Comment on Google Serves AI Slop as Top Result for One of the Most Famous Paintings in History
LainTrain@lemmy.dbzer0.com 1 month agoMeh. Art history - sure! Books, writing, films, music.
Paintings though are a bit more dodgy. Most of the ‘classical’ paintings (in the renaissance and baroque periods) were created through a system of patronage by monarchs, if not for a selfish ego-boost of endless royal portraiture, then as a strategic move to enhance the prestige of the monarchy to it’s neighbours.
Of course, music is often guilty of this too. Bach’s famous struggles with securing patronage for instance, but due to its more abstract nature, it is probably a better source of information about the author than a painting about the painter.
It will hardly inform you of what the people or the world was like at the time, but rather what the monarchs wanted to project, which of course is helpful too for understanding the political situation at the time, but it’s hardly an efficient way to acquire that information, nor do the ideas gained from such exploration likely to lead to concrete conclusions.
Kind of like judging a people by their government’s propaganda department commissions.
What is interesting though is the fact that AI art, and the LAION-5B dataset used to train the models is a true and earnest reflection of sorts of what images today really are, from works of art to private commissions for furries etc to plain stock photos and other corporate graphics.
It’s an interesting reality check to compare the kinds of images your mind conjures connected to an idea and the kinds of images stable diffusion produces instead, it’s revealing of one’s biases in a very unique way.
kat_angstrom@lemmy.world 1 month ago
While your points about the patronage system and its weaknesses are valid, you’re writing off several centuries worth of legitimate human endeavor because the systems that enabled it were dodgy. Guess what though? That’s literally all of history, dodgy AF, featuring an intrepid cast of characters more awful the deeper you look. That doesn’t make the art or music worth writing off though.
But honestly:
“Earnest” is definitely not the word you’re looking for. Derivative, maybe, because you said it yourself, they’re reflections; and as such, they’re going to reflect what images of today are; like you said. That makes them derivative, and I feel a vast artificiality that makes my heart sink when I look at the vast majority of them.
Choosing machine-created art over historical art is choosing a passing fad over centuries of culture. It’s your right; but to write off history with a wave of the hand means you’re missing out on truly expanding your horizons.
LainTrain@lemmy.dbzer0.com 1 month ago
Not at all, but as an accurate representation of who felt what? Yeah it’s not the best source. Music is also specifically less so, because it’s too abstract to really be propagandistic apart from the vague aesthetic of grandiosity.
It very much is, actually.
Earnest [adjective] - resulting from or showing sincere and intense conviction.
A sincere intense conviction in gathering a generic dataset representative of the very broad concept of “images” in order to be the source from which future researchers can train a diffuser model as a proof of concept, to create patterns represented in those images out of randomised noise, driven by scientific pursuit and uncontaminated with the subjectivity of artistic taste is about as fitting for “earnest” as can be.
Remember, I wasn’t talking about only the outputs of any given model, but the dataset itself.
Yeah, the outputs definitely are, but derivative of an earnest representation of us. That makes it an interesting and unbiased account of what our images really are like.
It’s interesting to go on SD and consider an idea, a concept, and what image it conjures in your head, then proompt and see what kind of images it conjures from the model. The difference is the difference in bias. It’s an interesting reality check.
So is most human works. If anything, the randomness of noise, even by processor’s famously incapable of any such thing, is going to be far more unpredictable than the copying done by humans.
In itself though I don’t think that inherently makes either less valuable, all originality only exists as both an evolution of and in contrast to the established and accepted, and that is achieved by derivative works, perhaps even creating a genre.
If anything, part of the problem is that AI art is too original, sometimes inventing 6 fingers, or 7, the form is broken, and the idea of any image no longer resonates.
Personally I don’t. I don’t think there’s anything that makes them any more artificial than any human work.
Ultimately all are a human vision - an image generator is just a lot of fancy matrices in a file without human input, neither it nor Photoshop can make everything by themselves.
All are ultimately .jpegs, products not of some singular vision but also of the tools developed and available to the human, all are concepts so far removed from nature, labeling one artificial but not the other is splitting hairs on a head freshly and cleanly shaved by a precision engineered mass manufactured machine shipped half-way across the world in system so complex most people don’t understand it.
Oh come on now. You can hate AI without resorting to delusion or ignorance.
Not only is AI art on the rise, but even a year or two ago the tools for generating images were good enough that it’s already seeing noticeable widespread use, including in the physical world. And that’s not even touching on the strides and accomplishments in the LLM space.
I’m not writing anything off as is hopefully evident by my writing here, it’s moreso that the value of it as I see it is perhaps overestimated by yourself.