Photographer Boris Eldagsen recently caused surprise and a flurry of discussion in photography when he won in a category of the Sony World Photography Awards with a computer-generated image he had made using AI or the DALL-E 2 generative neural network.
Eldagsen claims his intention was not to mislead and he rejected the award at the awards ceremony as he felt the organizers were not talking about the image being synthetic. His goal, he says, has always been to spark debate about the impact of these technologies on the way we think about photography. He also clarified his position, arguing that these computer-generated images are not photographs and should not be accepted in photography contests. But is it so simple?
In a later interview with the BBC, Eldagsen described these images as “promptography” and not photography, making the distinction that a real photograph is made from light reacting with a sensitive surface, whereas these photographs are the result of prompts entered into a neural network. However, this description obscures the rather more complex and murky reality of how these neural networks can generate these images.
In order to generate such impressive and realistic photographs, these neural networks are trained on huge datasets of millions of pre-existing photographs, allowing them to train the “neural” connections needed to take a textual prompt and transform it into a photorealistic image. . In a sense, these systems don’t produce anything new at all – they synthesize new images based on the data points of pre-existing photographs.
Through this they “learn” how light and lenses interact to create images in a conventional camera, but they themselves don’t, so in a way their outputs are almost closer to collage or 3D modeling than conventional photography. The problem here is that these systems struggle to generate images of things they haven’t been trained on, and so this will always be a major limitation to their creativity.
As Eldagsen said in a meeting “Photographic language has become a free floating entity separate from photography and now has a life of its own”. At the same time, it should also be noted that computational and generative photography is not exactly new, and we tolerate a wide range of post-processing effects being applied to photographs that have no direct relation with light, lenses and other things that we associate. with traditional photography. Mobile phones are increasingly using neural networks to enhance images from their cameras, sometimes changing them dramatically in the process and producing an image that would not be possible through optics alone. There is therefore also a happy medium between traditional photography and computer-generated imagery, that of “assisted” photographs which combine the best of both worlds.
However, perhaps part of the problem with this debate is that photography is used for so many different purposes, and to talk about them all in the same breath is too disgraceful to be helpful. There are genres where we might agree that undisclosed use of these images is problematic, such as photojournalism, where the potential for abuse is enormous and could have genuinely dangerous consequences.
Synthetic images of current events are already circulating widely on social media (such as a recent image of Presidents Putin and Xi), and in my own research I have found that there is enormous fear in the offices of newspapers about the dangers of news outlets using any of these images in error. Perhaps that matters far less in the context of art, where these generative neural networks are a potentially powerful tool for expression, as Eldagsen himself argues.
But a final question is whether the debate should focus less on whether these images count as photographs, and more on the moral good or evil of their functioning. There is growing evidence that the training data for many of these neural networks relies on copyrighted images of existing photographers, and a growing number of lawsuits are being filed against the companies to the origin of neural networks. Beyond the rights and wrongs of the images themselves, we should ask ourselves if it is fair that photographers end up losing out financially to systems that are only made possible in the first instance through their photographs.
