Take a look at this photograph that won a prize at the prestigious Sony world photography awards last week.
After being awarded the prize, the winner Boris Eldagsen declined it saying that the photo was AI generated and that he had submitted it to start a conversation about how to deal with AI in the art world.
“We, the photo world, need an open discussion,” said Eldagsen. “A discussion about what we want to consider photography and what not. Is the umbrella of photography large enough to invite AI images to enter – or would this be a mistake?
“With my refusal of the award I hope to speed up this debate.”
He said this was a “historic moment” as it was the first time an AI image had won a prestigious international photography competition, adding: “How many of you knew or suspected that it was AI generated? Something about this doesn’t feel right, does it?
Photographs, especially artistic ones, are not merely representations of the external world. There are always elements imposed by the artist, such as choice of setting, background, lighting, exposure, framing, etc., not to mention all the work that goes on the darkroom to modify it to get just the right effect that the photographer is seeking. The end result is that what finally emerges was never seen in real life.
The organizers of the competition seem to recognize the complexity of these issues, saying in a statement:
“We recognise the importance of this subject and its impact on image-making today. We look forward to further exploring this topic via our various channels and programmes and welcome the conversation around it. While elements of AI practices are relevant in artistic contexts of image-making, the awards always have been and will continue to be a platform for championing the excellence and skill of photographers and artists working in the medium.”
Eldagsen himself feels that AI images should not be allowed in such competitions.
“AI images and photography should not compete with each other in an award like this. They are different entities. AI is not photography,” Eldagsen wrote on his website. “Therefore I will not accept the award.”
“With my refusal of the award, I hope to speed up this debate.”
The idea of using computers, especially AI, to enhance an image may not seem like much of a stretch. It may depend on how much AI was involved in the end result. Did the photographer specify the way that the women should look, their relationship to each other, their expressions, and so on?
Eldagsen explains what went into getting the final result.
His photograph, “The Electrician”, features two women in a grainy sepia tone that gives it a 1940s style. The work was created in collaboration with an AI used by Eldagsen, who refers to himself as a “photomedia artist” on his website.
“The Electrician” is part of a series by Eldagsen called “pseudomnesia,” the Latin term for “fake memory.” The images are “fake memories of a past, that never existed, that no-one photographed,” created by putting them through AI image generators between 20 and 40 times, Eldagsen says on his website.
“I have been photographing since 1989, been a photomedia artist since 2000. After two decades of photography, my artistic focus has shifted to exploring the creative possibilities of AI generators,” Eldagsen wrote on his website. “The work SWPA has chosen is the result of a complex interplay of prompt engineering, inpainting and outpainting that draws on my wealth of photographic knowledge. For me, working with AI image generators is a co-creation, in which I am the director. It is not about pressing a button—and done it is.”
It seems to me that a lot of creative work went into getting this image. If they had been done in the darkroom or computer starting from a real photograph, it would not be controversial. But because it was all done on a computer, should that disqualify it as art?
No, it just disqualifies it as a photograph.
What is it about A.I. and hands. The person in back has two right hands.
😉
Right, not merely representations. But they are representations. For example, the pictures we see of galaxies, nebulae, etc are mappings of the frequencies of light received to frequencies we can see as colours. But one assumes that no galaxies, etc have been manufactured, added to or removed from the final product.
I was expecting a post about some ethical dilemma.
What this supposed ethical dilemma may be, I know not.
Which one is the electrician?
I don’t know who judges these and what they are looking for (or even looking at) but it took me a few seconds to know this was AI generated. Just looking at the hands was enough especially the one on the left as well as the wrist. Also, the woman in back looks cartoonish in a way that is common in those types of AI images.
Not impressed.
antaresrichard@#2:
What is it about A.I. and hands. The person in back has two right hands.
If you think about it, its a problem in detecting recursion. The lower parts of fingers are similar to the upper parts, and the nails are similar to the middle parts, and the arms are similar to other dangly things -- the way the AIs currently work is they subtract from a field of noise everything that does not match a specific prompt. So, the extra fingers appear where an extra finger or twenty would match the prompt acceptably. The current thing is to use negative prompts, like “no extra hands” which cause the AI to change its matching criteria in ways that might result in fewer extra hands.
The AIs do remarkably well given that they have no model (currently) for a body, or number of parts of structures. The training is all brute-force, which as Stalin said, “quantity has a quality all of its own unless you are invading Ukraine” There are already work-arounds in progress, such as inpainting and structural weights (LORA) etc. which repair various mistakes in various ways. For example, LORA permute the noise to make it a bit more hand-like, which greatly increases the chance that the resulting image will be a lot more hand-like. Inpainting allows the user to select a region they don’t like, describe what they want in it, and re-run the AI generation repeatedly to evolve toward the desired goal.
I play with this stuff a lot and I’m finding it is fun and stimulating -- writing good prompts and figuring out the various magical incantations is challenging. But it’s getting better really fast. I mean, really fast. These pecadilloes will be squashed within a year, two at the outside.
