Hacker News Reader

> 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

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.

> Modern Art was created

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.

Photography eventually led to the creation of an entirely new vidual medium though: films.

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.

It's possible to view movies as a new form of art not possible before photography, but one could also view them as automated plays and the thing that killed vaudeville.

I loved this comment so much, I had to toss the first sentence into a DALL-E Mini renderer. It came up with this: https://imgur.com/a/P6SPhlH. This one left me with the most sense of existential dread that I get out of your viewpoint for how digital art evolves.

That is scary...

Looks like Stephen Merchant.

That is really, all of that.

[Stable Diffusions interpretation of the second sentence](https://media.discordapp.net/attachments/1006094456799363082...)

> 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).

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 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.

Artists aren't horses because artists are actually intentionally participating in the economy. Horses just want to do horse stuff and don't particularly have any desire to be employed by carriage drivers.

I don't really see what the difference is. If I want to intentionally participate in the economy by dragging a carriage with people in it around in exchange for money, no one is going to employ me because my services are obviously inferior to those that a car can provide.

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.

Well, what I mean is the horses and carriages aren’t even produced anymore since we don’t want them, and they didn’t do anything to make themselves useful again, not being sentient.

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.

Exactly. Which is why Vinyl sales have grown, because of the digital abundance analog scarcity is attractive.

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.

> 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.

It'll be good for accessibility too; if you can make a text-to-image model then you can make an image-to-text model, and now you have a screen reader.

image-to-text models for captioning already exist. The most common one is CLIP from openAI. https://openai.com/blog/clip/ Jina AI has an out-of-the-box implementation for it https://clip-as-service.jina.ai/

I see it playing out this way too. Technology at societal scale is "and" not "or", and computers have long been cast in the scapegoat role because of a basic principle that is not unique to them, but is made all the more clear by their existence: it's easy to get wrong answers infinitely fast.

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.

Vinyl has come back for a variety of reasons, but digital abundance is one of the lesser reasons.

- 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".

It should be noted that many of your points are already hit by "buying digital music instead of using streaming platforms". Those points are important to me, yet I’d never buy a vinyl.

It honestly depends. I can name quite literally hundreds of digital releases that are objectively, measurably worse than their vinyl counterpart.

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 can name quite literally hundreds of digital releases that are objectively, measurably worse than their vinyl counterpart.

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.

> If I want to intentionally participate in the economy by dragging a carriage with people in it around in exchange for money, no one is going to employ me because my services are obviously inferior to those that a car can provide.

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.

Saying "The artist's job is to elicit emotion" is such a broad statement that it stops being a meaningful lens to look at the world and draw conclusions from. If you say that, then you say that a painter is the same as a cinematographer, or a fine chef, or a therapist, or my family. The conversation that this blog post brings up is about the very precise boundary of people who create art through static visual mediums. Saying that their job is the same as a cinematographer - when a professional artist has probably spent ten thousand hours learning to draw, and virtually zero on film, is meaningless.

I’m a professional artist and while this is true to an extent, most of my job - and a lot of what I enjoy about it - is “drawing”. Running a thousand prompts through an AI black box sounds infinitely tedious compared to drawing stuff. I don’t want to live in that world.

Don't worry, you won't have to. AI-generated images aren't possible to copyright, and as they're all derived from existing sources (read: plagiarized), you're not going to be replaced any time soon.

They're not necessarily plagiarized anymore than a human artist plagiarizes from looking at an existing image. If you accepted this, Disney would basically own every animation ever made.

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.

> DALLE is a continuation of Photoshop. Compilers didn't put programmers out of business, it just moved them one layer up.

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.

> DALLE allows any schmuck with zero artistic talent to generate their own art. It obviates the need for artists.

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?

Like many iterative technologies, Wordpress had the right idea with clunky, underdeveloped execution. Nevertheless it was wildly successful but the way that it works often requires some web development (or at least technical administration).

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.

Depends on what kind of complexity a website requires. Predominantly static portfolio or a website with basic ecommerce functions? Sure. Custom internal workflow and lots of externally interactive stuff with an API? Yeah, you're not going to be served well by Wix or Squarespace. Different tools for different tasks. It's silly to compare a website for the local salon with a platform/utility/tool/whatever.

> The same way Wordpress obviates the need for web devs.

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.

> Compilers didn't put programmers out of business, it just moved them one layer up

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.

Has anyone connected one of these models to zork or planetfall yet?

no, but I've used some descriptions as prompts (mostly on Craiyon) and got quite acceptable results. Plain descriptors like how you are standing in an open field to the west of a white house, with a small mailbox visible produced multiple usable candidates (if you didn't mind some quirky impressionism). 'A maze of twisty little passages, all alike' gave me all 2d mazes, but adding '3d first person, photorealistic' gave some entertainingly weird choices.

The good images come from detailed prompts, it requires creativity, visualising what you want out of it. I think anyone writing a prompt and getting what they want out of it (vs. something surreal and hilarious for its unexpected whackiness, at least while the novelty lasts) is 'doing art', just as a photographer chooses a subject and frames a shot.

