AI crap in compo entries?
category: general [glöplog]
Still not sure whether the unethical methods with which the data corpus was generated is the main problem, or (in a hypthetical scenerio without tech-bros but with ethically harvested AI data) that artists' work is devalued by prompt->image machines to use for everyone without artistic skills.
I am worried too and I know many who are worried too but don't speak up (here). The topic is too severe and it's good to have a stream of all opinions here.
@z5 at Nordlicht demoparty I try to vote for separate AI/non-AI compos. No idea if I can be successful because there are of course other orgas who have different (and relevant) opinions. And it's more work for us because you have to check all artworks used in all those production manually. And you might need to request proofs (timelapse, more worksteps) from artists. No idea if this is feasible. But maybe then compos are more just. Still, maybe this isn't even doable because then the number of prods is so low that the compo isn't interesting anymore ;) Let's see, at the moment I can't think about a perfect solution.
@z5 at Nordlicht demoparty I try to vote for separate AI/non-AI compos. No idea if I can be successful because there are of course other orgas who have different (and relevant) opinions. And it's more work for us because you have to check all artworks used in all those production manually. And you might need to request proofs (timelapse, more worksteps) from artists. No idea if this is feasible. But maybe then compos are more just. Still, maybe this isn't even doable because then the number of prods is so low that the compo isn't interesting anymore ;) Let's see, at the moment I can't think about a perfect solution.
I guess the only solace left for professional content creators is that in a creative feedback loop, these text based "AI search and remix" tools are probably worthless since these image and audio synthesizers lack fine-grained controls (correct me if I'm wrong).
From a technical point of view, one of the main challenges is the input context, which is severely limited to usually just a few (ten-)thousand tokens (for the *entire* "session").
The computing power would have to increase exponentially for an "AI" to be used as a "conscious" assistant for an entire project (and this may never happen, unless quantum computing really takes off -- then we are doomed. doooooomed I say !)
For some reason (boredom?) I just wondered what Cassius Coolidge's Dogs Playing Poker picture would look like with cats instead. And of course, someone already synthesized that:
("royalty-free stock illustration")
"lol" but these are just *three* cats when the original showed *four* dogs.
It's all fun and games until someone loses a job.
Entire call centers (/support jobs) are going to (or already have been) replaced by AI. In China they have these massive, deep-faked telemarketing shows (to save production costs plus you get to see your favourite hosts everywhere all the time). In South Korea the last presidential election was supported by "AI" deep fakes which told people what they wanted to hear depending on the district they live in.
I mean, I find the tech utterly fascinating but also think it should be severely regulated -- and effectively banned in cases where it replaces already underpaid humans with robots or attempts to manipulate public opinion.
One of the greatest achievements that incorporates "AI" are translation tools. And best of all, they do no harm to anyone.
For example, the comments on Ever so Blue's piano music compilation prelude to tsunami ♫ are entirely in Korean, even though the guy is from Gothenburg (nice music btw).
From a technical point of view, one of the main challenges is the input context, which is severely limited to usually just a few (ten-)thousand tokens (for the *entire* "session").
The computing power would have to increase exponentially for an "AI" to be used as a "conscious" assistant for an entire project (and this may never happen, unless quantum computing really takes off -- then we are doomed. doooooomed I say !)
For some reason (boredom?) I just wondered what Cassius Coolidge's Dogs Playing Poker picture would look like with cats instead. And of course, someone already synthesized that:
("royalty-free stock illustration")"lol" but these are just *three* cats when the original showed *four* dogs.
It's all fun and games until someone loses a job.
Entire call centers (/support jobs) are going to (or already have been) replaced by AI. In China they have these massive, deep-faked telemarketing shows (to save production costs plus you get to see your favourite hosts everywhere all the time). In South Korea the last presidential election was supported by "AI" deep fakes which told people what they wanted to hear depending on the district they live in.
I mean, I find the tech utterly fascinating but also think it should be severely regulated -- and effectively banned in cases where it replaces already underpaid humans with robots or attempts to manipulate public opinion.
