Really interesting to learn that even aside from AI, there's evidence many people prefer amateur poetry to expert poetry!! I've sometimes heard people say "LLMs are good for bad artists and bad (or neutral) for good artists"—this all feels consistent with your take, and the results of the quiz/others like it.
"I think part of the problem with AI art is that it produces stuff non-artists think look good but which on close inspection looks terrible, and so it ends up turning search results that used to be good into sifting through terrible stuff. Imagine if everyone got the ability to create mostly nutritional adequate meals for like five cents, but they all were mediocre rehydrated powder with way too much sucralose or artificial grape flavor or such. And your friends start inviting you over to dinner parties way more often because it's so easy to deal with food now, but practically every time, they serve you sucralose protein shake."
"When real pictures have details, the details have logic to them. I think of Ancient Gate being in the genre 'superficially detailed, but all the details are bad and incoherent'. The red and blue paint and blank stone feel like they're supposed to evoke worn-ness, but it's not clear what style this is supposed to be a worn-down version of. One gets the feeling that if all the paint were present it would look like a pile of shipping containers, if shipping containers were only made in two colors."
I also saw a couple of issues with the art included. In the first example, the painting on the left is an imitation of Renoir, a much better known impressionist than Gauguin. The Gauguin included is not his most known work. Some AI art included imitations of famous artists such as Van Gogh. For untrained people, it looked like an original but the real reason people picked it was exposure to the artist’s work. There are Van Gogh replicas everywhere on everyday objects. The analysis of the artist who was much better than the vast majority at picking AI from human art proves the point that training is necessary to appreciate art. Untrained people’s appreciation of art is limited by their exposure to good art. More, Renoir is more aesthetically pleasing to an untrained eye with his more naturalistic style and colour palette than Gauguin. Even a trained person might prefer Renoir to Gauguin.
This is a bit of a silly retort, I think, because your main point is basically "sure, AI can generate images that are more-or-less identical in form to human art, but the content of these images is so much worse!" Which... yeah, sure! But that's a silly objection, because there's nothing preventing someone with knowledge of composition, art history, etc, from generating an image that *is* on par with the best of human art, with carefully chosen details.
(*Why* no one with knowledge of composition, art history, etc, has gone and done that is maybe a more interesting question - but there's no technical reason some Google engineer with a sharp eye and Adobe Firefly access couldn't make a tech-world version of that Ingres painting of Homer and his influences by stitching together text prompts.)
You've misunderstood my argument a bit. I think a human making a pastiche imitation, or pure rip-off, of a painting that was innovative 150 years ago would also not be that impressive. Plenty of people can do that today. The claims by SA that we should judge art by pure impulse reaction without understanding history, context, composition method, and other factors is, in my view, silly. And mere imitation isn't really "creativity."
To your point, it becomes a bit of a philosophical question of when a human is doing so much of the work that it becomes human made art. SA picked digital images made with photoshop or similar programs for "human made." I certainly think one can use AI in a similar way. Where the line is though is up for debate I'm sure.
I think we agree here - there's definitely crappy AI "art" with no creative spark to it, but there's definitely art that was made using AI but with sufficient deliberation as to be human-made, and we don't really know how to distinguish between the two. The guy from this Vox video I feel comfortable suggesting is doing real creative work: https://youtu.be/K0ldxCh3cnI ... but I don't think the average AI ""artist"" is doing that much hands-on work!
(I do, however, object to the scare quotes here when describing digital art! I don't care much for Artstation Wacomtabletcore waifu pictures either, but the people who draw those anime girls for a paycheck are the ones who are going to be losing their jobs to Dall-E, not Van Gogh.)
Yes I think we probably don't disagree much if we teased this out! I didn't really mean "human made" as scare quotes, just as the category SA used. I don't think digitally made art is inherently bad or anything!
I think you're being too hard on pastiche, is the thing! I could see someone thinking that, e.g., Kehinde Wiley is an "uncreative" artist or "unimpressive" artist, but... that's not a majority opinion! And it definitely would be odd to say that there are plenty of people with the technical chops to make, e.g., an almost exact reproduction of a Jacques-Louis David painting except the subject has dreads and Timbs. That's just one example, but there many, many respected artists whose works are even more overt about being imitations. (This is, e.g., Elaine Sturtevant's whole deal!)
