An interesting read Andy & Alys. Thank you. It hadn't occurred to me in great detail, until this essay, how different AI generated imagery and the possibilities of it may be if regularly in the hands of photographers and those working in photography, rather than those who are just 'playing around'.
I am also left to wonder about the language around it all. I feel there's a distinction between the(a) 'photograph' and 'imagery' and perhaps is a distinction that needs to be made more often?
I disagree slightly with the idea that a photograph equals a reflection of reality. I often use extensive post-processing, which transforms the original photo into something that’s very far from reality. Yes, it was a record of reality at the moment the shutter was released, but the final result can be completely detached from reality.
Wonderful --and at most importance-- food for thought Alys and Andy. Especially the sentence: "One might say that where photography testifies to what has existed, generated images explore what could exist." I agree with some other readers that AI imagery should be marked as such. Though, should all the photoshopped photographs be marked as such too? ;-)
Enjoyed your piece Alys. The notion that photography's distinguishing feature is its link to the real physical world (the photon that bounces off the subject and into the camera, etc.) has always been a red herring. Even old-school photography (film darkrooms, chemicals) was less tightly bound to reality that some would care to admit. The great photographers of the pre-digital age were master manipulators of reality in the ways they controlled light at the moment of exposure, altered appearances through approaches to developing exposed film, and perhaps most notably, how they altered tonality, light and shadow and contrast through choices made into the printing process. It seems to me we only started expressing discomfort about the manipulation of reality inherent in photography when digital editing tools become commonplace, but of course the manipulative element had been there all along -- indeed Photoshop mimics a lot of analog darkroom techniques. When I photograph something and then edit it digitally (a process which may include removing items that appear inconveniently in the captured frame) there's very little that distinguishes this, conceptually, from an AI generated image that "looks like" a photograph. The difference is the starting point -- in my case, a real physical scene and in AI's case a storehouse of billions of tokens. If someone can produce arresting images with AI then more power to them! The hand-wringing over AI generated image-making (which I do not practice by the way; I prefer photography) sounds a lot like the empty debate about analog vs. digital photography which hopefully has been laid to rest.
Imagine a guy painting a still life. One artist paints what’s right in front of him, while the other just dreams it up. Should they have to slap a label on their work saying it’s not a carbon copy of reality?
If the difference between a photo and an AI-generated image is noticeable, labeling is unnecessary. And if it isn't noticeable, then it's actually unnecessary as well.
It is time to get real specific. See tree trunk: Left branch, manipulated digital photographs which started with an actual camera. Next branch analogue or straight photography. Somewhere else on the tree, next to NFTs, animation and other branches, you can have generative AI. They are not the same, and there can be no crossover or cross pollination unless clearly stated. And then there is the matter of ownership. My patent attorney says that the USSC has ruled that you cannot own something you did not make yourself, which would suggest a very limited market. Like Banksy, if the root is not visible (i.e. he remains anonymous) there can be no ownership. I have yet to hear how that is dealt with? After all things which cannot be monetised/protected have a way of dying on the vine…. or should I say branch.
Intellectual analysis of the image making spectrum is seductive. Boundaries are needed and labels. If an image is AI was its generation prompted by photos? Where did the photos come from? Generative AI is often creative and clever - then quite stupid. AI’s tendency is to never admit defeat. Filling in the blanks with garbage seems to be better than admitting a lack of knowledge. The photographers who pass off AI generations as photos are an embarrassment. Getting really good photos is difficult. I hope photographers aren’t giving up.
Hi Alys. I was expecting a lot of hate in the comments but they're surprisingly constructive. And that's good because this debate is important, whichever side you are.
I thought the punctum to index pivot was a very good move. You're right that Hoffman's argument is better read through Peirce than Barthes, and I like that you called the indexicality critique "the strongest" instead of strawmanning it.
But I finished your post feeling that the indexicality concession was doing a lot more than you acknowledged. You grant photography is an index, AI isn't, and then move to "two complementary regimes." But what about computational photography, which already blurs the line between recording and construction? Your phone doesn't show your a real picture, it's already a composite and has already been processed away from pure indexicality. And your own hybrid practice feeding photographs into AI models lives right on that boundary. I think the regimes framework would benefit from testing itself against the hard cases: an AI image trained on someone's likeness, or one passed off as documentary unlabelled. The framework doesn't tell you where those stand.
I'd love to read a follow-up that addresses this conversation instead of resolving it by categorisation, especially around the training-data question you highlighted in the opening and never returned to. That felt like an idea that needed more depth.
The labelling section is also very sensible. Concrete, actionable, and a real contribution to the norms conversation.
I am leery of the Ai trickery in creating a photographic image when it comes to documenting real events. I do not mind it when it is used to create surreal images or humorous ones that open up a fantastical image to enjoy.
What seems to be overlooked in evaluating the difference between photography and AI imagery is the existential aspect, the very foundation of what it is. I'm personally not into photorealistic paintings for the same reason I'm not into AI imagery. If the images in this post were photographs I might be interested in them in that they were a depiction of reality observed by someone (To be honest, this is also why I'm not interested in the staged photography of artists like Wall or Crewdson, but that's at least still a kind of photography).
So why the conflation? You could start a companion blog about your interest in photorealistic paintings, or papier-mâché, or mid-engined cars of the 1960s. I'd wager most of your readers are here to read about photography instead of those other things, so I always find these posts about something that is in no way photography a little incongruous.
An interesting read Andy & Alys. Thank you. It hadn't occurred to me in great detail, until this essay, how different AI generated imagery and the possibilities of it may be if regularly in the hands of photographers and those working in photography, rather than those who are just 'playing around'.