One of the greatest achievements that incorporates "AI" are translation tools. And best of all, they do no harm to anyone.
For example, the comments on Ever so Blue's piano music compilation prelude to tsunami ♫ are entirely in Korean, even though the guy is from Gothenburg (nice music btw).
Quote:
Still not sure whether the unethical methods with which the data corpus was generated is the main problem, or (in a hypthetical scenerio without tech-bros but with ethically harvested AI data) that artists' work is devalued by prompt->image machines to use for everyone without artistic skills.
I think both are problems that can be addressed separately. Mixing them together just makes people talk in circles.
The prospect of prompt-to-image machines becoming better and better is perhaps scary. However, what absolutely terrifies me is the prospect of diminishing level of human standards. If there are people today who cannot distinguish between LLM AI generated (and don’t care) and human made, then tomorrow there will be even more such people. The day after even more. So this tech doesn’t have to improve at all. It just has to saturate everything with crap, force the standards down and wait for The Sarges of this world to give up.
About whether it’s just an ethical problem of unsolicited data scraping or is it the artists being devalued. I agree it’s both. But about the second part. It’s not only that artists are being devalued, it’s that craft and ‘magic’ gets lost. You hone your skills by doing, you learn by doing, you grow artistically by doing. You don’t dream up “art” and call it up. It emerges from doing. It happens during a long process. Lifelong. Someone said before that someone “invented cubism”. These things don’t get invented. They emerge from experiments, from doing the craft. Prompt jockeying stalls all that, stalls development. If this was invented 150 years ago there would be no cubism. And a lot of other art movements also. These are statistical machines. Averaging machines. You can’t learn to code by using demomakers. Likewise, You can’t learn to “art” by using this crap. I can’t understand why is this so hard to grasp?
About whether it’s just an ethical problem of unsolicited data scraping or is it the artists being devalued. I agree it’s both. But about the second part. It’s not only that artists are being devalued, it’s that craft and ‘magic’ gets lost. You hone your skills by doing, you learn by doing, you grow artistically by doing. You don’t dream up “art” and call it up. It emerges from doing. It happens during a long process. Lifelong. Someone said before that someone “invented cubism”. These things don’t get invented. They emerge from experiments, from doing the craft. Prompt jockeying stalls all that, stalls development. If this was invented 150 years ago there would be no cubism. And a lot of other art movements also. These are statistical machines. Averaging machines. You can’t learn to code by using demomakers. Likewise, You can’t learn to “art” by using this crap. I can’t understand why is this so hard to grasp?
I´m quite convinced no LLM will ever grasp the true deepness intrinsic to endless scrollers inside a memorable c64 release.
There are no LLMs with enough training data to generate even a single c64 scroller.
@4gentE: You're wrong. Historically, the progress of technology has always driven the progress of art. For example, the discovery of the camera obscura led to the use of perspective projection in the Renaissance. And the discovery of photography is simultaneous with the emergence of impressionism. That of cinema to that of cubism. In the same way, the personal computer created video game industry and our demoscene. Etcetera etcetera... We still don't know what new forms of art the future will bring us thanks to AI.
And no art that exists today is going to disappear. Neither drawing nor programming nor literature will disappear.
And no art that exists today is going to disappear. Neither drawing nor programming nor literature will disappear.
Wait weren’t you telling us the AI is this great breakthrough singularity, transhumanism voodoo mumbojumbo blablabla hippie prophecy?
Are you really so rude? Or are you just drunk?
@ham:
Haha sorry sorry.
I know no art will actually disappear, let’s just hope it doesn’t change into something too crappy.
I’d wildguess that the thing that could disappear is the profile of today’s visual creatives who contribute to vfx, games, films, demoscene etc. The ‘arty’ profile. End up being replaced by the ‘SEO engineering’ profile. Go on LinkedIn and observe the profile of people that newly add ‘AI artist’ to their resume, look at their previous credentials.
Haha sorry sorry.
I know no art will actually disappear, let’s just hope it doesn’t change into something too crappy.
