I’ve read your essay and find myself agreeing with much of its anatomy of “neck vs necklace,” yet still questioning the conclusion you draw from it. You’re right that great art needs a thick, particular context, and your Guernica-in-the-woods thought-experiment makes the point vividly. But the essay treats generative systems as if they stand alone, when in fact every non-trivial AI artwork already arrives encased in multiple layers of human meaning: prompt, curation, framing, display, reception.
You also grant, late in the piece, that “someone will eventually create an AI masterpiece,” yet most of the argument reads as though this were impossible in principle. Photography offers a useful timeline check: in 1841, barely two years after Daguerre’s first public demonstration, it would have been premature to pronounce the absence of photographic masterpieces; Stieglitz’s The Steerage and Atget’s Paris were still half a century away. Declaring a deficit now, when generative tools have been publicly available for scarcely three years, risks repeating the same snap verdict critics made about early “mechanical” cameras.
Finally, the essay often equates “AI art” with raw, unedited machine output, then measures that output against canonical paintings. Yet the most serious practitioners work exactly where you say meaning is made: they choose or even shoot the training data, design prompts as scores, iterate, crop, print, mount, title and tell stories around the images. In other words, they restore the very specificity you claim is missing.
So I share your intuition that the first undisputed AI masterpiece will arise when an artist harnesses the medium’s native properties like stochasticity, scale, recursion rather than using it for mimicry. I just doubt we’ll recognise that moment if we decide in advance that no such work can yet exist. The history of new media suggests we’re still far too close to the invention to make that call.
It may exist already!! Actually, Celine Nguyen mentions a few such artworks that could be considered masterpieces. The part I didn't mention (because I didn't want to enter into that) is that the label "masterpiece" is hard to define and only attributed in hindsight or in comparison to things it can be compared to (e.g. Ulysses is a masterpiece in comparison to other novels so it can be described as such immediately but I'm not sure that's the case for any AI artwork yet). In short, I agree (the headline and the last sentence are to spark debate rather than to firmly assert that AI isn't ready yet to be the means for a masterpiece)
Whether or not there are any AI masterpieces, the important point you are making was the basis for philosopher Bert Dreyfus's objection to Minsky's representational approach to designing computer intelligence. Drawing on Heidegger's concept of being-in-the-world, Dreyfus argued that humans have a vast taken-for-granted understanding of the world a background that cannot be "programmed" into the computer. Every human (well, the vast majority) knows what people have necks in the first place, and that necks happen to be useful for hanging necklaces. This tacit background understanding of things is generated because things MATTER TO HUMAN BEINGS in ways that they do not to computers. Something is at stake for us. Heidegger even said that the very Being of the human is caring. We care about ourselves, Others, the future, and so on. Designing a robot to learn how to navigate in a complex setting (with layers of tacitly understood handy/necessary features) is a step in the right direction, because the robot would have to sense its surroundings (feel, see, hear, etc.), but the robot's Being is not care. Robots can be taught to SEEM to care, but what would it take to modify a robot to such an extent that it would care about itself, its future, other robots, people, whatever in ways somehow analogous to our own? Such robots would have to become human variants. Is that likely?
So true. Dreyfus's insights were the first influence on my views that AIs and robots without embodiment and social context and without living in the world are missing a fundamental piece that we humans take for granted. Intelligence or whatever is not enough.
I think there is a very subtle mistake in the concept of "a technique for producing aleatory artworks."
First, the choice of the term "aleatory" over its synonym "random" has information. The term's obscurity suggests unearned gravitas. Generative AI is random, with a clever corrective algorithm, and we should just call it random without the unearned pomp of a Latin term for gambling.
More importantly, it elides the distinctions in non-conscious creation. Automatic writing, Automatic drawing, Frottage, Grattage, Decalcomania, and other techniques are defensible methods of unlocking creativity without consciousness. I believe that AI folks are suggesting that generative art is a similar form of consciousness-less creation, but it is different.
The artistic approaches I listed above derive their "meaning" or "art-ness" from either organic flow (including the organic variation of evolution) or subconscious flow. They are not random; they are un- or non-conscious. Generative AI art lacks that, as it derives from a static pool, with the only interactive element being the prompt itself. This is insufficient.
Art is created in the collision of flow, between a writer and the page, the reader and the text, the organic form and its ecosystem, or the subconscious and its gnarly sources. These are systems that are independent mashing together. For the AI to offer that, it has to be more than an unbelievably complicated random generator from a static model.
All (human) art is by definition ‘derivative’ and appropriates from previous Art.
The History of art is "Iterative appropriation' or as Picasso has said much better "Good artists copy, great artists steal" (although that quote is often attributed to Steve Jobs 'appropriating' Picasso. Jobs was a great thief ;-)
Brilliantly written article! Bravo! Perfectly articulated lots of comments I've had on the topic.
