The Authorlessness of AI Art
“When I consume AI art, it … evokes a feeling. Good, bad, neutral—whatever. Until I find out that it’s AI art. Then I feel deflated, grossed out, and maybe a little bit bored. This feeling isn’t a choice,” writes Matthew Inman in a visual essay.1
The idea that the worth of a work of art depends on who created it may sound unreasonable or snobbish, but in this case it actually makes perfect sense, though maybe not in the way you might think. To understand Inman’s reaction to AI art it’s important first to define clearly what art actually is. This will help explain what, if anything, separates AI art from human art. The clearest definition I’ve come across is by Leo Tolstoy, who says that art is a human activity in which the artist, by means of words, pictures or sounds, transmits the feelings he or she has experienced to others, with the aim of uniting them in the same feelings.2
Now, you may disagree with Tolstoy’s definition, but it does point at a certain activity that can be clearly recognized as art, or at least a significant subset of it. This, I think, is precisely the activity that Inman is referring to when he speaks of art, and Tolstoy’s definition explains not only what makes AI art different from human art, but also why Inman finds it so empty.
To transmit a feeling, you must first be able to experience it. Assuming generative AI models aren’t sentient, the feeling that is transmitted cannot be that of the AI. If a feeling is being transmitted, then it is either that of the person writing the prompt, or that of the original artists whose work was used to train the model.
The former cannot be true because the person writing the prompt would have to fully express the feeling to the AI model before it could create a work to transmit it, but if the feeling could be fully expressed in a prompt then there would be no need to generate the art—the prompt alone would be enough, and would itself be art. Tolstoy actually mentions this in relation to summaries of works of art by noting that they can never adequately express the feelings they are talking about because those feelings cannot be expressed by anything less than full work. Art cannot be compressed.
This leaves us with the latter: the feelings of the artists the AI model was trained on. The AI model’s weights clearly reflect the choices of human artists, and it uses those choices when it generates its images. Can we go so far as to say that an AI model transmits human feelings? Because the diffusion process used by these models blurs the training data, what is being reproduced is that which is shared across a multitude of images: the superficial style and technique of the art, its “vibe” if you will, rather than its content. If a feeling is being transmitted, it is limited to this surface level style. The content, on the other hand, is dictated by the prompt, which, as we have seen above, is constrained by its brevity. But even the style does not really transmit human feelings directly, because while the output is guided by human decisions, it is not the work of any one artist. AI art erases human authorship.
A loose comparison can be made here to the advent of mechanical mass production in the 19th century, which was much criticized by prominent designers like William Morris for its effects on both craftsmen and their work. Craftsmen, who up to that point decorated their goods freely with their hands, were turned into the thoughtless cogs of the mechanized assembly line. Meanwhile, ornamental forms, which were difficult and time consuming to produce by hand, became cheap and easy when they could be cast or stamped by the new machines, and many designers abused this opportunity by plastering decorations all over their work. Mechanical mass production thus simultaneously took away the most creative and fulfilling part of craftsmen’s work and drove design towards an excess of ornament because it could now be cheaply made. Design movements of the early 20th century in turn revolted against this excess by stripping away all ornament from the surface of their work to leave only the simple, geometric forms beneath.3 What modern designers found objectionable wasn’t the ornament as such, but its thoughtlessness and excess—a consequence of new production methods. Even crude ornament, carved by hand, is valuable as an expression of the human spirit, but shoddy ornament that is mass produced is worthless.4
The problem with AI art isn’t that it’s bad, but that it’s authorless. It may evoke a feeling, but it doesn’t transmit a feeling, because there is nobody on the other end.5 When Inman discovers that an artwork was created by AI, he feels “deflated.” Even though moments ago he felt the emotions evoked by the illustration itself, the knowledge that a work is authorless strips from it the quality of human connection, the quality of being united with another human being in the feelings they’ve expressed. And it is this connection that is, ultimately, the whole point of art.6
Matthew Inman, The Oatmeal, “A cartoonist’s review of AI art.”
Paraphrased from Tolstoy’s 1897 book What Is Art? This definition also makes it possible to evaluate the moral quality of art, since it depends on what kinds of feelings are being transmitted.
Going, one might argue, too far towards the other extreme of minimalist blandness. Modern function-oriented design was, in turn, challenged by postmodern design, which brought back design elements for their own sake.
Another thing worth mentioning is that a work of art involves work. The effort expended on something is itself a signal of value. By delegating this effort, the perceived value of the work is reduced. This is something I touched on in my previous post on AI writing.
This is true in relation to wholly AI generated art, but artists can also use AI to assist their work without wholly replacing it. For example, they can use it to generate assets and textures to use in combination with their own drawings. This way of working isn’t new. Old masters often delegated less important parts of their paintings, like backgrounds, to their apprentices. Such hybrid art blurs the line between AI and human art, which means that the amount of feelings being transmitted will depend on the amount of human input there is in a work.
While the connection between the artist and the observer may not exist in an AI generated work, there remains the connection between the observers themselves. But the AI work here is akin to a natural phenomenon, such as a beautiful panorama, which unites those who share the experience of viewing it in the feeling it evokes in them, but does not itself have a human experience as its source. The beauty of art is that it reveals a panorama of the human soul.