AI-Generated Images.... Are We Fucked?

[Sebastian Cobb]

Any technology can be overcome by technology. It'll become an ongoing war of attrition so the general public will be easily fooled by fakes agreeing with their own biases.

I'm at the stage now where my default assumption is that any photo or video is fake.

Yep, they're using AI to game the AI with artefacts, so it seems pretty logical the counter step is to use the tool to generate a load of ai-gaming images of the same images and train the gamed AI that they're all the same again.

[Jerzy Bondov]

Maybe i'm a caveman with no imagination but i just don't see any practical use for this stuff. Apparently it costs a huge amount to run the servers, which are crunching huge amounts of information in order to produce a worthless, uninteresting picture of nothing. There's a couple of guys in my Banter Group Chat who will sometimes generate a picture of whatever we were talking about, for example one of us purchasing tools from Toolstation, and it's just a total bants vacuum. But no apparently this is going to end the world somehow.

[Zero Gravitas]

Maybe i'm a caveman with no imagination but i just don't see any practical use for this stuff. Apparently it costs a huge amount to run the servers, which are crunching huge amounts of information in order to produce a worthless, uninteresting picture of nothing. There's a couple of guys in my Banter Group Chat who will sometimes generate a picture of whatever we were talking about, for example one of us purchasing tools from Toolstation, and it's just a total bants vacuum. But no apparently this is going to end the world somehow.

A nuclear fusion breakthrough is needed to meet the vast energy requirements of future artificial intelligence, according to OpenAI boss Sam Altman.

The ChatGPT creator currently handles hundreds of millions of queries daily, with estimates suggesting that this consumes around 1GWh each day – equivalent to around 33,000 US households.

As generative AI tools like ChatGPT improve in functionality and increase in popularity, this energy consumption is forecast to increase significantly over the coming years.

"There's no way to get there without a breakthrough," he said at a Bloomberg event at the World Economic Forum's annual meeting in Davos on Tuesday.

"It motivates us to go invest more in fusion."

[touchingcloth]

Yep, they're using AI to game the AI with artefacts, so it seems pretty logical the counter step is to use the tool to generate a load of ai-gaming images of the same images and train the gamed AI that they're all the same again.

Who will block the ad blocker blocker blocker blocker blockers?

[perplexingprocrastinator]

Of course there's always a sense of hollow weirdness after you're done - you don't feel any real attachment to the image because you didn't do anything to make it beyond fuck with your prompt to wrangle the AI into giving you what you wanted, nor is there much reason to show it to anyone else, beyond the novelty of "hey, look, a computer made this".

This is incredibly well put.

AI images, like those you posted, can absolutely be fun or striking, or even beautiful. Although I will say - before you created those - didn't we already have tens of thousands of images which, perhaps weren't identical - but which were essentially the same as these? What have they added? (I'm not being a dick about your fun pictures, I'm just reaching for the example you provided and astutely critiqued yourself :)

They can't have added anything meaningful because they are literally composed of images which already existed. And let's face it, these kinds of images were already utterly generic even before a code started making them.

So - they aren't art, and they never will be art. Because art can only be done by a sentient person.

Now this reminds me of another thing, which is that documentary "Tim's Vermeer" produced by Libertarian halfwit Penn Gillette. The film concerns some rich plonker who figured out the techniques by which Vermeer (probably) achieved some of his art.

The plonker goes on to replicate the process, and produces something which is - at a glance - reminiscent of a Vermeer. But it's not a fucking Vermeer, because it wasn't done by Vermeer, at the time and in the place that Vermeer produced the art, in discourse with art history up to that point, and with the emotion and talent and crucially INTENTION which makes art art.

Art exists in dialogue with the creator and the viewer, with the society and culture in which it was made, with the art which influenced the artist. You can't have art without an artist. You can have images, and they might be interesting, or even moving - but it will be accidental. AI can create content, and we're about to enter a world which is saturated in it. Hollow, facile, oppressive, ubiquitous, ugly, pointless content. But AI can't make art. Perhaps it can be used to create some kinds of conceptual art. Which might even be really good. But that's a narrow cultural seam. In terms of what we generally mean by "art"... no. Just, no. It's all fucked.

Episode of the fantastic Michael & Us podcast which does a much better job of expressing these sentiments -

https://shows.acast.com/jacobin-radio/episodes/michael-and-us-penn-paintbrush

Oh, and plus everything people have rightly pointed out about energy consumption.

I demure on the topic of AI curing diseases and whatever else because I don't know enough about it. But in terms of culture, "synthetic" culture is a fucking travesty in my opinion.

[Sebastian Cobb]

So - they aren't art, and they never will be art. Because art can only be done by a sentient person.