I think part of the reason we haven't had a masterpiece of art is because art is inherently political. These AI models are neutered by the labs they come out of to minimize any transgressive political opinions to ensure they don't get heavily legislated. Even when I talk to these AI models, I feel like they have no personality, because they have no needs, no desires, no motivation.
Of course, that probably explains why the large models haven't produced anything subversive, but what about fine-tuned models? Surely there are people who can fine-tune models without guard rails to adopt a unique personality. Then the user will prompt it to create polarizing art.
I feel like with all art it's a negotiation between the creator and the viewer. When an AI model creates an image, we don't think of the prompt, we say "ChatGPT" made it. So the type of art the model can create is only meta-commentary on AI. When models advance to the point where the art can be a seamless extension of the prompter, will we be able to create more personally provoking art.
Here's my opinion, using art as an example: the probability that AI has *not* created a 'masterpiece'* is incredibly low. The issue is that as soon as people know it was AI generated, there's an immediate repulsiveness to it. We know from SSC's blog that when people are blinded they cannot tell what was human-made versus AI-made, and actually they preferred the AI-made overall.
SSC's survey gives us good objective data that people like AI-created art, they just don't like knowing it.
I mean, I agree there could be an AI masterpiece somewhere right now. I'm just saying Scott's post is not evidence for that in any way (it's evidence that people have no taste, though, and that they don't know how to spot AI. I got almost 90% correct)
What do you mean? I'm not following. I got 90% correct and I didn't like very much those AI made. They were mostly obvious too me through taste rather than pattern-matching
I remember complaining to an artist about the "artwork" that I once encountered on the floor of a museum. It was just a piece of cable. That was all. "How can that possibly art?" I said.
The artist drily countered, "But apparently, it's the only piece you're still talking about."
I think it's the fact that the AI artist has no ego, no stake in the game, you can't hurt it's feelings. it'll just keep saying "sorry sorry..." It has no balls. Moreover. Art is never called a masterpiece until after the artist dies. AI can't die
I’ve read your essay and find myself agreeing with much of its anatomy of “neck vs necklace,” yet still questioning the conclusion you draw from it. You’re right that great art needs a thick, particular context, and your Guernica-in-the-woods thought-experiment makes the point vividly. But the essay treats generative systems as if they stand alone, when in fact every non-trivial AI artwork already arrives encased in multiple layers of human meaning: prompt, curation, framing, display, reception.
You also grant, late in the piece, that “someone will eventually create an AI masterpiece,” yet most of the argument reads as though this were impossible in principle. Photography offers a useful timeline check: in 1841, barely two years after Daguerre’s first public demonstration, it would have been premature to pronounce the absence of photographic masterpieces; Stieglitz’s The Steerage and Atget’s Paris were still half a century away. Declaring a deficit now, when generative tools have been publicly available for scarcely three years, risks repeating the same snap verdict critics made about early “mechanical” cameras.
Finally, the essay often equates “AI art” with raw, unedited machine output, then measures that output against canonical paintings. Yet the most serious practitioners work exactly where you say meaning is made: they choose or even shoot the training data, design prompts as scores, iterate, crop, print, mount, title and tell stories around the images. In other words, they restore the very specificity you claim is missing.
So I share your intuition that the first undisputed AI masterpiece will arise when an artist harnesses the medium’s native properties like stochasticity, scale, recursion rather than using it for mimicry. I just doubt we’ll recognise that moment if we decide in advance that no such work can yet exist. The history of new media suggests we’re still far too close to the invention to make that call.
It may exist already!! Actually, Celine Nguyen mentions a few such artworks that could be considered masterpieces. The part I didn't mention (because I didn't want to enter into that) is that the label "masterpiece" is hard to define and only attributed in hindsight or in comparison to things it can be compared to (e.g. Ulysses is a masterpiece in comparison to other novels so it can be described as such immediately but I'm not sure that's the case for any AI artwork yet). In short, I agree (the headline and the last sentence are to spark debate rather than to firmly assert that AI isn't ready yet to be the means for a masterpiece)
I agree!
Whether or not there are any AI masterpieces, the important point you are making was the basis for philosopher Bert Dreyfus's objection to Minsky's representational approach to designing computer intelligence. Drawing on Heidegger's concept of being-in-the-world, Dreyfus argued that humans have a vast taken-for-granted understanding of the world a background that cannot be "programmed" into the computer. Every human (well, the vast majority) knows what people have necks in the first place, and that necks happen to be useful for hanging necklaces. This tacit background understanding of things is generated because things MATTER TO HUMAN BEINGS in ways that they do not to computers. Something is at stake for us. Heidegger even said that the very Being of the human is caring. We care about ourselves, Others, the future, and so on. Designing a robot to learn how to navigate in a complex setting (with layers of tacitly understood handy/necessary features) is a step in the right direction, because the robot would have to sense its surroundings (feel, see, hear, etc.), but the robot's Being is not care. Robots can be taught to SEEM to care, but what would it take to modify a robot to such an extent that it would care about itself, its future, other robots, people, whatever in ways somehow analogous to our own? Such robots would have to become human variants. Is that likely?
