AI in Photography: where do you draw the line?

[evacguy]

What about in-camera processing? Is that AI? Every time your camera spits out an image in jpeg or heif, it was processed by an electronic device making decisions.

I once had a guy who taught photography tell me that if you aren't shooting in M, you're in Auto. P, S, A, choose one. He set that forth as someone should be ashamed of their self for using auto anything. This guy was still stuck on an old second generation DSLR (talk about a luddite) and he failed to recognize that if you're using an in-camera meter, processing, exposure comp, or almost any of the others features, you're in some degree of auto regardless of what the mode dial says. Truthfully, I don't know if he failed to understand or purposely chose not to.

However, none of this nor the vast majority of this discussion is AI. They are programmed functions. They do not learn nor improve. They are subject to the user's input and settings. I take exception to (almost) none of it.

True AI is a different story. Don't take my word for it, look up Stephen Hawking's opinion.

Tim, I'm no expert on AI, so my understanding may be wrong, but as far as I understand AI and machine learning are similar but different. Machine learning is a subset of AI. An AI system mimics how a human makes decisions, it is a set of alorithms that have been trained through machine learning to undertake a particular task. For example the facial recognition system in the latest generation of cameras are AI systems that has been trained through machine learning in the Sony labs. They used thousands and millions of images to train the AI system to recognise human faces, bird faces, dog faces, etc and distinguish between them. When the lab based AI system achieves a high degree of accuracy, they graduate from the learning process and make their way into the AI system in your camera. The facial recognition AI system in your camera is not learning, it has learnt to identify faces of particular species to a given level of accuracy. When the next SONY firmware update comes along, they may have improved the accuracy of the faciial recognistion AI or expanded the AI system to also recognise trains and planes.

 

[Brownie]

Tim, I'm no expert on AI, so my understanding may be wrong, but as far as I understand AI and machine learning are similar but different. Machine learning is a subset of AI. An AI system mimics how a human makes decisions, it is a set of alorithms that have been trained through machine learning to undertake a particular task. For example the facial recognition system in the latest generation of cameras are AI systems that has been trained through machine learning in the Sony labs. They used thousands and millions of images to train the AI system to recognise human faces, bird faces, dog faces, etc and distinguish between them. When the lab based AI system achieves a high degree of accuracy, they graduate from the learning process and make their way into the AI system in your camera. The facial recognition AI system in your camera is not learning, it has learnt to identify faces of particular species to a given level of accuracy. When the next SONY firmware update comes along, they may have improved the accuracy of the faciial recognistion AI or expanded the AI system to also recognise trains and planes.

No doubt that AI is involved in the process, but once in the camera or processing software they don't continue to learn. The generally accepted definition of AI is that it continues to learn. The term has been coopted by marketing departments to make it sound like more than it is to entice you to part with your hard-earned money. I refuse to perpetuate the lie.

 

[Thad E Ginathom]

I was a lousy history student. At best, I enjoyed the stories and then forgot them; at worst, I daydreamed about something else.

But a braincell thinks that it recalls something about Luddites. They weren't quaint old folk drooling over their hand tools (and going home to their vinyl and tapes ) but an early incidence of resistance to "modernisation" because it threatened their jobs.

Which might, or might not, be relevant here. Anyone feel threatened in any way by AI in photography?

 

[Brownie]

Which might, or might not, be relevant here. Anyone feel threatened in any way by AI in photography?

Personally, no. I feel like what I enjoy most about photography is fairly well protected from directly being affected by AI. Of course, I'm sure someone will come up with a way to do it, show me, and screw up my whole world.

For the photography industry as a whole, for the world, with the implications of all the ways it can be misused in ways have that nothing to do with photography, and that haven't even yet been imagined, I feel very threatened.

 

[AlphaWorld]

I was a lousy history student. At best, I enjoyed the stories and then forgot them; at worst, I daydreamed about something else.

But a braincell thinks that it recalls something about Luddites. They weren't quaint old folk drooling over their hand tools (and going home to their vinyl and tapes ) but an early incidence of resistance to "modernisation" because it threatened their jobs.

Which might, or might not, be relevant here. Anyone feel threatened in any way by AI in photography?

Ned Ludd's Luddites were more like a trade union resisting changing that de-skilled their jobs - they were trained skilled workers who felt threatened by mills using machinery so they could employ less-skilled workers.

