For so many ppl "good" art is something that "looks like the thing" and this study just shows that depressingly well
Also, it is more accurate to say that this AI art is really just a derivative work, it's standing on the shoulder of artists from the past and it's not that difficult to create "art" when all the hard work has been done for you
I already think about this tweet a lot, but your post (which is excellent) really brought it back to mind in a big way: "As a teacher of poetry what I can tell you for sure is people want poems to rhyme. They want poems to rhyme so bad. But we won’t give it to them" (https://x.com/ursulabrs/status/1434791291653558275)
which is funny bc it's nearly impossible in my experience to get an LLM to generate poetry that *doesn't* rhyme, even when specifically prompted, which makes it stand out from most modern poetry
I did Creative Writing in college and specialized in poetry. I can’t recall ever writing a poem that rhymed except for a specific lesson or two to challenge us. And we had an amazing instructor. People don’t like poetry much. They do like song lyrics though.
Yep. Poetry is as popular as ever, it's just enjoyed as rap or song lyrics, which, surprise surprise, highly prize rhyme. It's almost as though high-art types were abandoned by the people after they abandoned the people first by arbitrarily trying to "challenge" them instead of bringing goodness into their lives 🤔🤔🤔🤔🤔
To be fair, I don’t think rhyming is required. Rhythm and stress can do a lot. Not all song lyrics rhyme. Accessibility is the biggest thing. And the poetry we have coming out of universities is fine. We just don’t have the accessibility layer between the masses and highbrow.
In fiction, we have a range of genre novels and literary fiction. So there’s multiple layers people can access according to taste.
Distilled a lot of thoughts I've had about this issue quite well. The argument that AI is democratizing artistic expression might be technically true but it's increasingly obvious that this removal of skill barriers is solely for the benefit of people who seek to create the equivalent of a Dr. Who / Rick and Morty crossover tee-shirt. Maybe that's what people want, but it certainly doesn't justify the weird gloating many tech enthusiasts seem to be doing about human made art being "over".
great piece. i think many within Rationalists communities are very tied to their one analytical way of thinking, and can't wrap their minds around other non-analytical forms of knowledge production, and because they can't understand alternative models they don't believe the models are really saying anything. so these tests basically confirm what they've long suspected, that there is ultimately something "bullshit-ty" about "art". i deeply respect Scott Alexander, and i have many friends who i'd consider to be part of this rationalist community, but this is one of the huge blindspots of these folks
I think there's a lot to respect about people who can take an idea seriously instead of just dismissing it as absurd. Sometimes that ability backfires, but most of the time it's pretty useful.
There’s virtually no public understanding that art does not exist without context. You see this particularly clearly in the evergreen gripe about abstract art: my six-year-old son could do that.
AI art is the perfect expression of that most horrible word into which all creative output has been packed: content. Devoid of context and meaning, like a gray porridge, ready to be poured into a website template, social media post, or whatever other Internet vehicle is at hand. AI art is like sausage meat in that regard, thousands of years of human culture thrown into a grinder and extruded at the other end in regular, easy to eat, pre-cooked chunks.
"...So, in the conversation concerning art, as I said before, something is likely to get lost and as funny as it is, what mostly goes unsaid is the creative human behind the work. That is, the artist. How comical it is, and typical, that the initial human spark of the work is forgotten, ignored, unmentioned, and the vomit that is created, the soul’s mandatory upchuck, is now debated, deliberated, picked apart, thoroughly dissected, and in many cases to such an extent that it causes the works own demise in the eyes of the viewer as well as the eyes of the artist. This is why, I take it, some individuals have such an apathy (and sometimes celebration) towards Artificial Intelligence in art—they have no concern for the process or individual, only for the output. Only for the “thing” in its physical or visual form. Either way, we take too much time for the meaning of the piece, rather than the meaning of the process."
This! Oh my goodness this, all day, every day. Thank you. That it was created by a human, to be experienced by humans, is essential. It might be the only essential thing. Art is, to me at least, most fundamentally a form of human communication.
Alexander's test wasn't a test of AI art. It was a test of AI art THAT WAS SELECTED BY A HUMAN BEING! Use an AI image generator to try and produce a specific image and any sense that they're a good substitute without human curation will immediately evaporate. He's a classic AI triumphalist-in-doomer-clothes so of course he's leaving that out. The rest of us shouldn't.