I’d wildguess that the thing that could disappear is the profile of today’s visual creatives who contribute to vfx, games, films, demoscene etc. The ‘arty’ profile. End up being replaced by the ‘SEO engineering’ profile. Go on LinkedIn and observe the profile of people that newly add ‘AI artist’ to their resume, look at their previous credentials.
these linked-in profiles and job applications are often full of buzz words and mumbo-jumbo. what people actually do when they are free-to-roam is much more interesting, IMHO.
and yes, human made art won't disappear. I have no doubt that at least some of the younger generation will still recognize its worth in the times ahead.
and yes, human made art won't disappear. I have no doubt that at least some of the younger generation will still recognize its worth in the times ahead.
@4gentE: Crappy art has always existed. But keep in mind that thanks to the Internet it now reaches everywhere faster. However, that trashy art is ephemeral (or at least we can hope so).
You are right that soon no one will add "AI artist" to their resume because it is too vague a term now and will be even more so in a few years.
We don't know what an AI artist can do or will be. Present days are to future AI art forms as the days of Lumière brothers were to the once unknown art of cinema.
You are right that soon no one will add "AI artist" to their resume because it is too vague a term now and will be even more so in a few years.
We don't know what an AI artist can do or will be. Present days are to future AI art forms as the days of Lumière brothers were to the once unknown art of cinema.
Perhaps we should try to nail down the real intelligence before we move on to the artificial stuff.
yep, garbage in, garbage out. solve your own issues before trying to force-feed a sixteenth-assed solution to the world. bad parenting, like I mentioned before..
(and now I am intoxicated and will bail out for today :D)
(and now I am intoxicated and will bail out for today :D)
Quote:
And as soon as there are, better be sure not to overtrain them with unrelated 6502 code, so keep those goddamn Apple II magazine type-in source snippets from Terminator (1984) out of the picture.There are no LLMs with enough training data to generate even a single c64 scroller.
Quote:
Hmm not so sure. Translator used to be a (possibly under-)paid job, and uh... "computer" used to mean actual people at some point, too.I mean, I find the tech utterly fascinating but also think it should be severely regulated -- and effectively banned in cases where it replaces already underpaid humans with robots or attempts to manipulate public opinion.
One of the greatest achievements that incorporates "AI" are translation tools. And best of all, they do no harm to anyone.
Banning some new tech because people lose their jobs doesn't seem like the right thing to do. Social market economy, fwiw, should be able to do better than that.
@neoman. I might have written it in previous topics or in another forum entirely, but with an organizers point of view in mind, I believe a solution for graphics (be old or newschool) would be for the organization to set some theme and generally conceptualize/contextualize the compos. This will prevent people from submitting random images and also add structural elements by which an entry will be judged that go beyond the blatant 'crafting' part.
Perhaps make an announcement, say, a couple of months prior to the actual party... So, a little bit like with the Evoke pixel compo, but with the context taken further from just the colour palette.
Perhaps make an announcement, say, a couple of months prior to the actual party... So, a little bit like with the Evoke pixel compo, but with the context taken further from just the colour palette.
Quote:
So you mean enough concept/context that it will be hard for AI jockeys to get something fitting with just one prompt?the organization to set some theme and generally conceptualize/contextualize the compos. This will prevent people from submitting random images
What bothers me about these posts that deploy the word "historically" is the fundamentally broken logic they seem to follow. What they imply goes something like this: Historically the black plague and the spanish flu didn't make humans extinct, therefore no virus that could do that can exist. Historically, the bow and arrow, the catapult and hand grenade didn't wreck the whole planet, therefore thermonuclear weapons won't. Historically, no asteroid destroyed the entire carbon based life on Earth, therefore no such asteroid can come our way. This logic is fundamentally broken.
Additionally, this thinking seems to work in a strictly linear fashion. Oblivious to the logic of tipping points.
