‘There is no such thing as a real picture’: Samsung defends AI photo editing on Galaxy S24
‘There is no such thing as a real picture’: Samsung defends AI photo editing on Galaxy S24
Yeah, this is a great example of a true statement that just serves to muddy the water of the actual argument.
A better way to think about it is: an AI-dependent photo is less representative of whatever is in the photo versus a regular photo.
It's not even a true statement. "A real picture of a pipe" has never once in history been interpreted as "my golly - there's an actual goddamn pipe trapped inside this piece of paper". We know it's a freaking representation.
The "real" part refers to how it's a product of mechanically capturing the light that was reflected off an actual pipe at some moment in time. You could have a real picture with adjusted colours, at which point it's real but manipulated. Of course with digital photography it's more complicated as the camera will try to figure out what the colours should be, but it doesn't mean the notion of a real picture is suddenly ready for the scrapyard. Monet's painting is still a painting.
Everyone knows exactly what you mean when you say a real picture. Imposing a 3D model over the moon to make it more detailed, for example, constitutes "not a real picture". Pretending this is some impossible philosophical dilemma is just a corporate exercise in doublespeak.
To play devil's advocate, even traditional photography involves a lot of subjective/artistic decisions before you get a photo. The type of film used can massively affect the image reproduced, and then once the photos are being developed, there's a load of choices made there which determine what the photo looks like.
There's obviously a line where a photo definitely becomes "edited", but people often believe that an objective photo is something that exists, and I don't think that's ever been the case.
An AI edited photo might not necessarily be less representative of whatever is in the photo. Imagine an image taken in a very dark room, then an AI enhancement makes it look like the lights are on. You can actually get a much better idea of what's in the room, but a less good idea of what the lighting was like. So it comes down to opinion, which one is more representative of reality? Because no photo since the beginning of time has been completely representative of what humans actually see with their eyes. It's always been a trade-off of: what do we change to give humans the image they want with the technology we have.
We are truly in a post-truth era
Lie on your resume.
Photo manipulation has been a thing since photos have been a thing
That's true, but they didn't used to sell you a camera claiming it would take a picture when in fact it just invented a picture it thought looked similar to the picture you were trying to take.
While that is true, it has gotten incredibly easy to alter and spread such photos. We all love interesting ways to take photos with optical illusions and practical effects.
Just like using a gun. It has gotten very easy to inflict disproportional harm compared to just 50 or 100 years ago.
Lie on your resume.
Wait... we weren't supposed to be doing this for decades already? No one told me.
How can the picture be real if your eyes aren't real?
The universe is a hologram projected from 19 dimensional space to look like it exists in 4 dimensional spacetime.
Settle down, Jayden
Hands Monopoly money to the clerk at a Samsung store
"I'll take the S24, money is a made up concept anyway"
Ha ha, very clever... money is just made up. But wait, so are borders, sovereignty, language, art, moral and commercial value, the law, logic, authority, human rights, culture, and government.
According to Jacques Lacan, experience itself is a fabrication; the worst thing that can happen to a person is to come into contact with the real.
The statement that "There is no such thing as a real picture" isn't wrong. It kind of missed the point though. It's true that, even when a photo attempts to make the most faithful representation possible, it can only approximate what it sees. The sensors used all have flaws and idiosyncracies and software that processes the images makes different decisions in different situations to give a good image. Trying to draw a line between a "real" picture and a "fake" picture is like trying to define where the beach ends and where the ocean begins. The line can be drawn in many places for many reasons.
That said, the editing that the S24 is going to allow may be going a little far in the direction of "fake" from the sounds of things. I'm not sure if that is good or bad but it does scare me that photos can't really be relied upon to give an accurate representation of a scene anymore. Everyone having access to ti's kind of AI is going to make it tremendously difficult to distinguish between realistic and misleading images.
At least sensors will be relatively consistently flawed, while AI can just completely make details up.
It's not just the sensors though. The software used to convert what the sensors saw into an image makes decisions. Those decisions are sometimes simple and sometimes complex. Sometimes they are the result of machine learning and might already be considered to be AI. This is just another step in the direction of less faithfulness in photos.
Capturing any data or making any measurement is an approximation, because every type of sensor has a limited degree of accuracy - with some more sensitive than others.
I think there is a clear enough line however between making an approximated record of a value, and making a guess at a value, the latter being essentially how these “AI” camera systems work.
I disagree. It's not that easy to draw a line.
First, current cameras that we consider to not use AI still manipulate images beyond just attempting to approximate the scene. They may not allow easy face swapping but they still don't faithfully represent the scene much of the time.