I found this essay interesting because it associates a view that I have much sympathy with with proponents of AI art, the view that claims of artistic discernment are just exclusionary methods of signaling and that highbrow art is a naked emperor. This view seems to me to have a lot of truth to it, but at the same time I dislike AI art and find most criticisms of it to be valid.
I've never really understood arguments that "kitsch" is bad because it is "cloying," "sentimental," or "comforting." What's wrong with comforting people? What's wrong with invoking big feelings? I do understand "shallow," I guess, I do appreciate art with layers.
I think one of things I hate about modern art is also what I hate about AI art. AI art is made by something that doesn't know what it is making, it is just associating colors and shapes with keywords. It feels like someone is trying to communicate an idea to you, but they aren't. Modern art seems the same way, it is intentionally making art that doesn't look like anything in order to confuse the viewer. It gives the impression that the artist is jerking you around, they are pretending to communicate an idea but are really just confusing you with nonsense.
To me the best art is the kind that can do both, it looks superficially pleasing like "shallow" art, but it also has additional layers that are revealed on further inspection (for example, a smart action movie with deep characters). AI can't really do this yet, although a lot of humans can't either. (Another type of "modern art" I dislike is the sort that tries to do deeper layers without the superficial layers resulting in an ugly, boring piece of art that can be rewarding if you stick with it, except why would you?)
"Modern art" is a capacious term that includes such a wide variety of artists, schools, media, ideologies, time periods, etc., (including lots of figurative work, i.e. work that "looks like something") it seems hard to make any kind of definitive statement at all about it.
I don't know that there's anything wrong with being comforting (in the sense of "flattering") or sentimental, per se. But is that what we should want out of art? To me "cloying," "saccharine," "comforting" imply shallowness, superficiality, unoriginality, anyway.
Why should it not be among one of the things to want out of art? The best part of art is the human emotion behind it. I can appreciate the masterworks for the sheer skill that went into creating them, and the powerful ideas or emotions they attempt to convey.
However, I used to pass by a makeshift art market, where people would set up stands and sell their bland, unremarkable pieces. The kind that I'd see in some random hallway. Someone might hang those in their toilet, for all I know. I still appreciate that sort of art, because those people took the time to learn the craft, and their only aspiration was to make a comforting or pleasant image that someone else might enjoy.
Bob Ross' paintings are kind of like that too. Are they incredibly original, complex, or meaningful? No, not really. You could imbue it with extra meaning by pointing out that some of those were made in the process of teaching others to express their own creativity, but I don't think that's necessary. Those paintings, as unoriginal as they are, a product of a human being using the mind and body to create visuals of things that they find beautiful. I feel like to reduce that to something that is simply "shallow, superficial, and unoriginal" would be an affront to what art is. It's like a teenager lashing out against media that targets them because they're an adult now, and they like mature art movies that have important things to say.
| taking the safest and least surprising path at all times.
I agree with the 'least surprising' bit (e.g. the models are predictive and will take the median path), but I think the 'safest' bit is more a product of the large and lucid models being trained by large corporations with liabilities to worry about.
I took the test after reading this post, but before seeing any of the pictures other than the two you share here, and got 38 right vs 11 wrong (and 1 that I can't remember) - or 76% accurate. That may be because I enjoy looking at art, I've seen plenty of it, and the styles that I enjoy are the ones AI struggles with - scenes of human figures in action, abstract works with ambiguous meaning, etc. In line with your point about bad art, the ones I got wrong were mostly those images most similar to *bad* human art (of which there is plenty already!).
One reason I suspect a lot of people are easily fooled by AI is that AI art produces something to be quickly glanced over. This is exactly what the internet favours and promotes - frictionless 'content' that can be swiftly and easily digested in an endless scroll (or, indeed, fill up space in a dentist's waiting room). The best human art, by contrast, invites the viewer in and rewards them with precisely those details that, in an AI image, fall apart under scrutiny.
Your second footnote is really the key here: people don't understand that these things work by averaging and interpolating data - they think they work by the model literally scratching out little imitations of humans , because they've been given every impression that's how it works by it's boosters ('it learns and imitates, just like you or me!')
I'd be interested for your thoughts on how the dynamics at play here interact with the adult-YA-fan/poptimist school. Are they an advanced deployment fifth column by time-traveling Super AIs?
