Recently, I heard someone say on CBC radio that artificial intelligence (AI) would help us “make better art”. The person was not connected to the arts and only made this statement in passing, while listing the potential benefits of AI. It struck me that only someone who fundamentally misunderstands what art is could make such a statement.
In the following, I use the word “art” to refer to drawing and painting. Similar arguments are likely to hold true for sculpture, photography, literature, music, dance, and other forms of art, but I only wish to write about what I know well.
When shown a selection of images, most people are unable to correctly identify which ones were produced by AI. This is considered a success for AI and proof that AI is really producing art. One problem with this argument is the assumption that quality does not matter in art. In other words, the human-produced work that AI is put in competition with does not have to be good. If you compare something produced by AI to something produced by a human who works mechanically, or who copies photographs, then it won’t mean very much that a viewer is unable to tell whether a human or a machine produced the image. After all, computers are better at repetitive tasks than humans.
What artists aim to create, and what excites art lovers, is not mediocre art. It is great art: art that speaks to the viewer and gets better the more it is looked at. Therefore, the real question is not whether AI can produce something that is indistinguishable from mediocre art but whether it can produce great art. That is what I address in this article.
Great art is about truth and honesty. It is an exploration of the inner and outer worlds of the artist. An artist making a drawing or a painting has something in mind that they want to capture or explore. The “something” is not as simple and straightforward as an idea, and it is not as precise as an image; it is something nebulous that the artist feels compelled to turn into something physical. The physical object almost never turns out exactly the way the artist intended. What makes the object great art is not how flawless it is or how impressive it is or how close the artist got to their original vision, but the truth and honesty with which the artist worked. The process, therefore, matters as much as the product. Or, to put it differently, the product will not be good if the process is flawed. If, for example, rather than being absorbed in the work and following the inner logic of the work, the artist aims to impress the viewer (something even talented and famous artists have done, producing works that hang in museums and impress viewers but don’t move them), the outcome will lack the magic of a work that was made with truth and honesty.
This point cannot be emphasized enough. Great art is born through truth and honesty. Lies cannot produce great art.
What is the process when we talk about an image generated with AI? It is certainly not the same process that leads to great art. The person using the software is not aiming to tell the truth or be honest. They are trying to produce something that impresses, that “passes for” art, that looks like something it isn’t: a work of art produced by a human.
Is the person who uses AI to produce images telling the truth and being honest? If they are, then why are they using AI, which feeds off the work of other artists? Isn’t that like stealing someone else’s product and selling it as one’s own? How is that honest?
The person who uses AI to produce images will argue that artists have traditionally copied other artists’ work. This is a fallacy. Great artists of the past did not copy parts of works by other artists into a single work and call it original art. They copied work to study it or to develop their skills, but they did not then try to pass off the result as original art. Or they saw another artist’s work and felt inspired to address the same subject in their own voice, either because they felt the original artist did not do the subject justice or because they felt the subject was important enough to deserve multiple treatments. That is not what AI is doing.
The person who uses AI may also argue that great artists of the past used assistants and that when they use AI, they are using the work of other artists as if those artists were their assistants. Anyone who claims that great art has been produced by assistants and apprentices cannot have spent much time in art museums. If they had, they would know that work produced by the artist is always considered more valuable than work produced by the artist’s studio. A preparatory sketch produced completely by the artist is more interesting than, and visibly superior to, the final painting produced by the artist and his assistants. In fact, the use of assistants was necessary and widespread only prior to the advent of ready-made supports and tubed paints. It was not based on the argument advanced by art dealers today that only the idea matters and that the painting itself can just as equally be executed by the artist’s assitants. An artist who aspires to create great art believes deeply in their talent, ability, and vision. They are not satisfied with the work of other artists, even if they may not admit this publicly. When great artists used assistants, they had them work on less important parts of paintings, or, in some cases, did not have them work on their paintings at all. And just because they used assistants doesn’t mean that using assistants is the best way to produce works of art. Is a less-than-honourable action justified because some famous figure from history sometimes engaged in it?