So true. Dreyfus's insights were the first influence on my views that AIs and robots without embodiment and social context and without living in the world are missing a fundamental piece that we humans take for granted. Intelligence or whatever is not enough.
Very happy you are covering what Dwarkesh said! It has been on my mind since then. Ok, gonna read the whole thing now.
Will publish on Google's AlphaEvolve tomorrow as well
I think there is a very subtle mistake in the concept of "a technique for producing aleatory artworks."
First, the choice of the term "aleatory" over its synonym "random" has information. The term's obscurity suggests unearned gravitas. Generative AI is random, with a clever corrective algorithm, and we should just call it random without the unearned pomp of a Latin term for gambling.
More importantly, it elides the distinctions in non-conscious creation. Automatic writing, Automatic drawing, Frottage, Grattage, Decalcomania, and other techniques are defensible methods of unlocking creativity without consciousness. I believe that AI folks are suggesting that generative art is a similar form of consciousness-less creation, but it is different.
The artistic approaches I listed above derive their "meaning" or "art-ness" from either organic flow (including the organic variation of evolution) or subconscious flow. They are not random; they are un- or non-conscious. Generative AI art lacks that, as it derives from a static pool, with the only interactive element being the prompt itself. This is insufficient.
Art is created in the collision of flow, between a writer and the page, the reader and the text, the organic form and its ecosystem, or the subconscious and its gnarly sources. These are systems that are independent mashing together. For the AI to offer that, it has to be more than an unbelievably complicated random generator from a static model.
The following link to attap.ai (All Things to All People)
https://attap.ai/chat/s/3fcf1830-1d1b-4649-ab6d-7d166932396b/
All (human) art is by definition ‘derivative’ and appropriates from previous Art.
The History of art is "Iterative appropriation' or as Picasso has said much better "Good artists copy, great artists steal" (although that quote is often attributed to Steve Jobs 'appropriating' Picasso. Jobs was a great thief ;-)
please give https://attap.ai/chat/7/genarts/ a spin, it can be quite fun.
Brilliantly written article! Bravo! Perfectly articulated lots of comments I've had on the topic.
I think part of the reason we haven't had a masterpiece of art is because art is inherently political. These AI models are neutered by the labs they come out of to minimize any transgressive political opinions to ensure they don't get heavily legislated. Even when I talk to these AI models, I feel like they have no personality, because they have no needs, no desires, no motivation.
Of course, that probably explains why the large models haven't produced anything subversive, but what about fine-tuned models? Surely there are people who can fine-tune models without guard rails to adopt a unique personality. Then the user will prompt it to create polarizing art.
I feel like with all art it's a negotiation between the creator and the viewer. When an AI model creates an image, we don't think of the prompt, we say "ChatGPT" made it. So the type of art the model can create is only meta-commentary on AI. When models advance to the point where the art can be a seamless extension of the prompter, will we be able to create more personally provoking art.
Here's my opinion, using art as an example: the probability that AI has *not* created a 'masterpiece'* is incredibly low. The issue is that as soon as people know it was AI generated, there's an immediate repulsiveness to it. We know from SSC's blog that when people are blinded they cannot tell what was human-made versus AI-made, and actually they preferred the AI-made overall.
SSC's survey gives us good objective data that people like AI-created art, they just don't like knowing it.
*this is also subjective and not measurable.
liking something =/= considering it a masterpiece
They weren't asked, how would we know?
Again, whether something is a masterpiece or not is subjective.
Not to mention that masterpieces are almost by default exclusively historical pieces and/or from famous artists.
With the above context, not even why would AI have created a masterpiece but how?
Where is the AI art museum where people come together to decide if something is a masterpiece whilst being blinded to whether it was AI or not?
I mean, I agree there could be an AI masterpiece somewhere right now. I'm just saying Scott's post is not evidence for that in any way (it's evidence that people have no taste, though, and that they don't know how to spot AI. I got almost 90% correct)
"...people have no taste...I got almost 90% correct" seems a slightly wild thing to say, especially given the context? Lol.
Re the survey, I think you misread my comment perhaps?
What do you mean? I'm not following. I got 90% correct and I didn't like very much those AI made. They were mostly obvious too me through taste rather than pattern-matching
I remember complaining to an artist about the "artwork" that I once encountered on the floor of a museum. It was just a piece of cable. That was all. "How can that possibly art?" I said.
The artist drily countered, "But apparently, it's the only piece you're still talking about."
That shut me up.
I think it's the fact that the AI artist has no ego, no stake in the game, you can't hurt it's feelings. it'll just keep saying "sorry sorry..." It has no balls. Moreover. Art is never called a masterpiece until after the artist dies. AI can't die
AI is not the artist though
Sure it is at least partially
I am unmoved by modern art - meh! I find the people who like modern art totally annoying!
Me too lmao
I kindly disagree. WhatWouldJesusDraw.com pumps out masterpieces on a regular basis.