Definitely the boundaries of these things are very blurry. Ignoring AI, where are the exact borders between straight plagiarism/copying, ripping off, uncreative pastiche, creative pastiche, and original style? No single answer, surely. AI adds a whole nother layer of confusion since sometimes AI truly copies an image (like the Joker example I posted) and other times it is remixing a bunch of work into something new if generic (the impressionist AI paintings in the quiz).
I don't think pastiche is bad at all. But I do think you could do this exact quiz with human artists, instead of AI, doing imitations or copies of impressionist paintings and most regular people--myself included!--probably could not always tell the difference. To me, these AI images show AI is close to imitating digital images in styles they've been trained on. That's very impressive, on a technical level. I don't think it shows "creativity" nor do I think regular people being unable to do better than 60-40 at distinguishing old innovative paintings from modern imitations (human or ai!) means much.
The ability of humans or AI to imitate old styles definitely depends somewhat on the style. Wiley paints in a technically difficult style. Impressionism is easier to imitate. And something like Rothko of course would be very easy to imitate, especially in small jpg form. Obviously his canvases are more impressive in their giant, real form.
Actually it would be pretty impressive if the replica is done with high fidelity. Being able to reproduce a Renoir or Gauguin means you’re a highly skilled artist.
It would be impressive--although that's not what is shown here, since these are small PNG files not full canvases--but it would not mean you have proved creativity or talent on par with the person you are imitating surely. In general, we tend to recognize a difference between technical skill and creative skill. A semi-talented guitarist can play most great rock songs, if not the most skilled metal guitar solos, but that's far different from composing those songs.
To be able to reproduce a small png of a semi-Gauguin requires billions of dollars, billions of training inputs, and many man hours of human interventions that power these genAI machines. Definitely easier for a human to achieve vs. current genAI tech.
I don't think humans and computers are really comparable on a 1 to 1 level here, but if you're going to do that you have to count the training hours and money spent on building these AIs right? The richest companies on earth have poured endless resources into training and building them.
Really interesting to learn that even aside from AI, there's evidence many people prefer amateur poetry to expert poetry!! I've sometimes heard people say "LLMs are good for bad artists and bad (or neutral) for good artists"—this all feels consistent with your take, and the results of the quiz/others like it.
For what it's worth, Scott Alexander published a followup that (somewhat) agrees with you. I especially enjoyed the quotes he includes from a friend who makes digital art: https://www.astralcodexten.com/i/151145038/but-others-might-genuinely-be-on-a-higher-plane-than-the-rest-of-us
"I think part of the problem with AI art is that it produces stuff non-artists think look good but which on close inspection looks terrible, and so it ends up turning search results that used to be good into sifting through terrible stuff. Imagine if everyone got the ability to create mostly nutritional adequate meals for like five cents, but they all were mediocre rehydrated powder with way too much sucralose or artificial grape flavor or such. And your friends start inviting you over to dinner parties way more often because it's so easy to deal with food now, but practically every time, they serve you sucralose protein shake."
Agree that was the most interesting comment. A nice analogy too.
"When real pictures have details, the details have logic to them. I think of Ancient Gate being in the genre 'superficially detailed, but all the details are bad and incoherent'. The red and blue paint and blank stone feel like they're supposed to evoke worn-ness, but it's not clear what style this is supposed to be a worn-down version of. One gets the feeling that if all the paint were present it would look like a pile of shipping containers, if shipping containers were only made in two colors."
What a fascinating exploration on creativity through this AI versus human art test. Thanks so much for this.
It takes personality to produce style. When computers get personalities, run for the hills : )
I also saw a couple of issues with the art included. In the first example, the painting on the left is an imitation of Renoir, a much better known impressionist than Gauguin. The Gauguin included is not his most known work. Some AI art included imitations of famous artists such as Van Gogh. For untrained people, it looked like an original but the real reason people picked it was exposure to the artist’s work. There are Van Gogh replicas everywhere on everyday objects. The analysis of the artist who was much better than the vast majority at picking AI from human art proves the point that training is necessary to appreciate art. Untrained people’s appreciation of art is limited by their exposure to good art. More, Renoir is more aesthetically pleasing to an untrained eye with his more naturalistic style and colour palette than Gauguin. Even a trained person might prefer Renoir to Gauguin.
This is a bit of a silly retort, I think, because your main point is basically "sure, AI can generate images that are more-or-less identical in form to human art, but the content of these images is so much worse!" Which... yeah, sure! But that's a silly objection, because there's nothing preventing someone with knowledge of composition, art history, etc, from generating an image that *is* on par with the best of human art, with carefully chosen details.
