Beyond the Frame 43/
Is Artificial Intelligence a dream come true for photographers? How well can you judge whether a picture was AI-generated? What does an AI future look like for creative people?
Artificial Intelligence and Photography
I have a series of recurring dreams. In one, it’s always 1943 and I am evading enemy troops in rural parts of occupied France. I hide in barns and inside damp culverts, listening to the sound of marching jackboots whilst I hold my breath and peer at tiny silk maps.
In another, I’m piloting a hot-air balloon over the Himalayas, using all my expertise to gain enough lift to clear the snowy peaks, rescuing a group of small children and furry animals huddled in the foot of the wicker basket.
In my dreams, as in life, I am fightin’ Nazis and savin’ kittens.
This post is not about those dreams.
This is about another dream in which I become a time-traveller. I return to my Primary school as a guest speaker. Naturally, my eight year-old classmates are impressed that I have travelled fifty years through time, back to 1975. However, being avid readers of sci-fi comics, they accept that time travel will be a normal part of life in the 21st century. They ask, “Where’s your jet-pack?”
I gloss over the disappointing fact that we still don’t have jet-packs in 2025 and, instead, place my iPhone on the teacher’s desk. The children gather round, wide-eyed.
I explain that everybody I know in 2025 carries something similar. I wake the phone, the screen lights up, thirty kids yell “Whoa!”
I explain that it’s possible to speak to any person in the world who has a similar device. Then I make a FaceTime call to a friend in 2025 (it’s a dream, I can do this) and when a face from the future appears on the screen, 30 eight year-old minds explode.
I demonstrate the camera, they pass the phone around and take photos.
“What, no film? No processing?”
“Nope. It’s all digital. I take the photo and can send it anywhere in the world at the push of a button.”
It’s clear that my revelations from the future have amazed my classmates. Hands shoot up. There are many questions. But, first, I have one more magical revelation.
“Watch this! I will speak to my device, describe a scene in words, and it will create a lifelike image, just like a photograph.”
I dictate to my phone:
“A young Asian woman outside the Louvre in Paris”.
”A man in an elaborate headdress celebrating Chinese New Year.”
“A bearded Indian man wearing an orange turban and looking up at the sky.”
I show the AI-generated “photos” and await the applause.
But the room is silent. After several seconds, one hand is raised and a voice pipes up from the back of the classroom.
“Why?”
That’s always the point at which I wake up.
Why, indeed?
And so, on behalf of 30 bewildered schoolchildren from the past, that’s the question I’m trying to answer.
How good is AI?
Before I delve into that question, let’s see how easily you can identify an AI-generated image. My prediction is that 90% or more of readers will be able to accurately select the AI image from these pairs below.
In each pair, one image is AI-generated, one is my own photograph. Click on the buttons to indicate which you believe is an AI image and I’ll reveal the answers in the next newsletter. (Don’t be misled if a tick appears beside your selection, it just indicates that your vote has been recorded.)
If you found that straightforward, congratulations.
I’m not sure that I was even aware of AI two years ago. Now it’s everywhere. Companies boast relentlessly about their AI enhancements. Every imaging app and software seems to have an AI function. AI is in my email, in my calendar, and in my word processor. The speed of AI’s advancement is eye-watering. Even if you correctly identified the AI images above, how confident are you that you’ll still be able to do so with 100% accuracy in two years’ time?
How does it work?
To be clear, I am not an AI expert and would struggle to explain how AI generates content. However, I’m in good company. In a recent podcast (which I can’t now locate but which I definitely did not dream), a number of tech company AI gurus were asked to explain AI’s generative process. Their responses can best be summarised with these emojis: 🤷🏼🤷🏼♂️
“Err, we’re not entirely sure.” is a fair summary of their responses. Nevertheless, I’ll have a crack at it.
Essentially, AI machines read millions and millions of website pages and images and videos and books from the internet and, using enough electricity to power a small industrial country, are trained to analyse the data in order to create predictive models that can generate new (I hesitate to say “original”) content.
AI machines working with images have been trained using the photographs and videos that you and I have shared on our websites and on social media. Nobody asked our permission. It just happened.
One of these videos of a starling murmuration in Rome was made by me, one is AI-generated. I’ll let you work out which is which. Should we be offended that tech companies have used our photos and videos to train their predictive models without our consent? Some people think that we should.
Objections
In a development that might accurately be described as trying to shut the stable door long after the horse has bolted, creative people are raising objections to AI companies’ uninvited use of their (our) work. In the UK alone, 48,000 creators are petitioning the Government, hoping to avoid an exclusion that would allow AI companies to scrape copyrighted work without asking or reimbursing the copyright holder.
What would this mean in practice? Andrew Lloyd-Webber, one of the creators speaking up against the bill, says, “AI companies would be allowed to take works, past and future, and use them as training data without consent or payment.”
Paul McCartney has “backed calls for laws to stop mass copyright theft by companies building generative artificial intelligence.”