I am also left to wonder about the language around it all. I feel there's a distinction between the(a) 'photograph' and 'imagery' and perhaps is a distinction that needs to be made more often?
I disagree slightly with the idea that a photograph equals a reflection of reality. I often use extensive post-processing, which transforms the original photo into something that’s very far from reality. Yes, it was a record of reality at the moment the shutter was released, but the final result can be completely detached from reality.
Thanks for the note, Pavel. You might appreciate this post: https://www.flakphoto.news/p/does-the-camera-lie
Wonderful --and at most importance-- food for thought Alys and Andy. Especially the sentence: "One might say that where photography testifies to what has existed, generated images explore what could exist." I agree with some other readers that AI imagery should be marked as such. Though, should all the photoshopped photographs be marked as such too? ;-)
An excellent question, Kinga!
Enjoyed your piece Alys. The notion that photography's distinguishing feature is its link to the real physical world (the photon that bounces off the subject and into the camera, etc.) has always been a red herring. Even old-school photography (film darkrooms, chemicals) was less tightly bound to reality that some would care to admit. The great photographers of the pre-digital age were master manipulators of reality in the ways they controlled light at the moment of exposure, altered appearances through approaches to developing exposed film, and perhaps most notably, how they altered tonality, light and shadow and contrast through choices made into the printing process. It seems to me we only started expressing discomfort about the manipulation of reality inherent in photography when digital editing tools become commonplace, but of course the manipulative element had been there all along -- indeed Photoshop mimics a lot of analog darkroom techniques. When I photograph something and then edit it digitally (a process which may include removing items that appear inconveniently in the captured frame) there's very little that distinguishes this, conceptually, from an AI generated image that "looks like" a photograph. The difference is the starting point -- in my case, a real physical scene and in AI's case a storehouse of billions of tokens. If someone can produce arresting images with AI then more power to them! The hand-wringing over AI generated image-making (which I do not practice by the way; I prefer photography) sounds a lot like the empty debate about analog vs. digital photography which hopefully has been laid to rest.
AI imagery needs to be marked as AI. Each image needs a watermark designated it as such. People need to know what they’re looking at.
Imagine a guy painting a still life. One artist paints what’s right in front of him, while the other just dreams it up. Should they have to slap a label on their work saying it’s not a carbon copy of reality?
If the difference between a photo and an AI-generated image is noticeable, labeling is unnecessary. And if it isn't noticeable, then it's actually unnecessary as well.
A remix of humanity's collective imagery running through a set of machines rolling the dice. That's my take on it, anyway… though never say never…
It is time to get real specific. See tree trunk: Left branch, manipulated digital photographs which started with an actual camera. Next branch analogue or straight photography. Somewhere else on the tree, next to NFTs, animation and other branches, you can have generative AI. They are not the same, and there can be no crossover or cross pollination unless clearly stated. And then there is the matter of ownership. My patent attorney says that the USSC has ruled that you cannot own something you did not make yourself, which would suggest a very limited market. Like Banksy, if the root is not visible (i.e. he remains anonymous) there can be no ownership. I have yet to hear how that is dealt with? After all things which cannot be monetised/protected have a way of dying on the vine…. or should I say branch.
Intellectual analysis of the image making spectrum is seductive. Boundaries are needed and labels. If an image is AI was its generation prompted by photos? Where did the photos come from? Generative AI is often creative and clever - then quite stupid. AI’s tendency is to never admit defeat. Filling in the blanks with garbage seems to be better than admitting a lack of knowledge. The photographers who pass off AI generations as photos are an embarrassment. Getting really good photos is difficult. I hope photographers aren’t giving up.
Hi Alys. I was expecting a lot of hate in the comments but they're surprisingly constructive. And that's good because this debate is important, whichever side you are.
I thought the punctum to index pivot was a very good move. You're right that Hoffman's argument is better read through Peirce than Barthes, and I like that you called the indexicality critique "the strongest" instead of strawmanning it.
But I finished your post feeling that the indexicality concession was doing a lot more than you acknowledged. You grant photography is an index, AI isn't, and then move to "two complementary regimes." But what about computational photography, which already blurs the line between recording and construction? Your phone doesn't show your a real picture, it's already a composite and has already been processed away from pure indexicality. And your own hybrid practice feeding photographs into AI models lives right on that boundary. I think the regimes framework would benefit from testing itself against the hard cases: an AI image trained on someone's likeness, or one passed off as documentary unlabelled. The framework doesn't tell you where those stand.
I'd love to read a follow-up that addresses this conversation instead of resolving it by categorisation, especially around the training-data question you highlighted in the opening and never returned to. That felt like an idea that needed more depth.
The labelling section is also very sensible. Concrete, actionable, and a real contribution to the norms conversation.
I am leery of the Ai trickery in creating a photographic image when it comes to documenting real events. I do not mind it when it is used to create surreal images or humorous ones that open up a fantastical image to enjoy.
What seems to be overlooked in evaluating the difference between photography and AI imagery is the existential aspect, the very foundation of what it is. I'm personally not into photorealistic paintings for the same reason I'm not into AI imagery. If the images in this post were photographs I might be interested in them in that they were a depiction of reality observed by someone (To be honest, this is also why I'm not interested in the staged photography of artists like Wall or Crewdson, but that's at least still a kind of photography).
So why the conflation? You could start a companion blog about your interest in photorealistic paintings, or papier-mâché, or mid-engined cars of the 1960s. I'd wager most of your readers are here to read about photography instead of those other things, so I always find these posts about something that is in no way photography a little incongruous.