Also, I don't even think it is clear where we can draw a line between "normal" algorithms and "AI" algorithms. What level of machine learning is required before we consider an alrogitm to be AI?
Simple non-AI algorithms and generative AI are on a spectrum of comlexity. They aren't discrete from one another such that they can be easily categorized.
Whether or not this feature is on Samsung phones it will still be accessible. It already is really. You can't hold back the tide.
Yeah, you're right. It still scares me somewhat, though. What happens when courts fall behind and continue to rely on photo evidence after photo evidence becomes easy for anyone to fake. What happens when the courts finally do realize that photos are unreliable?
I don't think this change can or should be stopped. It is just worrisome and thought should be put into how to mitigate the problems it will inevitably cause
Do you think that's air you're breathing?
“There was a very nice video by Marques Brownlee last year on the moon picture,” Chomet told us. “Everyone was like, ‘Is it fake? Is it not fake?’ There was a debate around what constitutes a real picture. And actually, there is no such thing as a real picture. As soon as you have sensors to capture something, you reproduce [what you’re seeing], and it doesn’t mean anything. There is no real picture. You can try to define a real picture by saying, ‘I took that picture’, but if you used AI to optimize the zoom, the autofocus, the scene – is it real? Or is it all filters? There is no real picture, full stop.”
If your epistemological resolution for determining the fakeness of the moon landing photos is to just assert that all photos are in a sense fake so case closed, then I feel like you aren't even wrong about the right thing.
The moon photos they’re talking about are specifically the AI enhanced zoom moon photos of previous Samsung models that caught controversy because taking a picture of a marginally round object against a black background with their zoom enhancement would produce a moon photo even if it was in someone’s dark basement and the object was a dimly lit bottle cap.
if you used AI to optimize the zoom, the autofocus, the scene – is it real?
To me, using AI to optimize zoom, focus, aperture (or fake aperture effects), framing etc. That's composition. The picture isn't fake, but software helped compose the real image in a better way.
When you change the image (remove objects, distort parts of the image not the whole, airbrush etc) then the image isn't based on reality any more.
That's where I see the line drawn, at least. Yes, drawing a line also makes the image not real any more.
Beyond this, we get to philosophy. In which case, I'll refer to my other comment on another post about this story. Our brain transforms the image our eyes receive (presumably to be able to relay it around the brain efficiently, who knows?). So we can take it to Matrix philosophy. When we don't know if what we're seeing is real, what is real?
I think the reality is that there is no reality, there is only perception. Composition does add to remove things from the photo. Light, both the amount and its wavelength, is a thing. Whether the lens picks up the pores on a person's face is a thing. Whether The background seems close or far as a thing. But I agree that camera makers would tow any philosophical line to help them drive profits.
There are certainly purposes for which one wants as much of the raw sensor readings as possible. Other than science, evidence for legal proceedings is the only thing that comes to mind, though.
I'm more disturbed by the naive views so many people have of photographic evidence. Can you think of any historical photograph that proves anything?
Really famous in the US: The marines raising the flag over Iwo Jima. It was staged for the cameras, of course. What does it prove?
A more momentous occasion is illustrated by a photograph of Red Army soldiers raising the soviet flag over the Reichstag. The rubble of Berlin in the background gives it more evidentiary value, but it is manipulated. It was not only staged but actually doctored. Smoke was added in the background and an extra watch on a soldier's arm (evidence of robbery) removed.
Closer to now: As you are aware, anti-American operatives are trying to destroy the constitutional order of the republic. After the last election, they claimed to have video evidence of fraud during ballot counting. On one short snippet of video, one sees a woman talking to some people and then, after they leave, pull a box out from under a table. It's quite inconspicuous, but these bad actors invented a story around this video snippet, in which a "suitcase" full of fraudulent ballots is taken out of hiding after observers leave.
As psychologists know, people do not think in strictly rational terms. We do not take in facts and draw logical conclusion. Professional manipulators, such as advertisers, know that we tend to think in "narratives". If a story is compelling, we like to twist neutral snippets of fact into evidence. We see what we believe.
The situations that drive me nuts are the conspiracy idiots who zoom in super hard on some heavily compressed image they pulled off of the web. They then proceed to claim that compression artifacts, optical flares, noise, etc are evidence of whatever crap they are pushing.
Taking things out of context is another issue. It has become painfully common online. I would see it all the time when pushing the "all police are bad!" narrative. They will deliberately edit out the violence that triggered the arrest then make it look like the arrest was unwarranted and overly physical. People will do this with dashcam videos and show road rage but edit out the part where they triggered it with their own aggression.