I think the elephant in the room is lots of graphic design isn't really art, or doesn't really have to be, when it's just a vehicle to shift some units. And ultimately that's what's going to get decimated first.

[Goldentony]

yes

[perplexingprocrastinator]

I think the elephant in the room is lots of graphic design isn't really art, or doesn't really have to be, when it's just a vehicle to shift some units. And ultimately that's what's going to get decimated first.

First they came for the journeyman graphic designers.

And I did not speak out, for I was not a journeyman graphic designer.

But then they came for ME and that fucking sucked!!!!!!

[jamiefairlie]

I think the elephant in the room is lots of graphic design isn't really art, or doesn't really have to be, when it's just a vehicle to shift some units. And ultimately that's what's going to get decimated first.

Quite, I used to work on FIFA and 99% of the art on that was churned out of factory in China. Could all be done by AI now.

Most art is commercial and just needs to be good enough.

[jamiefairlie]

This is incredibly well put.

AI images, like those you posted, can absolutely be fun or striking, or even beautiful. Although I will say - before you created those - didn't we already have tens of thousands of images which, perhaps weren't identical - but which were essentially the same as these? What have they added? (I'm not being a dick about your fun pictures, I'm just reaching for the example you provided and astutely critiqued yourself :)

They can't have added anything meaningful because they are literally composed of images which already existed. And let's face it, these kinds of images were already utterly generic even before a code started making them.

So - they aren't art, and they never will be art. Because art can only be done by a sentient person.

Now this reminds me of another thing, which is that documentary "Tim's Vermeer" produced by Libertarian halfwit Penn Gillette. The film concerns some rich plonker who figured out the techniques by which Vermeer (probably) achieved some of his art.

The plonker goes on to replicate the process, and produces something which is - at a glance - reminiscent of a Vermeer. But it's not a fucking Vermeer, because it wasn't done by Vermeer, at the time and in the place that Vermeer produced the art, in discourse with art history up to that point, and with the emotion and talent and crucially INTENTION which makes art art.

Art exists in dialogue with the creator and the viewer, with the society and culture in which it was made, with the art which influenced the artist. You can't have art without an artist. You can have images, and they might be interesting, or even moving - but it will be accidental. AI can create content, and we're about to enter a world which is saturated in it. Hollow, facile, oppressive, ubiquitous, ugly, pointless content. But AI can't make art. Perhaps it can be used to create some kinds of conceptual art. Which might even be really good. But that's a narrow cultural seam. In terms of what we generally mean by "art"... no. Just, no. It's all fucked.

Episode of the fantastic Michael & Us podcast which does a much better job of expressing these sentiments -

https://shows.acast.com/jacobin-radio/episodes/michael-and-us-penn-paintbrush

Oh, and plus everything people have rightly pointed out about energy consumption.

I demure on the topic of AI curing diseases and whatever else because I don't know enough about it. But in terms of culture, "synthetic" culture is a fucking travesty in my opinion.

It doesn't matter what we philosophise as the meaning of art, most people will consume this stuff and see no difference whether it's made by humans or AI. The games up, all that will remain is the kind of 'human made' artisan expensive stuff for well off snobs who like to show off to their friends. Like the stuff Enfield sells in his "we saw you coming" shop. Mass market artists have no future.

[jamiefairlie]

Maybe i'm a caveman with no imagination but i just don't see any practical use for this stuff. Apparently it costs a huge amount to run the servers, which are crunching huge amounts of information in order to produce a worthless, uninteresting picture of nothing. There's a couple of guys in my Banter Group Chat who will sometimes generate a picture of whatever we were talking about, for example one of us purchasing tools from Toolstation, and it's just a total bants vacuum. But no apparently this is going to end the world somehow.

I honestly think you're at the stage equivalent to the coach drivers staring at the sputtering monstrosity in front of them and laughing about how superior their horses are.

The thing about AI is how fast it can learn and adapt. It will become amazing efficient and effective in a relatively short time span. It doesn't have to be great, it just has to be good enough and it's really close.

The big challenge is not the AI itself, it's how corporations utilize it to save money and that is being actively worked on like crazy.

[perplexingprocrastinator]

Most art is commercial and just needs to be good enough.

Again, as a working illustrator - already in abject despair about the quality of children's picture books - I still think the future is pretty dispiriting.

And I can't just say "Hey-ho" about this... For a long time, the majority of art and design has been produced for commercial purposes, because someone has to pay the bills. But even in that context, when it was down to human talent, companies would pay a premium for creativity and skill, and we all benefited from having nicer images on the tat we fill our lives with. Reaching for an example, Quality Street used to look like this:



Since photoshop came about, and design went to shit, and everyone stopped caring, Quality Street now looks like this:



Basically, like puke. Because there's no craft any more - no actual human pushing against the limits of their creativity and skill. I have to assume that AI is going to accelerate this collapse into total banality in the vast majority of images and design we're surrounded by.