PS -- evolution is in action bigtime in AI land. One easy way to get a picture with good hands is to ask the computer to make 20, then you pick the one you like best. Or another 20. Repeat. It’s an incomplete feedback loop. There are projects in research now that are attempting to capture human thumbs up/down as inputs into the feedback loop to adapt the AI’s weightings. That will, eventually, work. The problem with evolution as an algorithm is you have to wade through a lot of failure first. A whole lot. And in the end, you still have occasional failure.
It should not be included in a photography contest, as it is not a photograph. Easy. The tricky thing will be detecting frauds. Easy for the time being, but AI image generation is only going to improve over time, as this story and others shows it can already fool unsuspecting people.
I can only be sure that there are two focus points for the same lens. One on the eyes of each face; nice, believable bokeh blur elsewhere.
I wouldn’t have noticed it at all if I hadn’t been primed to look for AI artefacts.
Whiskey Tango Foxtrot?! That “photo” screams “fake” (along with “creepy and ugly”) from the uttermost depths of the Uncanny Valley! We need to have a conversation about how it got past people who are supposed to be experts judging photos for a prestigious award. Where did Sony find those jokers?
If a CGI-doctored “photo” of, say, Emma Watson’s face on a porn-star’s body doesn’t qualify as a “photograph,” then neither should imitative AI “art.” And yes, what we’re seeing here is not “generative,” it’s imitative. There’s a difference.
Raging Bee,
It’s both.
I found the image to be intriguing but would never have guessed that it was AI-generated.
There is something vaguely unsettling, sinister even, in the expressions of the two women but that is what made the image compelling to look at, as it made me wonder what their backstory might be.
@3:
But then in almost every article about some astrophysical discovery you end up with a big leading picture “artist’s impression of what black hole X would actually look like”. Argh! Just die already! Either show me what it actually looks like if you can (which for most of these things is a complete impossibility) or show nothing. Because most of this shit is just the somnorific attempt to induce the impression of having learnt something and that *really* fucks me off.
@13:
I definitely would have said AI after more than a cursory look. But then that’s now, in an environment full of AI images and all the well known artifacts that entails.
I think the problem for the judges, one that readers here might not have appreciated, is that there are an awful lot of “legitimate” photographic techniques spanning the entire path from the glass to the paper that can produce all manner of weirdnesses into the final image. So while a lay person might be able to tell the difference between “commercial magazine art” and AI generation pretty well, I suspect their (and my own) accuracy might not be quite so good if exposed to the entire breadth of the photographic art scene.
Well, Mano, now you know: THERE IS NO BACK-STORY. Unless, of course, you ask ChatGPT to “generate” one.
All I know is that I have generated hundreds of AI images and I don’t consider myself an AI artist.
I definitely would have said AI after more than a cursory look.
So would I. But another possibility would be a bad artist trying to paint over a photograph; like the cover art for Songs from the Wood by Jethro Tull (1977, so no AI to blame there, except maybe the one that they used to fake the Moon landing and the round Earth /s).
@17:
I have personally produced a lot of omelettes, but don’t consider myself a chicken murderer.
Regarding the introduction of photography:
(https://www.artinsociety.com/pt-1-initial-impacts.html)
I prefer ‘pure photography’, although nothing is ever completely pure. However, by this, I mean I like to use film (not digital) and choose the f stop, ISO, exposure and compose the photo in my mind and then in the camera. I do no further processing, so to a certain extent it is pure. It is what it is and stands or not on its merits.
Digital processing and/or AI photography are not pure photography.b They involve the hand go the human to guide them. The AI machine has to be programmed and sqwtiched on, and digital orcessng in Lightroom or Phtoshop again is manipulation.
That said, I don’t think AI or Digital photography is unethical -- it is just not phtographic art for me.
.
[I did wait]
I noticed the hand (claw) groping whatever might be in the breast area.
Weird.
@Tabby Lavalamp, 17:
I apologise in advance if you didn’t mean what I think you mean by this, which is more or less “AI images aren’t art”.
I’ve plonked the keys on a piano probably thousands of times and I don’t consider myself a pianist. Merely generating an image using AI is as trivially easy as picking out a dozen randomly selected keys on a piano. Art comes in first, learning to manipulate the output meaningfully to achieve a desired effect, and second knowing what to reject and how to improve.
No matter how good AI image generators get*, they will still be the equivalent of a musical instrument. Some people will be able to do the equivalent of chopsticks, and some will be able to do Rachmaninov. And if some people can’t tell the difference, well, some people are philistines, eh?
*… at least until they can literally read my mind directly and transfer the image there to a screen. I’m not convinced this will ever be possible.
There have been many comments about the hand in front. I thought that it belonged to the woman in front and so did not think there was anything odd about it.
The woman in front obviously has no arm to support the hand in front. Which looks quite male, BTW.