Image: Let the Children Play
Julie Kirsch via DALLE-2
Let us assume for the sake of argument that AI-generated imagery is art. If this is true, who is the artist? The AI generator or the person who uses the AI generator? Neither? Both? Let us explore this question by considering a thought experiment. Amy fashions herself an artist but often finds herself lacking in imagination and inventiveness. She has tried to cultivate her creative powers in various ways—even experimenting with marijuana and psilocybin—but finds herself sadly uninspired. In comparison with Amy, Amy’s younger brother, Felix, is wildly creative and imaginative. Although he creates artwork for fun, it is not his true passion. He is pursuing a career as a science researcher and has no interest in making his own artwork public. He also loves his sister dearly and would do almost anything to help her. Amy would like to submit a piece to an upcoming show and competition for emerging artists but is struggling to come up with something that is unique and technically impressive. She asks Felix if he might help her develop a piece for the show.
They meet over a series of days, Amy presenting Felix with a set of word prompts, and Felix translating these into imagery. Amy asks Felix to create something about “nostalgia,” “childhood,” and “friendship.” Felix created a pen sketch of two young friends walking away from each other with one of the friends, as an older person, looking upon the scene with a slightly melancholic expression. “That’s perfect… you just hit the nail on the head…. Now paint this in cool but muted watercolors and I will be all set. Thank you! You knew just how to put on to paper what I was struggling to say!” Amy submits the piece to a local art competition and wins first place. This helps her budding art career tremendously. She continues to pursue her dreams as an artist with Felix by her side, giving visual expression to her idea prompts.
How does Amy’s use of Felix resemble an artist’s use of an AI art generator? It might be argued that Amy’s use of Felix is different from an artist’s use of an AI art generator insofar as Felix is human and AI art generators are not. While this is undoubtedly the case, we might wonder what, if anything, hangs on Felix’s humanness. There might be moral, social, and even legal issues that surround Amy’s use of Felix, but in both cases the agent as artist is using the “intelligence,” or “creativity,” of something (or someone) external to her for the purpose of producing a work of art that is inspired by word prompts. My suspicion is that many of us would say that Felix, not Amy, created the work of art that she submitted to the art competition. And if, in the Amy-Felix case, we want to say that Felix is the artist who created the painting that he made, then why should we not also say that the AI art generator is the artist, or partial creator, of the art that it produces via its machine learning algorithms?
One problem with claiming that AI art generators are artists or creators is that AI generators are not even agents with mental states, goals, desires, and so forth. We generally reserve the term “artist” for agents. We can, of course, imagine a revised version of this case in which Felix is not a human brother but an intelligent robot with mental states. In this version of the case, we might be more prepared to say that that the AI is an artist. So, perhaps we should think of AI art generators in their current manifestation as quasi-artists or partial creators of the work that they generate. Alternatively, we might devise a new term to encapsulate the causal role that they play (qua non-agents) in the production of artwork through interactions with a user. Putting these issues aside, the Amy-Felix thought experiment shows that Amy’s role in creating a work of art is different from, and arguably more limited than, Van Gogh’s was, for example, in painting The Starry Night. Even if the AI art generator is not an agent or artist, it nevertheless contributes causally to a significant portion of the finished work.
If we make this move, then what should we say about photographers? Are photographers mere partial authors of the work that they create? And is the camera best understood as a quasi-artist? When we compare a Frida Kahlo self-portrait with, say, one of Cindy Sherman’s untitled film stills, I think that we do recognize an important difference between the two. In the case of photography, the camera, like the AI art generator, makes an important causal contribution to the finished work. Still, there is much left for the photographer to do. While the camera may be an astute observer, it is lacking in creativity and imagination, among other cognitive abilities. The photographer not only conceives of the piece, but also controls the lighting, composition, and so forth. The user of an AI art generator, it seems, does less of what the photographer does. After the user enters word prompts into the AI art generator, the AI art generator takes over –makes the magic happen—and produces its results, i.e., a set of images.
Of course, it is even more jarring for us to refer to a camera as a quasi-artist than it is for us to refer to an AI art generator as one. Perhaps as our tools become more human-like—more intelligent and creative—we are more inclined to think of them as agents, authors, artists, or creators. In comparison with AI art generators, cameras may seem “dumb” and unsophisticated. If, in the future, we have the ability to use intelligent robots to create art – robots whose imagery producing powers are accompanied by mental states like our own – then we might be more willing to grant them author, artist, or creator status.
None of this is to deny that users play an important role in creating the images produced by AI art generators. I thought that it would be wise, in writing this series of posts, to give the technology a try for myself. I began with Midjourney, which is attached to Discord, but could not tolerate the interface or figure out how to get started or even where to go from the Discord landing page. (This is perhaps because I am old, impatient, and not very tech savvy.) I found DALLE-2 to be comparatively straightforward and easy to use. However, having researched award-winning works of AI generated art for these blog posts, I was astonished to see how awful my own AI artworks were in comparison. I find Jason Allen’s Théâtre D'opéra Spatial to be truly beautiful and inspiring; it contains multiple figures, a highly detailed composition, and dramatic lighting. The images that I produced via DALLE-2 (which accompany these blog posts) were not off the same caliber. I found that the figures that DALLE-2 generated often had blurry and contorted faces. (See, for example, the faces of the children in the image accompanying this post.) And the images had a tacky and almost cartoonish quality about them. Digital artists who use DALLE-2 and other AI art generators, often edit and improve the images on Photoshop. As a non-digital artist myself, I did not have the ability to do this.
After playing around with DALLE-2 for even a few minutes, I realized that one could be better or worse at producing images by understanding which prompts to use. And so, to be a really great AI artist, one would nevertheless need to possess right set of skills and abilities.
One upshot of this discussion is that different forms of art require different skills and abilities. One could be an excellent oil painter but a terrible photographer; one could be an expert charcoal artist, but a mediocre AI artist. And thus, for many purposes, it may be wise to keep the categories of art separate. For example, it may not be fair, or in our in the best interests of the artworld, to allow AI artists to compete for prizes or recognition with photographers. While there may be a place for AI art in the world, it may not be in the same competitions and shows as photographers. This is nothing new. Many art competitions are limited to a single medium and have strict submission requirements. For example, a watercolor competition may refuse to accept or consider watercolor paintings that are “impure” insofar as they contain colored pencil or gouache.
It also follows from this discussion that, even if AI generated images are art, the artists who created them may not be especially talented, skilled, and praiseworthy. Again, this is nothing new. We might embrace Fountain as a work of art without thinking that it took much talent or skill to produce. After all, Duchamp didn’t construct the urinal himself; he simply flipped it over and signed it. Similarly, a person might hastily splash some colorful paint on a canvas and call it art. They may be right that it is art even if it is not great art – even if it took virtually no talent to produce.
To sum things up, AI art leaves a significant amount of the art making process in the hands of the AI art generator. If an artist simply uses an AI art generator, and does nothing more to alter the image, then the AI is responsible for making almost all the magic happen, as I put it. For this reason, it would be wise to keep AI art separate from not AI art for at least certain purposes. In my next post, I will consider whether the existence of AI art poses an existential threat to artists and creative professionals.