A friend recently pondered via a social media post whether we will have photography as we know it in the future, or if artificial intelligence (AI) will soon generate all of our images. With tens of millions of people now capturing snapshots on their phones’ cameras and instantly applying AI-generated filters to enhance or modify the images, we can certainly observe an increasing trend toward computer involvement in the making of photos. But I don’t believe AI will replace the artist’s eye in the making of fine-art photographs for quite some time to come. Here are a few semi-random musings on this theme.
A machine can certainly generate bad art. In college in the mid-1980s, I wrote a program for my Computer Science final exam that composed musical canons (pieces in which each voice plays the same melody together, but starting at different times). My code used a semi-random configuration of musical intervals as the opening melody, then applied a simplified set of the rules of counterpoint (how musical lines are allowed to fit together) to complete the canon. I received an “A” for this project, but truth to tell, any listener familiar with classical music could instantly discern that the pieces composed by my program weren’t anything like the lovely canons written by Telemann, for example. In other words, my AI didn’t pass the Turing Test.
In the more than 30 years since I wrote that program, AI has progressed by leaps and bounds. Computers can now generate poetry, classical and jazz music, and even paintings that many non-experts judge as products of human artistic creativity. I’m fascinated by the progress, but so far the best of the AI-generated “art” is really just imitation and trickery: it takes a seed of something original such as a photograph or a melody, and transforms it using a set of complex rules that could be described as a pre-programmed artistic style into something pleasant enough but not inspiring.
In his landmark 1979 book, “Gödel, Escher, Bach: An Eternal Golden Braid,” Douglas Hofstadter amazed the world by demonstrating comparable interlocking themes of grace and elegance among the very different disciplines of mathematics, visual art, and music. He even speculated on the ability of machines to create works of great insight. But Hofstadter’s proposed approach differed from that of the AI field that has developed since then in that he favored teaching machines to create via an understanding of how the human mind creates, as opposed to today’s AI approach of taking mountains of data and throwing brute force calculations at it. To my eye, ear, and mind, this brute force method is the reason most of today’s attempts to artificially emulate the creative process are not insightful and do not add anything to their genres. And so far, the vast majority of these attempts fail their respective Turing Tests. That is, humans can tell it is a machine and not a human generating the “art.”
Applying these musings to the art of photography, what do see today? To be sure, more images are being generated today than ever before in human history, and the art of photography is being devalued by its sheer pervasiveness. Everyone captures images now, and most of them believe that makes them photographers. While photographers have always required the involvement of a machine in the creation of their art, good photographers have always relied on their artistic vision, the so-called artist’s eye, to create images that are special. I don’t believe that all the Meitu and similar AI filters that abound today are creating any photographic art that adds insight or helps interpret the world around us.
One very central component in photography is composition. How does the photographer choose what elements to include in the image, and how will these elements be combined? Read my recent post on composition here: Post on Composition. This vital aspect of photography does use some “rules”, such as the Rule of Thirds, Leading Lines, Framing Elements, Point of View, Foreground/Background, and Symmetry and Patterns. Rules, of course, can be programmed into an AI so that the machine can emulate the way humans create. But in photographic composition, the “rules” are really just guidelines for getting started. A good photographer knows when to break the rules for artistic impact.
Even the dumbest devices are capable of generating images. Security cameras can capture images that we would consider to be rudimentary documentary photographs. Given long enough, a security camera might accidentally capture what we would consider to be a good street photography image, because after capturing millions of dull scenes, sooner or later the camera will catch a random alignment of interesting elements. It’s like thousands of monkeys typing random characters: given enough time, one of them will coincidentally type out a Shakespeare sonnet or even a full play. As wearable computing devices become more pervasive, many people’s lives will be documented in real-time via the capture of millions of images. Some of these may be interesting to their friends and perhaps the general public. A few may even have artistic value. But true artistry isn’t characterized by coincidence.
I don’t doubt that eventually we will get to the point where machines can create images as good as much of what humans can create. I think we’ll get there, but it will take a long, long time. And in the meantime, the role of photographer as artist, experimenter, and interpreter of the world around us will continue to be central to our society’s need for communication and expression.
What do you think about the future of photography? Will we soon see machines creating much of our imagery? How about our good, artistic imagery? Please share your thoughts here.