There is a suggestion that creators could “opt out”. “Under the proposals, tech companies will be allowed to freely use copyrighted material to train artificial intelligence models unless creative professionals and companies opt out of the process.” (Guardian)
Let’s be honest, the chances of people bothering to opt out are slim. It ought to be an opt-in process, where content creators who are happy to share their work for AI generation can say so. But the tech companies are too powerful and that’s never going to fly.
I hope the campaign to overturn the suggested legislation succeeds but, to me, it seems like a moot point. Creators demanding reimbursement from AI companies are like a lorry-load of hens demanding payment for eggs they’ve previously laid, not realising that they’re being driven at some speed towards the abattoir.
The eggs are already omelettes and we’re all destined to become Coq au Vin.
Authenticity issues
One doesn’t need a vivid imagination to think of areas where the use of AI-generated material could (will) be problematic. But here’s one example that I’ve encountered personally.
The nice people at Adobe run Adobe Stock, a platform that started life as a traditional image library. I’ve sold images through Adobe Stock and it’s been profitable for me in the past. Not so much, I suspect, in the future.
AI is heavily promoted on the Adobe Stock website. Users can alter an image, generate new content, change style and composition… There are infinite possibilities.
Adobe Stock is not an editorial or news library. However, it does host thousands of news-style images created by AI.
“A young Israeli woman, wounded, clinging to a soldier’s arms in anguish. A Ukrainian boy and girl, holding hands, alone in the rubble of a bombed-out cityscape. An inferno rising improbably from the tropical ocean waters amid Maui’s raging wildfires.
At a glance, they could pass as iconic works of photojournalism. But not one of them is real. They’re the product of artificial intelligence software, and they were part of a vast and growing library of photorealistic fakes for sale on one of the web’s largest stock image sites until it announced a policy change this week.” (Washington Post)
I did a very quick search on Adobe Stock, using terms like “Gaza conflict”, “Ukraine war” and “Syria refugees” and found hundreds of AI images. Here’s a random selection of 15 images. 12 are AI generated, three are not. Some are probably easy to distinguish, others are not so obvious.
If you have an Adobe account, you can see the images in a slightly larger format in this public library collection I’m sharing.
You can see which three images are “genuine” at the foot of this post but before you scroll down, please read on.
Custom libraries or galleries are a common way for researchers and editors to share collections of images. You might notice (if you have an Adobe account) that in the public library I created there’s no obvious indication that any of these images are AI-generated. It does say “AI generated” on the Adobe Stock website but that disclaimer does not appear beside images in shared libraries.
Adobe say they’ve changed their policy about AI images but, ultimately, it’s in Adobe’s interests to sell image files, so there’s arguably little incentive for them to limit the potential market!
There are two reasons why this bothers me. The first is partly emotional.
Why does it matter?
All of those AI images of conflict, civil unrest, refugees etc. have been generated using models trained on the photographs made at great personal cost by photojournalists. When Lynsey Addario or James Nachtwey or Marcus Yam head towards a war zone or to the site of a natural disaster, should it be acceptable for a keyboard warrior eating Cheetos in their underpants from the safety of their parents’ basement to churn out a pastiche of the photographers’ hard work in a few seconds with a quick text prompt?
It’s a rhetorical question, at best, because it’s already happening and, whether intentionally or not, those AI images are finding their way onto websites, where they are presented as editorial news photographs.
Everything about that pisses me off.
The creators of the original photographs are not recognised or reimbursed. Their livelihoods are under threat from AI knock-offs. Somebody eating Cheetos in their underpants is potentially being paid more handsomely than a photojournalist in a war zone. I’m not suggesting that eating Cheetos in one’s underpants is entirely risk-free but if a Cheeto-eater ends up in hospital, I fear they only have themselves to blame and questions should be asked.
Secondly, it raises the question: what happens next? Perhaps the media outlets with bigger budgets will continue to commission photographers to document world events. Although that’s by no means guaranteed as budgets diminish and agencies shrink.
Beyond Reuters, Associated Press, the New York Times, and Agence France-Press, how many media outlets and online platforms will be able to justify sending a photographer to cover news events?
Even if events with global impact are covered, what about smaller, local events. Is an editor looking for images to accompany an article about the annual return of cattle to summer meadows in the Alps really going to pay for a stock image, let alone commission a photographer, when a reasonable facsimile can be generated in seconds for a few cents?
(This AI example was created in five seconds using an old, free and relatively basic model. More sophisticated AI engines are being released every day.)
If there’s a good argument for why photojournalists and stock photographers can remain optimistic, I would love to hear it.
Reasons to be cheerful
In an attempt to understand how AI might be used more positively, I’ve been using it to help manage my Lightroom archive. Even though I try to carefully caption my images and I spend time adding relevant keywords, there are hundreds of thousands of images in my archive and even my best efforts inevitably fall short.
I’ve learned that AI can do a pretty good job at searching my archive really quickly for specific content. For example, off the top of my head, I searched for “green trousers” because the colour of trousers is not the level of detail I’d include in a caption or accompanying keywords. AI has done a reasonable job of searching my entire archive and returning images which include green trousers. The process took one or two seconds.
