“As far as I am concerned, this is not photography”, some opinionated guy said to his bemused friends. A lively discussion followed, before everyone decided that the meeting had overrun, and it was time to stop. This happened last week on the regular call that I have with my fellow students. The guy? That would be me.
We were discussing a topic that, unless you’ve been living in the wilderness for the past year, has been a regular news item on almost every news outlet; that of Artificial Intelligence, or AI. We got onto this slightly controversial chat because one of our group had uploaded an AI ‘photograph’ for our image challenge, just for the fun of it. The conversation that followed was fascinating to me because my strong, and not particularly humble, opinion was met with some great arguments in support of the technology. Our group is made up of very talent photographers, each with different experiences, genres of interest and styles of learning, which results in us all learning from each other. After this call, I was left reflecting on what AI really means for the creative arts and whether I will ever likely use it.
Artificial Intelligence, outside television and the blockbuster movie space, has been around for decades. The idea of using machines to learn and think was postulated by Alan Turing, widely recognised as the father of modern computing, in his 1950 paper Computing Machinery and Intelligence, a lengthy debate about whether intelligence can be truly ascribed to a machine and how it might be tested. Turing was limited by the technology of his time which though, unknown to most civilians, included computing machines that he developed during his code-breaking work in the Second World War. With the evolution of computers, came the inevitable continuation of AI research, and eventually it started to become part of our everyday lives at some level or another. We have the obvious examples, such as satellite systems in our cars, that can plan optimal routes and automatically adapt them with changing traffic conditions, smart speakers that can recognise voice commands, shuffle music that we might like etc., but we also have the more subtle uses too. Social media, for example, uses sophisticated learning algorithms to present us with recommendations and tailored advertising based on what it they see as our interests. We also have plugins for commonly used software like, well, Adobe Photoshop. Being a regular user of its sister program for many years, Lightroom, I’ve seen the tools become more sophisticated, but its only recently that they added AI in the shape of noise reduction to the suite. This kind of software has existed for a few years in other produces like Topaz Denoise AI, but it was only this change in my own workspace where I noticed the considerable difference in performance to how I was working previously. Prior to the current version, reducing noise in a photograph was a process of trial and error, selecting the pixel size, deciding how much detail to preserve, masking etc. Without using batch editing, it could be a fairly time-consuming process. The new version presents the simple option of ‘how much do you want?’ as a percentage and takes care of everything else. I used it for the first time a couple of weeks ago and, despite causing my elderly Macbook Pro some pain, it yielded impressive results. So, AI for supportive tools that take more and more workload off is brilliant and harmless, right?. Well, maybe not. In his paper, Turing said of the objections to machine intelligence “The consequences of machines thinking would be too dreadful. Let us hope and believe that they cannot do so.", referring to the idea that ‘thinking’ machines could eventually lead to our subjugation or, worse still, our destruction. Such fears have been creative fodder for the movie industry, with such films as Terminator, or the scarily realistic Ex Machina. The news items that are around us daily, talk of people getting ChatGPT to write something that is passed off as original, or as in the context of my student call, creating ultra-realistic images that could be mistaken for actual photographs. The image that was submitted in our photo challenge was created by an online AI research platform called MidJourney, and in essence it works by using a series of text descriptions, similar to metadata, to create an image. The software can also analyse an existing image to determine a set of text descriptors, and this is what my friend did to respond to the challenge’s brief. He got MidJourney to analyse our ‘theme image’ and then used its own description as the information for it to create a new image. When I was preparing to write this blog, I asked him to do the same with one of my photographs to see how close to the original it might be and, as I hope you agree, the results were remarkable.
Abbey Mill, Tewkesbury, by Richard Fletcher
This photograph was shot with on film using a square format medium format camera. Apart from dust removal and small adjustments to create the final photograph, it’s pretty much as shot.
Output from MidJourney using the text: "black and white image of mill and bridge up river, in the style of dramatic movement, contemporary landscapes, hasselblad 1600f, vibrant airy scenes, dark sky-blue and white, shiny, suburban ennui capturer"
MidJourney created a number of versions for me to choose from and I’ve included the ones that best represents the original. We see a building on a bridge, over a fast moving river, set in dramatic stormy conditions, much like mine. The key elements work, such as the long exposure of the water, the light is coming from a consistent direction, the look and feel of contrasty black and white photography, and the aspect ratio. I think it’s a pretty good effort with relatively little actual data.
Did I have an epiphany when I saw these, to the extent that I re-evaluated my opening comment? Well, not really. I certainly do not see this as ‘photography’ because the term inherently means ‘drawing with light’. I don’t see photography in the traditional terms of being pure, truthful or incorruptible, because that’s all frankly nonsense. Artists have been using photography to subvert stories or perspectives on the world since its conception. We’ve been able to modify images to create rhetoric and alter narratives with ease as digital camera and editing technology has advance. At its heart, though, it is a medium that represents what a real object or situation, even if the artist conceptually plays with how that representation might be interpreted by the viewer. What I acknowledge after this exercise, and with some further prompting from my fellow friends, is that AI’s birth as a creative tool is very similar to that of photography, which suffered derision and suspicion from the painters of the mid-19th Century. It evolved into a creative art in its own right because an artist used a tool to be creative. All that has changed in the case of AI is the ability to create in a non-visual way. The output is digital art, which is still art if it speaks to people about something related to the human condition. If the spark of creativity comes from a different source and the computer does the leg work, who really cares? After reading an article about how the use of text influences the context of a photograph, something the viewer is unaware of when looking at an AI image, I got to thinking about how this technology might work when in the hands of a true wordsmith. If, for example, a poet were to write the text descriptors, what would the resulting image represent? Would the AI be able to articulate, in some form, the poet’s ideas visually? If a brilliant writer like the late, great Terry Pratchett had used AI, would we have a different sense of wonder when thinking about his fantasy realms? Where would this ‘intermediate’ interpretation leave the ideas that Barthes’ postulated in The Death of the Author, where the burden of interpretation shifts to the reader (or viewer in this case)? Sometimes, as in the case of William Fox Talbot, experimental technology unwittingly takes our skills in other areas to whole new levels. Fox Talbot was a frustrated sketch artist, who used a camera obscura to help guide his pencil, before wanting to find a way to permanently freeze the image. His technical mind created one of the inventions acknowledged as the beginning of photography, the silver and salt process. Could a similarly inventive person take AI and digital art into a whole new direction by trying to overcome a creative problem? I think so, which is why my perspective has shifted from where I belligerently started. Yes, there are valid concerns about where AI images could find their way into a chain of evidence or derail a criminal prosecution. It will undoubtedly get the ‘Fake News’ brigade excited too, who are keen to present a ‘truth’ that could do with some much-needed credibility. Then again, if we need technology to protect core values like our courts, schools, media, etc, that will need to be developed alongside. The media portrayal of the evil genius waiting to take over the world has been with us throughout the work on the human genome, the development of fertility treatment and genetic engineering of vaccines for so long that we are naturally afraid of anything new and complicated. For now, I think AI is actually a potentially powerful tool, as long as we see it for what it is and what it is not. Also, the engineer in me can’t help but want to play with it myself, now that I’ve seen that potential. As a photographer though, I’m not interested in it forming part of my work, choosing instead to happily shoot with my ever-growing collection of film cameras. The former may influence the latter in the end, though. If that happens, I guess I’m going to Photography Hell. Could someone put the kettle on when I get there?
Link to the article about AI and Text: https://www.lensculture.com/articles/gregory-eddi-jones-49-23-considering-technology-ai-and-photography
My thanks to Mark Racle for driving MidJourney and to my fellow students for widening my gaze.