In photography that's less about AI and more about production line Santa photos - put the box in the viewfinder over the kid's face, push button...

Arguably, AI based tools threaten retouchers more than photographers, although a lot of us are both.

 

[evacguy]

No doubt that AI is involved in the process, but once in the camera or processing software they don't continue to learn. The generally accepted definition of AI is that it continues to learn. The term has been coopted by marketing departments to make it sound like more than it is to entice you to part with your hard-earned money. I refuse to perpetuate the lie.

Tim, I don't think your definition is strictly correct as a formal definition of AI, at least not one that I am familar with. As I suggested earlier, machine learning is a sub-set of AI which means machine learning is definitely a form of AI, but not all AI needs to actively involve machine learning (a Robin is a bird, but not all birds are Robins). For example, rule engines and expert systems are all part of AI, but don't involve active machine learning in their application. Machine learning could have been involved in the training of the AI, but the AI product does not need to actively continue to learn and adapt to be classed as AI. These are not concepts cooked up by marketing departments but by scientists and engineers. However, as you say, marketing departments may take these concepts and bend them to whatever purpose they consider will make them money, but that does not mean that AI MUST involve adaptive learning. So your camera facial recognition system is a form of AI, even though it is not adapting and learning. The Photoshop AI based content replacement algorithm probably continues to learn with use (e.g., users selecting the option that best suits their needs trains the system leading it to make better judgements), but both the facial recognition in camera and photoshop content replacement are examples of different types of AI. At the end of the day, it is just a defintion and one form of AI may be more threatening or concerning to some people than others.

 

[Brownie]

Tim, I don't think your definition is strictly correct as a formal definition of AI, at least not one that I am familar with. As I suggested earlier, machine learning is a sub-set of AI which means machine learning is definitely a form of AI, but not all AI needs to actively involve machine learning (a Robin is a bird, but not all birds are Robins). For example, rule engines and expert systems are all part of AI, but don't involve active machine learning in their application. Machine learning could have been involved in the training of the AI, but the AI product does not need to actively continue to learn and adapt to be classed as AI. These are not concepts cooked up by marketing departments but by scientists and engineers. However, as you say, marketing departments may take these concepts and bend them to whatever purpose they consider will make them money, but that does not mean that AI MUST involve adaptive learning. So your camera facial recognition system is a form of AI, even though it is not adapting and learning. The Photoshop AI based content replacement algorithm probably continues to learn with use (e.g., users selecting the option that best suits their needs trains the system leading it to make better judgements), but both the facial recognition in camera and photoshop content replacement are examples of different types of AI. At the end of the day, it is just a defintion and one form of AI may be more threatening or concerning to some people than others.

You're missing the point, stop thinking like a scientist for a minute. It's not what is 'strictly correct', it's the way AI has been described on the news and in documentaries, dumbed down for absorption by the masses. It's what the general public has come to expect. Manufacturer's can now exploit the term and use it however they like with a simple spin.

 

[iamdlewis]

David, your opinon as to what constitutes acceptable editing in this photograph is fair enough, it is your photograph after all, but I think we will all draw the line between what is acceptable photographic editing and what is creating digital art, at different points. To me, all the proposed edits are acceptable in creating an interpretative photographic landscape, they are not creating what I and others would call 'digital art'. If the sky was changed to show a stary night sky with the Milky Way, that would be digital art, not photography. Alternatively, if your aim in taking the photograph was to create a factual photographic record of this place at this time, then none of the edits would be acceptable.

So if I were to combine a photo of the night sky over the city and a photo of the city on the same evening that would be digital art but removing permanent structures is "interpretative photographic landscape" not art? To me saying that something is "interpretative" means that you are modifying to an ideal or imagined setting regardless of the reality of the situation which would make it art.

 

[evacguy]

So if I were to combine a photo of the night sky over the city and a photo of the city on the same evening that would be digital art but removing permanent structures is "interpretative photographic landscape" not art? To me saying that something is "interpretative" means that you are modifying to an ideal or imagined setting regardless of the reality of the situation which would make it art.

If you combined a photo of the night sky over a city, with a photo of the same city on the same night, I think I'd call that acceptable editing. If I combined a night sky from one location with a completely different location taken at a different time, I'd call that digital art.