I think the adult YA fan/optimistic school does have a good point a lot of the time. People do have a tendency to confuse genre with artistic quality. To someone who bothers to look, YA books and pop art often display the same kind of depths as "high art."
You can take the same critical tools and apply them to anything to find hidden depths. I think the reason that "high" art lends itself to this process better is that it is often stripped of anything obviously appealing so it had no value except for hidden depths.
Note the homogenized aggressively bad taste displayed in ChatGPT poetry seems to be the result of RLHF (reinforcement learning from human feedback, i.e., hiring people to rate model output and heavily tuning the model on those ratings) - pre-"chat"/pre-RLHF GPT models that were true statistical imitations didn't sound like that at all. I guess you could say the study results are confirming that the RLHF does indeed give the people what they want.
It's interesting that Scott is kind of a philistine (outside his opinions on AI art, he said that Horatius [https://englishverse.com/poems/horatius] is better than any poem of the last 50 years) because he actually writes pretty funny and effective satire sometimes. I think the art appreciation muscle is just not very tightly connected to the other brain muscles, at least in some people.
For so many ppl "good" art is something that "looks like the thing" and this study just shows that depressingly well
Also, it is more accurate to say that this AI art is really just a derivative work, it's standing on the shoulder of artists from the past and it's not that difficult to create "art" when all the hard work has been done for you
There’s not that many creative geniuses around. Isn’t all art standing on the shoulders of giants
Yes but it still involves choices, AI art isn't making any choices it's using templates
100%
I already think about this tweet a lot, but your post (which is excellent) really brought it back to mind in a big way: "As a teacher of poetry what I can tell you for sure is people want poems to rhyme. They want poems to rhyme so bad. But we won’t give it to them" (https://x.com/ursulabrs/status/1434791291653558275)
which is funny bc it's nearly impossible in my experience to get an LLM to generate poetry that *doesn't* rhyme, even when specifically prompted, which makes it stand out from most modern poetry
Art elitists
Like to shit
On things people like
Because "feeling good is bad, actually."
Fuck them
And fuck their stupid rhymeless poetry.
If they can get away with that,
I can declare this poem
The best one ever written.
At this point,
Making something
That anyone would actually like
Would be the truly bold, challenging move
And anything less
Makes them utter hacks.
I will shit on their abstract paintings
Hear them praise my shit-splatters for challenging their noses
And wipe my ass with their poetry.
Fuck them.
I did Creative Writing in college and specialized in poetry. I can’t recall ever writing a poem that rhymed except for a specific lesson or two to challenge us. And we had an amazing instructor. People don’t like poetry much. They do like song lyrics though.
Yep. Poetry is as popular as ever, it's just enjoyed as rap or song lyrics, which, surprise surprise, highly prize rhyme. It's almost as though high-art types were abandoned by the people after they abandoned the people first by arbitrarily trying to "challenge" them instead of bringing goodness into their lives 🤔🤔🤔🤔🤔
he was a skater boy
she said 'see ya later boy'
To be fair, I don’t think rhyming is required. Rhythm and stress can do a lot. Not all song lyrics rhyme. Accessibility is the biggest thing. And the poetry we have coming out of universities is fine. We just don’t have the accessibility layer between the masses and highbrow.
In fiction, we have a range of genre novels and literary fiction. So there’s multiple layers people can access according to taste.
Which of course is the reason that poetry has, for all intents and purposes, died as a industry. Give the people what they want!
Distilled a lot of thoughts I've had about this issue quite well. The argument that AI is democratizing artistic expression might be technically true but it's increasingly obvious that this removal of skill barriers is solely for the benefit of people who seek to create the equivalent of a Dr. Who / Rick and Morty crossover tee-shirt. Maybe that's what people want, but it certainly doesn't justify the weird gloating many tech enthusiasts seem to be doing about human made art being "over".
It’s auto-generated fanfic, forever.
great piece. i think many within Rationalists communities are very tied to their one analytical way of thinking, and can't wrap their minds around other non-analytical forms of knowledge production, and because they can't understand alternative models they don't believe the models are really saying anything. so these tests basically confirm what they've long suspected, that there is ultimately something "bullshit-ty" about "art". i deeply respect Scott Alexander, and i have many friends who i'd consider to be part of this rationalist community, but this is one of the huge blindspots of these folks
Yep…I know the type well. I second this notion!
Why would you respect anyone from the community that shat themselves over an incredibly stupid thought experiment (look up “Roko’s Basilisk”)?
