Worthless images

When you look at a drawing made by someone, you seek the meaning and intention conveyed by the person who made it. Aesthetics and form are like a language of their own. In the case of AI generated images, the prompt given to generate something is the intent itself materialized in words, written language. The visual aspect loses any relevance in this case, as it becomes a mere obstacle between the maker and the viewer. It would be easier to simply read the prompt that generated that image and, perhaps, use your own imagination to render the image in your own head, thus leaving some space for subjectivity and subtext. Do this now, imagine the image generated by the following prompt: “generate a Van Gogh painting of a dragon surfing on Copacabana beach”. What did you think of this image formed in your head? Funny or unusual perhaps? Two things are certain: it will not be the same image I imagined, and you will never know what a Van Gogh painting of this dragon would have looked like, since he has been dead for over 130 years and did not have a specific taste for painting this type of scene. I could write that prompt into any AI and visualize something created from there, and if you saw it too, we would all have the same concrete image of what a dragon surfing in the painter’s style would look like, without room for any individual subjectivity. However, that doesn’t even seem to be the worst part of this case.

What is especially strange to me is that when I see a Van Gogh painting, I look closely, I look from afar, and I try to understand what led this mad Dutch man to paint in that way. What was afflicting him that day that made him make a brushstroke in this or that way? Deep down, if Van Gogh hadn’t existed, there would be no prompt to generate his own works. And I don’t say this lightly, to say that AIs don’t generate anything new. No! I say this because the painting, in this case, is a manifestation of that exact historical and personal moment in which he found himself. There is no intermediate text where the painter gave himself precise instructions on how to produce his painting. If there had been, it would perhaps be even more interesting than the painting itself! Since it would allow us to access the truth of what lies hidden behind the visual enigma. In the absence of a generating text, the enigma is the meaning itself, as it is an enigma even for the author! He himself often doesn’t have a clear understanding of what he is doing and wouldn’t be able to concretize what is produced in a descriptive text. It would be a true torture to force an artist to first, before beginning to paint, objectively describe what he wants to do, even in terms of style, facial expressions, and brushstrokes. Even in the case of commissioned or planned works, as in the case of the Catholic Church during the Renaissance, artists like Michelangelo could not hide their subjectivity and spontaneity when painting what was commissioned. He was not hired because he could perfectly reproduce what was described, but because he produced something considered divine at the time, something no one else could do. When I look at his paintings, I come into contact with an enigma that is aesthetic, personal, and historical. “Did Michelangelo want to say something with this? Was it something profound or something banal? What did he feel when he painted this?” We probably won’t know the answer, but we do know that a person who died 460 years ago left us this enigma, and it’s up to us to try to decipher it (or not). One thing is clear: when we look at these paintings, it’s not just the image that matters, but the entire process of creating it and getting us there.

In the case of an artificially generated image, the only true material is the textual prompt itself, which is usually intentionally hidden, as if it actually revealed the truth of what one wanted to convey, and this was of less value than the image. Unlike paintings or drawings made by people, the person’s intention is linguistically materialized, giving us a unique opportunity for the author to explain explicitly and concretely what they wanted to produce. On the other hand, the generated image has low information density. Even if the image generally reproduces what was described, the details don’t convey information and don’t carry any specific meaning. Making an inaccurate mathematical analogy, it is as if the Fourier transform of the image in terms of “spatial distribution of meaning” was only high in the low-frequency range. If the prompt I wrote earlier had been used to generate the image, we couldn’t interpret in any specific way the fact that one of the dragon’s eyes was painted red or green. Or that a brushstroke was made heavier or lighter. These things have no meaning and thus discourage any attempt to interpret them, since it’s most likely that there is no enigma at all. Furthermore, we know that the enigma, if it exists, has a concrete solution (the text), thus we want access to that text. It’s much better if we try to decipher the text itself! In this case, the textual prompt becomes much more interesting, since it can reveal something about the person who wrote it and can also bring nuances and mysteries about them. The text is the only true material, and the generated product is irrelevant since it is simply an interpretation of instructions, something I could have done myself. Therefore, instead of looking at AI images, I would rather simply look at the prompt. Instead of images generated in this way, I would prefer that people leave blank images on the internet, with a text prompt in the middle describing what one would like to have there; I use my imagination for the rest.