So far, so good.
Beauty is in the AI
More problematic, perhaps, is the claim that this AI product can measure an image’s aesthetic value.
I asked AI to show me the most aesthetically pleasing images in my archive. The three most appealing photographs, according to AI, are a long-forgotten landscape photo of the Gatesgarthdale Valley in Cumbria, a portrait of a woman praying at Gangte Goemba in Bhutan and a naked, weed-smoking Sadhu at the Kumbh Mela in India.
I’m on board with the first and second pictures but is the naked Sadhu really the third most aesthetically pleasing photo in my archive? Of course, beauty is in the eye of the beholder and I’m not suggesting he isn’t aesthetically pleasing. It’s just not a photo that I’d put at the top of my list. Hey ho, it’s all subjective, but I’m tempted to ask AI if it’s been smoking the same substance that the Sadhu was enjoying.
The company, Excire, recently ran a competition for landscape photographs. Entries were judged instantaneously by their AI machine as soon as each image file was uploaded. And this is where they lose me.
Using AI to locate specific content or to assist with the process of adding keywords is helpful — providing we ignore the fact that photographers have been cannibalised in the process of creating the AI model and the power required might speed up the destruction of the planet — but is judging the artistic merits of a photograph something that we’re happy for AI to do? I really feel that humans, with all our inconsistency and subjectivity, are best left to make these sort of judgements. Where will it end? How long before Instagram is mostly hosting AI-generated images that are “liked” by an army of AI bots in a self-perpetuating AI spiral? I’d give it a week or so.
Just because it can be done doesn’t necessarily mean that it should be done.
Excire’s competition results suggest that to increase chances of success in a contest judged by AI, one should include the Northern Lights.
Conclusion
For what it’s worth, this is what I’ve concluded:
The process of building AI models has already ripped off creators and it will continue to do so.
There may be some mitigations, like Nightshade, which claims to protect images from AI scraping. Perhaps the proposed “opt-out” legislation will be the answer. But, really, let’s be honest, we are never going to slow the inexorable progress of the AI beast.
I can’t think of a future model where the endeavours of creative humans are protected. That might sound melodramatic but in a world driven by the insatiable desire for commercial gain, what space will remain for writers, photographers, filmmakers, actors, poets, etc.? Some might claim that the difference between human and AI-generated content is obvious and that’s where our value lies but even whilst that remains true — and it won’t be true for much longer — I’m not sure most people really notice or even care that much.
I might ask Chat-GPT to write my next newsletter and see if anybody notices. Perhaps AI wrote this. Who can say for sure?
Answers
The AI images are highlighted in red. Even though the other three are not labelled as “AI”, I was suspicious about the one of the Ukrainian flag, middle row on the right. It turns out that this was “3D rendered”.
All this AI talk has been tiring and I’m heading to bed. Whilst I sleep the AI machine will continue to gobble up our work and regurgitate it in infinite variations. Please, don’t have nightmares.
I’m hoping not to have the recurring time-travelling dream. I don’t think I have a good answer for those 1975 kids. They were hopeful for a future world with digital cameras and mobile telephones and electric cars and rockets to Mars and jet-packs. Instead, they’ll get climate change, artificial intelligence and Elon Musk. And not a jet-pack in sight.
Sorry, kids.
If you have thoughts about how AI will affect our future, for better or worse, I would love to hear. If you have views on AI and copyright, please share a comment. And if you have any specialist knowledge, tell me what you know.
Until next time, go well.
The Directory contains a full list of newsletters, tagged and searchable by content.
The Resources page contains links to recommended photographic resources, including books, camera gear, accessories, software, contests, grants, and online tools.
Beyond the Frame is a reader-supported publication. The newsletter is a labour of love that only exists thanks to the patronage of readers. Please consider joining the community of supporters. Your support makes all the difference.
A nicely written, thought-provoking piece, Gav.
AI has the ability to completely change our world for the better. Unfortunately, it can also make it worse... depends on how it is used, because there will always be those who use it for their own gain, either at a personal or corporate level.
There is no doubt that AI will take away jobs and, sadly, those at the creative end of the working spectrum look like they are most likely to be affected. I doubt if anything will - or even can - be done to protect them.
Some of those photos you have shown, look fantastic, yet, as soon as I know they are AI generated, I lose interest in them.
Knowing that it was created in seconds by an algorithm, using a multi-billion pound infrastructure of data centres and computing power, does nothing for me (other than being impressed by the technical aspects of it all).
Whereas, someone with a 500 quid camera and a laptop, and who has probably spent hours in finding and photographing that particular subject and then possibly even more time in cropping, tweaking and honing it, so as to present us with the best picture they possibly can... well, that gives me something I can relate to.
But, as you say, that only applies if I can still tell the difference.
Keep up the good work.
Thanks for the engaging, fantastical journey, Gavin. If only AI could remove toxic public figures from our consciousness...demons who are everywhere in our hair these days.
I will keep praying for that moment in time!