I think there's a lot to respect about people who can take an idea seriously instead of just dismissing it as absurd. Sometimes that ability backfires, but most of the time it's pretty useful.
Only when the idea isn’t built on a foundation of absurd cultish assumptions.
Are you mixing up Scott Alexander and Eliezer Yudkowsky?
I’m taking a dunk on the rationalist community as a whole. That’s why I said “community” and not “Scott Alexander”.
I mean the artistic community hasn’t always had flawless judgment either.
Implying there is such a thing as a coherent “artistic community”.
Fair enough. So what’s an intellectual school, or whatever you want to call them, with no embarrassing episodes?
oh boy the fact people liked that AI walt whitman poem is bleak. although the people do hate poetry, so that’s not super surprising
There’s virtually no public understanding that art does not exist without context. You see this particularly clearly in the evergreen gripe about abstract art: my six-year-old son could do that.
AI art is the perfect expression of that most horrible word into which all creative output has been packed: content. Devoid of context and meaning, like a gray porridge, ready to be poured into a website template, social media post, or whatever other Internet vehicle is at hand. AI art is like sausage meat in that regard, thousands of years of human culture thrown into a grinder and extruded at the other end in regular, easy to eat, pre-cooked chunks.
I should qualify “public understanding” as “compulsively-online public understanding.”
"...So, in the conversation concerning art, as I said before, something is likely to get lost and as funny as it is, what mostly goes unsaid is the creative human behind the work. That is, the artist. How comical it is, and typical, that the initial human spark of the work is forgotten, ignored, unmentioned, and the vomit that is created, the soul’s mandatory upchuck, is now debated, deliberated, picked apart, thoroughly dissected, and in many cases to such an extent that it causes the works own demise in the eyes of the viewer as well as the eyes of the artist. This is why, I take it, some individuals have such an apathy (and sometimes celebration) towards Artificial Intelligence in art—they have no concern for the process or individual, only for the output. Only for the “thing” in its physical or visual form. Either way, we take too much time for the meaning of the piece, rather than the meaning of the process."
--"Of Masterpieces and Scraps"
https://judsonvereen.substack.com/cp/150209507
At the risk of coming off as glib and immature, fuckin’ A.
This! Oh my goodness this, all day, every day. Thank you. That it was created by a human, to be experienced by humans, is essential. It might be the only essential thing. Art is, to me at least, most fundamentally a form of human communication.
Alexander's test wasn't a test of AI art. It was a test of AI art THAT WAS SELECTED BY A HUMAN BEING! Use an AI image generator to try and produce a specific image and any sense that they're a good substitute without human curation will immediately evaporate. He's a classic AI triumphalist-in-doomer-clothes so of course he's leaving that out. The rest of us shouldn't.
I found this essay interesting because it associates a view that I have much sympathy with with proponents of AI art, the view that claims of artistic discernment are just exclusionary methods of signaling and that highbrow art is a naked emperor. This view seems to me to have a lot of truth to it, but at the same time I dislike AI art and find most criticisms of it to be valid.
I've never really understood arguments that "kitsch" is bad because it is "cloying," "sentimental," or "comforting." What's wrong with comforting people? What's wrong with invoking big feelings? I do understand "shallow," I guess, I do appreciate art with layers.
I think one of things I hate about modern art is also what I hate about AI art. AI art is made by something that doesn't know what it is making, it is just associating colors and shapes with keywords. It feels like someone is trying to communicate an idea to you, but they aren't. Modern art seems the same way, it is intentionally making art that doesn't look like anything in order to confuse the viewer. It gives the impression that the artist is jerking you around, they are pretending to communicate an idea but are really just confusing you with nonsense.
To me the best art is the kind that can do both, it looks superficially pleasing like "shallow" art, but it also has additional layers that are revealed on further inspection (for example, a smart action movie with deep characters). AI can't really do this yet, although a lot of humans can't either. (Another type of "modern art" I dislike is the sort that tries to do deeper layers without the superficial layers resulting in an ugly, boring piece of art that can be rewarding if you stick with it, except why would you?)
"Modern art" is a capacious term that includes such a wide variety of artists, schools, media, ideologies, time periods, etc., (including lots of figurative work, i.e. work that "looks like something") it seems hard to make any kind of definitive statement at all about it.
