We certainly live in a peculiar time in human history with regard to Artificial Intelligence (AI) and the revolution it is bringing to many aspects of our lives. From speech recognition to internet search engines, through autonomous vehicles and up to sophisticated chatbots capable of answering almost any question, AI techniques are becoming the heart of many systems we use daily. This, of course, raises many philosophical and ethical questions. Google’s AI chatbot generator LaMDA was recently reported as passing the famous Turing test, leading to a debate between field experts in the question whether it is sentient. One of the most impressive applications of AI that made a drastic improvement in its capabilities is text-to-image generators, such as DALL-E2 and Midjourney, producing digital visual imagery based on any combination of input words. In just the last few months, we witnessed a multitude of such images published on social media.

As a street photographer, I came across many online discussions about this new form of producing images – often indistinguishable from street photographs taken by photographers in the real world – delving into the question “Can these be regarded as street photography?” And more generally, can these AI-generated digital images be regarded as photography of any genre? Is this a threat to the future of photography? Is this even a form of art? I haven’t given these questions any serious thought until I recently failed to identify an AI-generated image as such and believed it was a street photograph. This made me consider the similarities and differences between them. In this article I would like to share my thoughts about these topics. I will immediately start with the short version of my view here for those who are not keen on reading longer texts. My personal opinion is that AI-generated images can not be regarded as photography. They can be considered as a form of digital art, but not as photographs. Ok, wait a minute. Some of you might be thinking “Oh, sure, she is probably one of these stringent old-school technophobic purists, who thinks street photography is nothing but 100% candid unprocessed shots”. Well, I am not. My attitude towards street photography which I shared many times in the past is that merely the photographer’s presence in the scene is by itself part of reality and surely affects it. In fact, the connection between the photographer and her protagonist is often a key element in my work. And hey, I even edit my photographs using Adobe Lightroom, God forbid. I am also an algorithms engineer in my profession (though I admit, not an AI expert), so I can assure you that I am not intimidated by complex technological developments. So now that we’ve established that, let’s dive in.

Starting with the question regarding the genre of street photography, I think most of us can agree that AI-generated images, even when their content includes street scenes, are quite the opposite of this genre’s principles. Street photography aims to capture everyday human life in public places. It is often about finding beauty in the most mundane actions of people in their surroundings. It is about capturing this rare passing moment, that will never repeat again (recall Henry Cartier Bresson’s famous term “decisive moment”, referring to that moment when all the elements in a photograph come together perfectly to create a meaningful image). Being physically on the street, instinctively ready to reveal any unfolding scene and pressing the shutter exactly at the right fraction of a second are the core skills of a street photographer. And those are undoubtedly not the foundations of any AI-generated digital image.

One may question “Ok, maybe not street photography. But what about other genres of photography? Still life with its static nature? Portrait photography? Landscape photographers don’t care much about candid moments”. It is my belief that creating AI-generated images cannot be considered as photography, regardless of the genre. First of all, let us remember the word’s origins. Created from the Greek words “phos” meaning “light” or “to shine”, and “graphe” meaning “to draw” or “to write”, the compound literally means “to write (or paint) with light”. In AI-generated images, no action of recording light returning from any object is taking place. This may sound like a mere technicality, but it is not. It is directly linked to the basic principle of photography.

In his influential book “Camera Lucida” published in 1980, Roland Barthes is examining the very essence of photography. He is interested in deciphering the features that characterize photography and its uniqueness. Here are a few quotes from his writing I collected in order to demonstrate the answers he came up with during his long study. ”In Photography I can never deny that the thing has been there. There is a superimposition here: of reality and of the past. And since this constraint exists only for Photography, we must consider it, by reduction, as the very essence, the noeme of Photography”. ”The name of Photography’s noeme will therefore be: ‘That-has-been,’ … what I see has been here, in this place which extends between infinity and the subject”. AI-generated images are not another leap in photography’s technology (as opposed, for example, to the shift from analog to digital cameras). It is a completely different technology that lacks this fundamental feature of photography Barthes calls “That-has-been”, the pure evidence that something or someone existed in the past. Barthes is fascinated by the tension between the present (the photograph we currently view) and the past (depicted in the photograph). This tension can never emerge from an AI-generated image, since its content never existed at any point in the past.

This is not to say Barthes considers a photograph as an exact replication of reality. He is aware of the photographer’s intent and its effect on the resulting photograph. He states: ”Photography never lies, or rather, it can lie as to the meaning of the thing, being by nature tendentious, never as to its existence”. But AI-generated images may very well lie about the existence of the thing being captured. In fact, if we can be sure of anything with AI images, it is that their content never occurred in reality. There is a built-in lie within them. ”In the photograph, the power of authentication exceeds the power of representation” Barthes claims. In an AI-generated image, I think the opposite holds – the power of representation exceeds the power of authentication. This is why I think they cannot be regarded as photographs at all.

What about art? It may not be considered as photography, but can it be regarded as a form of art? Well, many artists claim inserting a few text sentences as input to a software to produce a visual image is too lame or too easy to be considered a piece of art. I disagree. I think that these images may be considered as a new form of digital art. In many ways the creation of such an image is not very different from many digitally manipulated creations, such as mixed-media art, collages, fantasy worlds created in Photoshop and other forms of digital art. I think it is a valid tool for self-expression, synthesis of ideas and conveying emotions. As such, I think it is a legitimate form of art. Photography and AI-generated images would probably dwell side by side as two different art forms. This is why I don’t believe AI-generated images are a risk to the future of photography.

However, a completely different question is evoked in the context of this new digital art form: Who is the artist? By this I refer both to the ethical question of who is titled the creator of the artwork, as well as to the copyright ownership question. There isn’t a straight-forward answer to these questions. Is the creator the user of the software, that thought of the input sentences that initiated the process? Is it the senior data scientist that developed this AI model, trained it day after day, engineered its features, and optimized its parameters to perfection? Or maybe it is the many photographers that took real photographs and published their work on the internet, without even knowing their photographs will be used to train this specific AI model?

I will not delve into the more ethical aspects of AI technologies and the implications it will most certainly have on one of the fundamental building blocks of human society, which is trust. I feel these topics are out of the scope of this short article, but I do want to end with a call to my fellow photographers, to encourage trust between us all. If you create your own AI-generated images, make sure to mention it properly in the caption when you publish it, and not to use street photography or general photography hashtags. Form a clear separation between your photography work and your digital artwork. If you submit it to a photography competition, make sure you are doing it in the “digital art/digitally manipulated photography” category. Also make sure the contest rules do not require you to be the sole copyright owner of the image. Let us be responsible, transparent and clear to our viewers and to each other.

Sharon Eilon

First published: March 2023

Blog

Can AI-generated Images be Regarded as Photography?