"a photograph of a raccoon playing the banjo on a boat" – *enter*
As DALL-E 2 processed the millions of pictures it was trained on to turn my prompt into a photorealistic image, it was hard not to feel slightly overwhelmed by this marvelous machine. It looks like magic.
Amid the excitement of seeing this state-of-the-art system, the raccoon looked at me with their synthetic, AI-generated eyes as if asking, "what kind of art am I? Am I even real?"
"Of course!" I replied, "what's a better representation of reality than an art built of fragmented, out-of-context information and influenced by the preexisting societal dynamics of our time?"
Is it art?
My perception of AI-generated photos as magical is driven by the fascination that machines – previously perceived as merely computational – have now learned to make sense of our language. Let alone be creative.
But as my excitement around AI image-generating platforms grew, so did the public debate around its legitimacy. When an "AI-generated picture won an art prize," many did not see it as magical but artificial. After all, a synthetic means of creation can never be as real as human creativity. Right?
To answer that, we need to understand what AI is – not in the technical sense, but the social sense. In her book Atlas of AI, Kate Crawford argues that "AI is neither artificial nor intelligent. Rather, artificial intelligence is both embodied and material, made from natural resources, fuel, human labor, infrastructures, logistics, histories, and classifications". In other words, AI is not this mysterious system sitting in the "cloud", but a human-made system that is influenced by the social and political forces of our society. Furthermore, Crawford claims that in the process of collecting data, it is often stripped away from its original context. The data then is given a new meaning in the classification process, which can reinforce biases and other historical influences.
Looking at "AI art" from this perspective strips away some of its magic. It raises questions about how the millions of images used to train DALL-E 2 were chosen. Which images were not chosen? How where they classified? Which biases were taken into account? What do the images that AI systems produce really represent?
Is it real?
The medium of AI-generated photos will never generate "real" images, but so do human-taken images. In her 1977 text, In Plato's Cave, Susan Sontag addressed this question of "realness" in photography and stated that "photographs are as much an interpretation of the world as paintings and drawings are." That is because, she explains, in capturing photographs, the photographer frames the world, leaving most of the image outside the camera's lens.
Using Sontag's logic, one might argue that synthetic media (with AI-generated images included) drives humans even further away from reality, as it blends millions "of interpretations of realities" into something entirely out of context. I, however, believe it is not the case. Instead, synthetic media presents a new kind of realness – one that does not come from the observation and interpretation of a single artist but a collection of all the data used to create it and the influence of those who classify it.
The image of the raccoon playing the banjo on a boat is all the images that were chosen to train DALL-E 2. It is the representation of all the images that did not make it. It portrays the way the people at Open AI classified the data.
If our online communication is fragmented and lacks context, so does the data used for AI art.
If our society is prone to biases, so does how we classify images.
If anything, this raccoon conveys meaningful messages about our culture in a more real way than many other art forms.
Roose, Kevin. An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy. New York Times, 2022.
Crawford, Kate. Atlas of AI. Yale University Press, 2021.
Sontag, Susan. On Photography. New York: Delta Books, 1977, pp. 3-24.