I don't know that there's anything wrong with being comforting (in the sense of "flattering") or sentimental, per se. But is that what we should want out of art? To me "cloying," "saccharine," "comforting" imply shallowness, superficiality, unoriginality, anyway.
Why should it not be among one of the things to want out of art? The best part of art is the human emotion behind it. I can appreciate the masterworks for the sheer skill that went into creating them, and the powerful ideas or emotions they attempt to convey.
However, I used to pass by a makeshift art market, where people would set up stands and sell their bland, unremarkable pieces. The kind that I'd see in some random hallway. Someone might hang those in their toilet, for all I know. I still appreciate that sort of art, because those people took the time to learn the craft, and their only aspiration was to make a comforting or pleasant image that someone else might enjoy.
Bob Ross' paintings are kind of like that too. Are they incredibly original, complex, or meaningful? No, not really. You could imbue it with extra meaning by pointing out that some of those were made in the process of teaching others to express their own creativity, but I don't think that's necessary. Those paintings, as unoriginal as they are, a product of a human being using the mind and body to create visuals of things that they find beautiful. I feel like to reduce that to something that is simply "shallow, superficial, and unoriginal" would be an affront to what art is. It's like a teenager lashing out against media that targets them because they're an adult now, and they like mature art movies that have important things to say.
Art elitists
Like to shit
On things people like
Because "feeling good is bad, actually."
Fuck them
And fuck their stupid rhymeless poetry.
If they can get away with that,
I can declare this poem
The best one ever written.
At this point,
Making something
That anyone would actually like
Would be the truly bold, challenging move
And anything less
Makes them utter hacks.
I will shit on their abstract paintings
Hear them praise my shit-splatters for challenging their noses
And wipe my ass with their poetry.
Fuck them.
Fuck you.
| taking the safest and least surprising path at all times.
I agree with the 'least surprising' bit (e.g. the models are predictive and will take the median path), but I think the 'safest' bit is more a product of the large and lucid models being trained by large corporations with liabilities to worry about.
I enjoyed this Read Max very much, and I'd like to think I liked it more than I would a similar AI-generated blog!!
I took the test after reading this post, but before seeing any of the pictures other than the two you share here, and got 38 right vs 11 wrong (and 1 that I can't remember) - or 76% accurate. That may be because I enjoy looking at art, I've seen plenty of it, and the styles that I enjoy are the ones AI struggles with - scenes of human figures in action, abstract works with ambiguous meaning, etc. In line with your point about bad art, the ones I got wrong were mostly those images most similar to *bad* human art (of which there is plenty already!).
One reason I suspect a lot of people are easily fooled by AI is that AI art produces something to be quickly glanced over. This is exactly what the internet favours and promotes - frictionless 'content' that can be swiftly and easily digested in an endless scroll (or, indeed, fill up space in a dentist's waiting room). The best human art, by contrast, invites the viewer in and rewards them with precisely those details that, in an AI image, fall apart under scrutiny.
Your second footnote is really the key here: people don't understand that these things work by averaging and interpolating data - they think they work by the model literally scratching out little imitations of humans , because they've been given every impression that's how it works by it's boosters ('it learns and imitates, just like you or me!')
I'd be interested for your thoughts on how the dynamics at play here interact with the adult-YA-fan/poptimist school. Are they an advanced deployment fifth column by time-traveling Super AIs?
I think the adult YA fan/optimistic school does have a good point a lot of the time. People do have a tendency to confuse genre with artistic quality. To someone who bothers to look, YA books and pop art often display the same kind of depths as "high art."
You can take the same critical tools and apply them to anything to find hidden depths. I think the reason that "high" art lends itself to this process better is that it is often stripped of anything obviously appealing so it had no value except for hidden depths.
Note the homogenized aggressively bad taste displayed in ChatGPT poetry seems to be the result of RLHF (reinforcement learning from human feedback, i.e., hiring people to rate model output and heavily tuning the model on those ratings) - pre-"chat"/pre-RLHF GPT models that were true statistical imitations didn't sound like that at all. I guess you could say the study results are confirming that the RLHF does indeed give the people what they want.
It's interesting that Scott is kind of a philistine (outside his opinions on AI art, he said that Horatius [https://englishverse.com/poems/horatius] is better than any poem of the last 50 years) because he actually writes pretty funny and effective satire sometimes. I think the art appreciation muscle is just not very tightly connected to the other brain muscles, at least in some people.