Perhaps not. Perhaps people will figure out how to make it produce images, the likes of which we have never seen before. But it still would be the product of a lifeless machine and therefore, I will dogmatically loathe it even if totally owned by an AI produced picture of devastating power. If a robot did it, get to fuck

[ASFTSN]

Again, as a working illustrator - already in abject despair about the quality of children's picture books - I still think the future is pretty dispiriting.
...If a robot did it, get to fuck

Same - but really all I can do is carry on 'coz there's nothing I can do to adapt. It's not hey-ho, but all I can do for now is keep going. Fucked or not (industry-wise) no-one can stop me making images on paper, so I don't really see any choice but to keep going.

EDIT: I suppose to put it another way, I was told what I was doing was pointless anyway before AI art started popping up

[Lemming]

They can't have added anything meaningful because they are literally composed of images which already existed.

Agree with most of your post but to expand on this: the interesting thing is the way in which the AI "learns" concepts, styles and objects, and the user's imagination then intersects with the pre-existing data to create entirely new things. You can use it to create images which are completely unlike anything in the training data.

To just come up with an example image made of totally discordant, incoherent concepts, which taken together will have no extant analogue in the training data - a middle-aged woman with bionic eyes, wearing a burkha, flying through space on a giant cat, holding an old PC, with the word "LEMMING" written in pink glitter text below her, drawn in the style of 90s fantasy art.



I think it'd be very hard to find images in the training data that correlate to any part of this; every element is likely to be original. The PC, the woman's face, her clothes, the cat, the planet's atmosphere below, the starscape, the text, the colours, the composition: none of it is taken directly from any existing image in a way that would be recognisable. The AI isn't copying and pasting from existing sources, rather it's drawing on what it "knows" about a vast number of different concepts to provide something that matches my absurd prompt. It is an amalgamation of pre-existing images... but it's also sort of not. This isn't to step into the debate of whether or not it's "art" (I'd agree that it self-evidently isn't), but there's quite a bit more going on than just regurgitation of existing images wholesale.

You can see this in action if you train your own Stable Diffusion concepts. You can put your own face in with about five photographs and then generate images of yourself on trial at Nuremberg or whatever for a laugh, and you'll see yourself replicated in poses, situations, art styles and angles that you never included in the small training data sample. The AI "learns" you as a concept, and can then apply you in all kinds of ways that go far beyond the photographs you showed it.

Large language models are even more exciting than the image generators for similar reasons; you can quite literally do or say anything, and generate sentences, situations and even words that are entirely original. It amazes me that you can come up with completely indecipherable situations and the AI will still "know" what you mean and react in a coherent and sensible way, even if you start writing a story about giving birth to a thousand screaming demonic hellhounds through your nostrils or something.

[touchingcloth]

Is there any proof that Shakespeare was objectively better of a writer than ChatGPT?

[bgmnts]

Honestly yeah, I quite like the fact that I can make an image in my head become a real image without having to have any artistic talent.

Drawing that well isn't really something everyone on earth can do, so it's nice I can realise my own imagination visually.

It takes no skill to capture a nice image with the click of a camera, before that you'd have to have stayed in that spot for ages and painted it by hand, which only a few people could do.

[Zero Gravitas]

Can you though? Or are you just presented with something that sort of fits your textual description of it?

[perplexingprocrastinator]

EDIT: I suppose to put it another way, I was told what I was doing was pointless anyway before AI art started popping up

It'll be harder to get paid but the work you do will always have value - the value is what your effort, skill and imagination put into it. You being you, means your work will always have a type of value impossible to replicate by machine. The DOING of it creates value. The YOU of it creates value. Authenticity - though bastardised into mere currency by the market - does have meaningful, spiritual, innate value. Your work could not be done by anybody or anything else.

Solidarity, hey :)

[bgmnts]

Can you though? Or are you just presented with something that sort of fits your textual description of it?

It'd probably be the same level of discrepancy as if I gave that textual description to a human artist, no? Just sort of what I wanted.

I've never commissioned art obviously but I imagine it being exactly what you looked like in your head to be very rare.

[ASFTSN]

It'll be harder to get paid but the work you do will always have value - the value is what your effort, skill and imagination put into it. You being you, means your work will always have a type of value impossible to replicate by machine. The DOING of it creates value. The YOU of it creates value. Authenticity - though bastardised into mere currency by the market - does have meaningful, spiritual, innate value. Your work could not be done by anybody or anything else.

