction was a lack of sophistication on his part when it came to recognizing computer effects. He predicted a future where audiences would become such connoisseurs of computer generated effects that much of it would be noticed and rejected, which is exactly what happened. It still happens 30 years on. It's good to keep in mind when confronted with new technology.

Artists are going to respond to this. They'll use it as a tool, and they'll make art in ways that the AI isn't good at. If you are looking for a manifesto, I'd suggest making art that focuses on images that are difficult to describe in words (think the illustrations from those Where's Waldo books where there are hundreds of things going on, or similarly complicated pixel art drawings that you used to see), or making art that isn't digital (the AI can make art that looks like a scan of a painting, but to make a real painting they'll need to build a robot too).

It seems pretty clear to me that whole swathes of commercial artists are about to be directly harmed. Dall E 2 is already making images that are good enough to replace finished artwork at a certain level, and there's no reason to think that it wont get much better really fast. This is the foothills, not the peak. If I made digital art with commercial aspirations, I would find it very difficult not to be looking at AI researchers as a bunch of supremely evil motherfuckers, right now, nevermind that it all might be for the best as far as humanity is concerned.

I spent yesterday working on some designs for giant robots for a painting I want to do. It's a middling hard problem because I want a robot that looks credible, isn't a copy of someone else's design, doesn't already exist, but still clearly reads as a robot. It sucks because every time I do a thumbnail that doesn't work, I get a feeling of failure. It requires plugging away with no guarantee of success, and I don't find it fun at all. If I could just tell an AI "Draw a 1940s giant robot" and get 30 copyright free designs which I could then tweak to my needs , I'd do it in an instant, and whatever painting I ended up making would probably be better for it. (Obviously I signed up to test Dall E2 the instant I learned that I could.) But people get paid to do that kind of design now, and they get paid because it's hard...

Characters on Star Trek would sometimes talk about how they "programmed" the holodeck, and I always used to think they were a bunch of fucking liars. The UI for the holodeck was natural language. Tom Paris was like "give me a 1930s pulp sci-fi adventure, but stripped of overt racism, as far as a 1990s sensibility allows" And suddenly he's Flash Gordon. He wasn't hacking away in Assembly. But maybe that counts, and my views on the subject are hopelessly backward. Beginner artists will sometimes feel shame about tracing, but the truth is that tracing is fine and way way way more prevalent than a person might naively think.

AI is coming for most creative fields, and it's coming really uncomfortably fast. If you are a writer, or an artist, or a musician, or a poet, or a programmer, it's coming. People are going to fight it, and they'll win, much like John Henry won, which is to say, for a while, with difficulty, and nowadays nobody even knows what a steel driver is.
posted by surlyben at 10:39 AM on May 6, 2022 [6 favorites]

Another thing that concerns me about Herndon and Dryhurst is that their artistic practices seem to be backed by a lot of organisations in Russia; a Russian art/cultural-theory publisher named Strelka publishes some of Drhyurst's writing and he has lectured in an online course run out of Moscow. It could well be that this is just the vigorous civil society that the Putin regime hadn't yet crushed, and the kleptocracy hadn't yet corrupted, or it could be that they are willing or unwilling pawns in some information-warfare operation, and their frenetic shilling of crypto/blockchains to people who see themselves as more left-wing futurists than techbro ancap libertarians doesn't incline me to give them the benefit of the doubt.
posted by acb at 10:39 AM on May 6, 2022 [1 favorite]

derailing it with "btw they have bad opinions on something unrelated" does the post a disservice I think.

I like Holly Herndon's music a lot and some of the stuff she does is super interesting, but I really don't think this is a derail: crypto/web3 shilling utterly pervades her public presence of late, and I don't trust anything she does to not have this angle lurking somewhere. E.g. what are the odds that the stuff in this article does have something to do with whatever her "AI twin DAO" alt twitter account has been hinting at? Pretty far from zero. (For that matter the article itself is posted via "The essential web3 toolkit for sharing and funding anything" that is primarily designed for crowdfunding via crypto, it has a link to an eth wallet where I'd expect a byline.)
posted by advil at 12:38 PM on May 6, 2022 [1 favorite]