The greatest minds of today are wrestling with this. They are thinking, feeling, discussing, watching, learning, adding, subtracting, predicting, warning, agonizing over it. But the "historically" people don't need all that, they don't need the data, the feedback, they KNOW. Because "historically". Like they are stuck in a pre-cybernetic system. To me this sounds more like a religious belief or an ancient wisdom than true thinking. Romantic.
One thing in common these posts exhibit is the authors need (conscious or not) to be and sound profound, enlightened. Many of them seem to be fond of wisdoms and quotes from the bumper stickers. You know, "what doesn't kill you makes you stronger". Sure. It's sometimes true. However, most of the times, what doesn't kill you, cripples you.
Additionally, this thinking seems to work in a strictly linear fashion. Oblivious to the logic of tipping points.
The greatest minds of today are wrestling with this. They are thinking, feeling, discussing, watching, learning, adding, subtracting, predicting, warning, agonizing over it. But the "historically" people don't need all that, they don't need the data, the feedback, they KNOW. Because "historically". Like they are stuck in a pre-cybernetic system. To me this sounds more like a religious belief or an ancient wisdom than true thinking. Romantic.
One thing in common these posts exhibit is the authors need (conscious or not) to be and sound profound, enlightened. Many of them seem to be fond of wisdoms and quotes from the bumper stickers. You know, "what doesn't kill you makes you stronger". Sure. It's sometimes true. However, most of the times, what doesn't kill you, cripples you.
"Come down, the pixel artists will not go extinct". Rejoice:
https://civitai.com/models/304097/sdxl-commodore-64-style-pixel-art
https://civitai.com/models/304097/sdxl-commodore-64-style-pixel-art
Quote:
And your posts are no exception there, my friend. (And no, i don't have any personal beef against you.)One thing in common these posts exhibit is the authors need (conscious or not) to be and sound profound, enlightened. Many of them seem to be fond of wisdoms and quotes from the bumper stickers.
Quote:
And no, i don't have any personal beef against you.
Good, I'm glad. Because I have this giant respect for you and your work man!
Now please look at the link in my last post, and tell me one thing: if you like what you see there, wouldn't one be forgiven to think that it seems like you have a beef with pixel artists in general?
Quote:
enough concept/context that it will be hard for AI jockeys
Pretty much, yes.
Would also affect people who don't care about the conceptual part though, and just want to submit a nicely done pic to a compo.
So there's going to be additional curation needed; I mean, you end up with, say, two pics, both nicely done, but one nails the concept while the other is more of a 'I added an Amiga ball to it'. What do you do? Do you take the latter out of the compo? Do you give additional (let's say) 'experts' points to the former so as to avoid crowd voting being the decisive factor?
Dunno, perhaps doing something like that sounds way too much for the demoscene. On the other hand perhaps it's a step up that will take things to new levels than just regulate AI/plagiarism.
Quote:
@neoman. I might have written it in previous topics or in another forum entirely, but with an organizers point of view in mind, I believe a solution for graphics (be old or newschool) would be for the organization to set some theme and generally conceptualize/contextualize the compos. This will prevent people from submitting random images and also add structural elements by which an entry will be judged that go beyond the blatant 'crafting' part.
I will suggest something like that in the team. SunSpire also suggested to split the voting in "skill" and "creativity" or something like that. But we really need to find a way to rate the craft otherwise the compos do not make sense if a big part has been created "automatically" or by entering prompts because then we can also make a separate AI compo where "Prompt engineers" can compete against each other! Also on our party many people are from the C64 scene where this kind of human craft seems to be more appreciated anyway so I hope we still get enough entries.
Quote:
So there's going to be additional curation needed; I mean, you end up with, say, two pics, both nicely done, but one nails the concept while the other is more of a 'I added an Amiga ball to it'. What do you do? Do you take the latter out of the compo? Do you give additional (let's say) 'experts' points to the former so as to avoid crowd voting being the decisive factor?
lol I know exactly which picture you have in mind with "I added an Amiga ball to it". I know those artists who do that also redithered a lot of images by Boris Vallejo but even then there's still a lot of manual work involved. 3D modellers also mostly work with reallife assets as templates for their models.