Ok the one hand, yeah. Actions have consequences. On the other hand, no amount of aggressive driving "deserves" to be responded to in the way some do, and no amount of someone doing something dangerous or illegal justifies police using unnecessary force (or else we wouldn't call it that). Once they've been subdued, it should be done.
The Moon landings? Hello?
Seriously? At least clarify that you mean film and not photographs. The effects of lower gravity and not atmosphere take at least some effort to get right. Photos can be staged anywhere.
Have you ever looked at the arguments of moon hoaxers? A lot of them would be good questions, if they were questions. Why can't you see the stars? Why are there multiple shadows if the sun is supposed to be the only light source? The boot prints are so perfect, as prints only are in wet sand. How can the flag wave without an atmosphere?
You can easily find answers (and more questions). How do you "prove" these answers? How do you go from that to actually turning photos, and even film, into proof positive of the moon landing?
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
I support automatically face swapping everyone's faces in photos with the gingerbread man from Shrek.
After all, there's no such thing as a real photo anyway, so Samsung editing them in unusual ways is completely reasonable.
Reminds me of jREG's hyper-self rant
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
At this rate with this mentality, we are going to have photo "evidence" of bigfoot and UFO's before 2030.
Early XXI century seems to model late XIX century quite well, if you think about it. A tremendous leap followed by gimping and all kinds of such stuff.
We've had photo (and film!) "evidence" of both for many decades. "It has to be real, because they did not have CGI back then" is something that people actually argue.
Here's a famous Bigfoot film.
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
Sorry, but I don't want to see fake "real" images just because people think it looks good.
edits made using this generative AI tech will result in a watermark and metadata changes
The metadata is easy to erase. It’s only a matter of time until we start seeing some open source projects come out that can remove the watermarking the AI players are starting to try.
FWIW, The Samsung Boss said:
As soon as you have sensors to capture something, you reproduce [what you’re seeing], and it doesn’t mean anything. There is no real picture.
I understand this as talking about a definitive original, as you get with trad analog photography. With a photographic film, you have a thin coat (a film) of a light sensitive substance on top of a strip of plastic. Taking an analog picture means exposing this substance to light. The film is developed, meaning that the substance is chemically altered to no longer be light sensitive. This gives you a physical object that is, by definition, the original. It is the real picture. Anything that happens afterward is manipulation.
An electronic sensor gives you numbers; 1s and 0s that can be copied at will. These numbers are used to control little lights in a display.
As far as I understand him, he is not being philosophical but literal. There is no (physically) real picture, just data.
I feel most of this is a slippery slope / negative sum spiral.
See e.g. Liv Boeree's video on beauty filters.
Here is an alternative Piped link(s):
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source; check me out at GitHub.
Altering the photo even further only makes it worse though ...
I'm altering the photo, pray I don't alter it any further.
This photo is getting worse all the time.
Enshitification of the photo
They do have a point when they say AI is here to stay, and what they propose (A 'watermark' in the metadata for AI edited content) is at least a way forward. There should be also some electronic seal/signature for this to be effective though (md5?) , metadata so far is easy to tweak
What absolute hogwash.
So, I'm just going to not buy their garbage.
What's a Polaroid then?
A Polaroid is the best representation that can be made of a scene on Polaroid photo film. The lens, the paper, and other factors will always make the representation, to a degree, not real. That was the Samsung exec's point. It's a little disingenuous, though. The discussion shouldn't be about "real" vs "fake" it should be about "faithful" vs "misleading".
The Samsung Boss said:
As soon as you have sensors to capture something, you reproduce [what you’re seeing], and it doesn’t mean anything. There is no real picture.
A Polaroid photograph is a real picture, in the sense that it exists as a single, definitive, physical thing. Whether what it shows is real is a different question, though.
I don't really understand what is unethical about AI photo edit to begin with? Like photo editing before AI existed and you could make anything you want with a ps.
Because it's not automatic and is outside of the user's control, to start.
It's not?
Besides potential copyright infringement, there isn't really any issue with it I'd say. This is still an incredibly stupid take from Samsung however.
Im 14 and this is deep
When corporations dabble in philosophy, you know they're trying to muddy the waters and skirt an ethical issue. It's not a genuine inquiry going on here; it's a "whatever argument serves the bottom line" situation.
I guess there's no such thing as intellectual property either, when you really think about it. Hence nothing wrong with me making and selling pirated samsung phones.
Can i copyright my face and get an ai to trawl the internet for any pictures of me and demand people take them down or pay me?
If you're prominent enough and can afford to pay for the lawyers, you can assert ownership of your own likeness, yes.
That either is necessary is a travesty, though.