(*Why* no one with knowledge of composition, art history, etc, has gone and done that is maybe a more interesting question - but there's no technical reason some Google engineer with a sharp eye and Adobe Firefly access couldn't make a tech-world version of that Ingres painting of Homer and his influences by stitching together text prompts.)
You've misunderstood my argument a bit. I think a human making a pastiche imitation, or pure rip-off, of a painting that was innovative 150 years ago would also not be that impressive. Plenty of people can do that today. The claims by SA that we should judge art by pure impulse reaction without understanding history, context, composition method, and other factors is, in my view, silly. And mere imitation isn't really "creativity."
To your point, it becomes a bit of a philosophical question of when a human is doing so much of the work that it becomes human made art. SA picked digital images made with photoshop or similar programs for "human made." I certainly think one can use AI in a similar way. Where the line is though is up for debate I'm sure.
I think we agree here - there's definitely crappy AI "art" with no creative spark to it, but there's definitely art that was made using AI but with sufficient deliberation as to be human-made, and we don't really know how to distinguish between the two. The guy from this Vox video I feel comfortable suggesting is doing real creative work: https://youtu.be/K0ldxCh3cnI ... but I don't think the average AI ""artist"" is doing that much hands-on work!
(I do, however, object to the scare quotes here when describing digital art! I don't care much for Artstation Wacomtabletcore waifu pictures either, but the people who draw those anime girls for a paycheck are the ones who are going to be losing their jobs to Dall-E, not Van Gogh.)
Yes I think we probably don't disagree much if we teased this out! I didn't really mean "human made" as scare quotes, just as the category SA used. I don't think digitally made art is inherently bad or anything!
I think you're being too hard on pastiche, is the thing! I could see someone thinking that, e.g., Kehinde Wiley is an "uncreative" artist or "unimpressive" artist, but... that's not a majority opinion! And it definitely would be odd to say that there are plenty of people with the technical chops to make, e.g., an almost exact reproduction of a Jacques-Louis David painting except the subject has dreads and Timbs. That's just one example, but there many, many respected artists whose works are even more overt about being imitations. (This is, e.g., Elaine Sturtevant's whole deal!)
Definitely the boundaries of these things are very blurry. Ignoring AI, where are the exact borders between straight plagiarism/copying, ripping off, uncreative pastiche, creative pastiche, and original style? No single answer, surely. AI adds a whole nother layer of confusion since sometimes AI truly copies an image (like the Joker example I posted) and other times it is remixing a bunch of work into something new if generic (the impressionist AI paintings in the quiz).
I don't think pastiche is bad at all. But I do think you could do this exact quiz with human artists, instead of AI, doing imitations or copies of impressionist paintings and most regular people--myself included!--probably could not always tell the difference. To me, these AI images show AI is close to imitating digital images in styles they've been trained on. That's very impressive, on a technical level. I don't think it shows "creativity" nor do I think regular people being unable to do better than 60-40 at distinguishing old innovative paintings from modern imitations (human or ai!) means much.
The ability of humans or AI to imitate old styles definitely depends somewhat on the style. Wiley paints in a technically difficult style. Impressionism is easier to imitate. And something like Rothko of course would be very easy to imitate, especially in small jpg form. Obviously his canvases are more impressive in their giant, real form.
Actually it would be pretty impressive if the replica is done with high fidelity. Being able to reproduce a Renoir or Gauguin means you’re a highly skilled artist.
It would be impressive--although that's not what is shown here, since these are small PNG files not full canvases--but it would not mean you have proved creativity or talent on par with the person you are imitating surely. In general, we tend to recognize a difference between technical skill and creative skill. A semi-talented guitarist can play most great rock songs, if not the most skilled metal guitar solos, but that's far different from composing those songs.
To be able to reproduce a Gauguin as a human artist takes years of training. And even so, not everyone can do it.
To be able to reproduce a small png of a semi-Gauguin requires billions of dollars, billions of training inputs, and many man hours of human interventions that power these genAI machines. Definitely easier for a human to achieve vs. current genAI tech.
I don't think humans and computers are really comparable on a 1 to 1 level here, but if you're going to do that you have to count the training hours and money spent on building these AIs right? The richest companies on earth have poured endless resources into training and building them.