Solidarity, hey :)

Brilliant post for a Friday afternoon, agree on all points!

gonna go piss on a computer (not really)

[Jim_MacLaine]

It takes no skill to capture a nice image with the click of a camera, before that you'd have to have stayed in that spot for ages and painted it by hand, which only a few people could do.

Not on nodding terms with the concept of composition then?

[Zero Gravitas]

It'd probably be the same level of discrepancy as if I gave that textual description to a human artist, no? Just sort of what I wanted.

I've never commissioned art obviously but I imagine it being exactly what you looked like in your head to be very rare.

I thought you were vividly imagining something and saw that turned into an image, but it's just the google images experience of typing in "Dog wearing a Party Hat" and getting this:



That's not a new thing though.

[bgmnts]

Not on nodding terms with the concept of composition then?

What I'm saying is before photography, the only way to capture real images was painstaking years of learning to paint and then painstaking hours of sitting down and brushing it out.

Then anyone could click a button on a camera and get the image what they were looking at, now it's a phone with a camera on it as standard. I'm not talking about the skill of photography.

I thought you were vividly imagining something and saw that turned into an image, but it's just the google images experience of typing in "Dog wearing a Party Hat" and getting this:



That's not a new thing though.

That is searching for an existing image, not generating a new one.

[perplexingprocrastinator]

This isn't to step into the debate of whether or not it's "art" (I'd agree that it self-evidently isn't), but there's quite a bit more going on than just regurgitation of existing images wholesale.

Yeah, I definitely agree with you. That's why I left wiggle-room for what I called "conceptual" art, and I do acknowledge that this tech absolutely can be used for (potentially great) art and creativity... of certain kinds.

But it's still not a replacement for the thing it will destroy, which is a world of craftspeople with a set of human, flesh and blood skills, evolved and shared and developed over countless generations, hundreds of years. Art with humanity behind it. And effort, frankly. And failure, trial and error. These things are good, they are part of art.

Human creativity means something, and not just in a woo-woo way. The process of making the art, of working with the materials, of pushing against your own limitations, influences the art. So without people being supported to pursue this type of practise, there's a whole international culture of (human) art, centuries in the development, which is being snipped off with a shrug. AI will never replace it - but we won't know what we have lost because it's never going to exist in the first place.

I have sympathy with @bgmnts point above though, that they can now conjure things up from their imagination which they otherwise couldn't. I'm all for that, happy about it. But I still think we're likely to lose far more than we gain overall from this... ersatz, synthetic world we're being hurled into with absolutely fuck-all consideration.

[ZoyzaSorris]

'AI' is just glorified brute force data analysis on a grand scale and is generally crap, prove me wrong

[perplexingprocrastinator]

gonna go piss on a computer (not really)

Bzzzzzzzzzzzzt / 'ow, my peehole!'

Here lies @ASFTSN. They pissed on a computer

Joking apart, big yourself up. Art forever!

[ZoyzaSorris]

And I say this as someone who works in that general field

[Video Game Fan 2000]

i hate the demiurge

[Cuellar]

Is AI any different than those lens device things that great 'artists' like van eyck used to use that basically project whatever you've pointed it at onto the paper in front of you so you just have to trace it like a child?

[FeederFan500]

I am a bit ambivalent on this. If you were to look at most mass market paperbacks, for example, they are already a pastiche or cliche of the type of book that has come before them, in the supermarket atm I expect there is at least one title on a pastel background with a lone young woman in early 20th century clothing in the foreground, with a picture of some terraced houses representing her upbringing in poverty. However, the Noma Bar Murakami artwork is some of my favourite cover art, and while you could probably imitate them, I can't believe that current AI could produce the Underground one, with bold lines but the small touch of the tracks too. No doubt there are also connoissuers of the Nadine Dorries type books, but they have a very stock image feel to me.

So recreating something that has been done to death already feels a bit like cheating, but also no big loss. I've tried some AI image creation, but I lose interest in it out of diminishing returns. I don't really like digital/gaming art but a lot of the images I see shared are that style, I suppose when dealing in fantasy and non-realistic situations you can get away with imperfections a bit more. I am generally impressed by the compositions, the Microsoft one will generally put together a scene as long as there is a lot of source material. But try your B-tier artists and it can't manage it yet (even if someone like Pisarro had styles that changed over time).

So my impression so far is that it is capable of a lot, but gravitates towards the most common/popular depictions of things which means it comes across as overly cliched to me. All the women and (great hairline) men are attractive, anyone playing football is in Adidas or Nike gear, and the lighting effects are a little overdone.

But for all sorts of things where the presentation is just about giving the right impression rather than a bespoke set of images, it seems to do the job of graphic designers at a much lower cost. Things like food packaging in supermarket own brands where you aren't marketing yourself as premium.