This is prompt engineering, which as I said in the previous message, I 100% believe is not going to be a thing anymore very soon.

There's no algorithm that generates artistic meaning. These neural nets allow you to employ arbitrary styles and techniques, but no AI can't come up with meaningful messages on its own...yet

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

You just need to set up a generalized adversarial model that optimizes for meaning!

I'd agree with you, but I have yet to come across a salient, non-circular definition of "meaning" in this context that makes it somehow unobtainable by robots. What is the meaning you speak of? This isn't meant to come across as snarky either, I'm genuinely curious to hear your thoughts.

I think of meaning in terms of Saussure's sign and signified. "The Earth orbits around the sun" for example, is literally a string of characters. As a sign, however, its correspondent signifier could be a diagram of the solar system or something like that.

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 expect little to no digital art (images, videos, films, music, writing) will be sold for profit by human artists 10 years from now

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

Yeah anyone claiming they understand what these things do, and how they do it, and then expecting them to replace digital writing/film is just...not going to happen. Not in 10 years at least. We're not even close.

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.

Is it far-fetched to think that AI animating comic books / manga into cartoons is that far away?

Yes. I don't think we'll even have AI "Drawing" comic books in 10 years, at least not the way most people here are thinking of it.

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.

I see how AI movies could replace every single Marvel movie since Endgame so. Toss the AI the comic story lines, give it endless compute capacity to create CGI effects and deep fakes if actirs and you have it.

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.

At the end of the day, either AI art is "good enough" (whatever your criteria are - "novelty", "soul" etc), or not. If it's not, then human-made art will continue to exist, to the extent there's demand for it. But if AI really is that good at art, it would seem that it's not really a problem from the perspective of society at large. It would be a problem for artists who rely on art as their source of income, but that's something that is fairly easy to solve, since all it takes is money.

In my estimate, the common joe cannot distinguish between "real" and AI-generated art today. We're only seeing the first generation of AIs right now with the tech such as Midjourney, DALL-E 2, Imagen and Parti. I think it's pretty presumptuous to assume this trajectory would not continue to the point where most people would have a hard time identifying AI-generated images from real ones.

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.

*Average Joe, just FYI.

> 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?

The awareness for Dall-E / Midjourney is exploding. These tools are incredibly compelling and things like game art will definitely be generated by AI's in the near future. The possibilities are barely explored.

Leonardo da Vinci didn't just pick up a paint brush, knock out the Mona Lisa, and call it a career. He built his skills as an apprentice in workshops doing background figures and lesser commissions. Later artists (to the present day) have created prints and works for clients.

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'll add in that sculpture too might see challenge, thought it's materials might see resistance for some time. 3d Printing is already at a state where the output is of extremely high detail, and still on it's ascent (resin printing specifically). While not a like-for-like comparison, it's seen heavy use in favor of traditional sculpting in multiple industries, such as miniature production, where it completely replaces or heavily reduces the use of the traditional skillset.

An AI could make something look like a Picasso, but it could never invent cubism. Without thought behind it, it is just advanced regurgitation.

Could an AI, and some selective process (think YouTube algorithm) over billions of iterations evolve cubism?

> An AI could make something look like a Picasso, but it could never invent cubism.

I wouldn't bet on this

"Never" and "AI" in the same sentence requires a high level of confidence in a specific future. I don't see why AI of the future couldn't generate a mind with all of the complexity and life experience (simulated) that Picasso was inspired by. Do that a trillion times and we might get things even cooler than cubism.

Well conceptual artists get a big boost you used to need to have a budget to hire artisans to build out your ideas now all it takes is an app.

Let's not stop here.

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?

Blue collar jobs that require dexterity

I think nursing will be the last job to get automated away. It's all about dexterity and human interaction.

Dull, dirty and dangerous jobs will be the last. Why waste a perfectly good and expensive robot if a cheap human can do it?

Disagree on 2/3. Dull tasks are some of the best for robots - their attention never wanders, they never get fatigued, and their performance is always going to stay consistent. Dangerous jobs are already being given to robots - just look at the military. Even outside of that, humans getting hurt means liability/expensive insurance (not to mention bad publicity).

Humans will probably be stuck with the dirty jobs, though.

From a song in an old Soviet movie: “all worries are forgotten, all rat races have stopped: robots are toiling, humans are happy”.

Your last paragraph is one potential use for blockchain that migjt actually work. Assuming a manually taken picture (regardless if digital or film) has some inherent value that an AI recreation does not, using blockchain to proof an actual person took that picture on site serves a purpose. Personally, I think those "orginals" have value.

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.

While I agree with your sentiment, I don't see why this couldn't be accomplished with just standard cryptographic hashing verification based off of, say, a scan of the artist's iris salted with a timestamp that is then encoded into the file. What do you think a blockchain would add to it?

Sure, some artists, like Glen Cook, can be replaced by AI, but honestly they are already not quite readable.