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.
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.
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.
There’s a clear difference between a processing mistake and an intentional addition. That’s a fairly clear line.
Grain on a photo is not the same as making you look like a human head on a shark’s body.
Yes, no photo is 100% accurate, just as no statement will ever capture an incident perfectly. That doesn’t mean there’s no such thing as lying.
There is definitely a line.
Is the tech trying to make the photo more accurate, or change the narrative?
Sure, there’s some tech that’s awkwardly in the middle, like skin smoothing, but we don’t NEED that, and it’s not directly changing the narrative of the story, unless you’re selling acne medication.
I still disagree that there is a clear line. Yes, it is obvious that photo grain is different from making you look like a human head on a shark’s body. The problem is somewhere in the middle. Determining where that line is drawn is going to be difficult and is only going to become more difficult as this technology advances.
I think the line (while the details may be certainly difficult) is along “are you making the existing image/story clearer or are you changing the narrative of the media?
When the story you get from the image changes, then you’ve crossed the line.
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
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.
I think that’s disingenuous…
There’s a clear difference between a processing mistake and an intentional addition. That’s a fairly clear line. Grain on a photo is not the same as making you look like a human head on a shark’s body.
Yes, no photo is 100% accurate, just as no statement will ever capture an incident perfectly. That doesn’t mean there’s no such thing as lying.
There is definitely a line.
Is the tech trying to make the photo more accurate, or change the narrative?
Sure, there’s some tech that’s awkwardly in the middle, like skin smoothing, but we don’t NEED that, and it’s not directly changing the narrative of the story, unless you’re selling acne medication.
I still disagree that there is a clear line. Yes, it is obvious that photo grain is different from making you look like a human head on a shark’s body. The problem is somewhere in the middle. Determining where that line is drawn is going to be difficult and is only going to become more difficult as this technology advances.
I think the line (while the details may be certainly difficult) is along “are you making the existing image/story clearer or are you changing the narrative of the media?
When the story you get from the image changes, then you’ve crossed the line.
I generally agree with you but that is still a fuzzy line to draw that is likely very dependent on circumstances. The devil is in the details.
I can concede to that… there will be some grey area, but the idea that “there is no true photo” or “there is no truth” feels wrong.
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
That’s not how courts work. It’s not like there’s a list of acceptable evidence that gets updated once in a while.
Prosecutors and defenders will present evidence to jurors in their contemporary context.
Basically, we all need to acknowledge that images do not convey “truth”, and really they never did.