Others claim that AI can help artists by providing an image for them to start with, so that they won’t have to start with a blank canvas. Anyone making this argument doesn’t understand the artistic process. If you are an artist who has mastered your medium, and you are facing a blank canvas and can’t seem to put anything on it, there are two possibilities. The first possibility is that you are suffering from anxiety. The solution is not to start with an AI-produced image but to deal with your anxiety or figure out a way to work through it. If you don’t do that, you will produce work that is timid and overly cautious. The second possibility is that you are not ready to make that painting, and that your brain needs to process some information or feeling or resolve a problem before you are ready. In that case, the only solution is to give your brain the time it needs. Any attempt to circumvent this process will produce mediocre art.
As much as I am loath to pollute my brain by looking at AI-generated images, in preparation for this article I looked at some that are supposed to show how amazing AI is. They simply don’t hold my attention. My eyes glance at the image and move elsewhere; they don’t linger within the image. They don’t go around and around within the image with the pleasure of a child skipping about a playground the way they do when I look at a great work of art. (This is similar to what happens when I try to read prose written by AI. My mind wanders, and I have trouble concentrating.) When I try to force my eyes to stay within the image, I have feelings of discomfort, as I encounter elements that seem subtly wrong, that don’t fit, that don’t feel right.
One problem with AI that I haven’t seen anyone mention1 is what will happen if AI puts human artists out of work. To produce useful output, AI needs human-produced input. Without continued input from humans, the output of AI will be frozen in time. Art is an artist’s contemplation of their world, and it evolves as the world changes. Imagine what the world would be like if all we had was art created prior to the 16th century. While there are some great works of art that have withstood the passage of time and remain relevant today, most of the art that speaks to us today was produced in the 19th century and later. And this is not just a problem for fine art. Imagine what ads would look like today if AI had been invented in 1989 and had put all commercial artists out of work. Not only would all ads have 1980s hair styles, but they would also carry the zeitgeist of the 1980s. This argument is easier to understand in the case of literature: for most people, contemporary literature holds far more interest than medieval literature. Some great artists, like Shakespeare, created works we still appreciate today, but most did not, and even Shakespeare’s works are disturbingly sexist, racist, and anti-Semitic in places. If humans are foolish enough to allow AI to put artists out of business, our artistic environment will deteriorate over time. Countries that don’t support the arts and sciences inevitably lag behind those that do, both economically and culturally. If the world loses its artists as a result of greed and stupidity, it will lose everything.
Underlying the question of whether AI is making art is the unspoken premise that a drawing or painting is nothing more than an image. Indeed, everyone seems to believe that flipping through low-quality digital images on a smart-phone screen is a valid way to view art. No one seems to think that a work of art needs to be looked at for more than a second or two. People who have never bothered to see a work of art in person, who do not go to art shows and who do not have any original art on their walls nonetheless consider themselves art lovers for merely following a few artists on social media. The world seems to have accepted that the image is all that matters. Who needs artists in such a world? If humanity is that shallow, then AI-generated images may indeed be what it deserves.
If people really cared about art, if they went to museums not to take photographs to post on social media but to actually look at the art, if they spent more than a few minutes in front of each work, they would have a clearer idea of what art is, and the question of whether AI can create art would not be seen as deserving discussion. Looking at a painting, say by Rembrandt, is like meeting Rembrandt, having tea with him and hearing his innermost thoughts. You are not going to look at an image produced by an app and have the feeling that you are having tea with the person who fed prompts to the app.
I have not worked on machine learning or AI, but I programmed professionally for many years, first as a physicist and later in private industry. People who are not programmers place too much trust in computers. In fact, a computer is not much more than a fancy calculator. Although a calculator can calculate square roots faster and more accurately than a human, it will produce the wrong result if the human using it types the numbers incorrectly. AI, in spite of the mystical way its proponents talk about it, is ultimately a computer program that executes instructions given by a human. AI is never going to make science. There are too many combinations of any meaningful data sample for a computer program to know what to look for and to come up with anything original, correct, and believable on its own. Similarly, creating a painting involves a huge number of decisions. The composition, the colours, the brush strokes: there are countless choices that an artist makes, sometimes without thinking consciously. Just as a computer is never going to accidentally write Shakespeare’s Hamlet, a computer is never going to get all the choices made by a visual artist right.
AI is not conscious. One could imagine that one day, humanity could produce a machine that is truly conscious, like Isaac Asimov’s R. Daneel Olivaw, a robot with its own thoughts, opinions, and feelings. Such a machine could conceivably make its own equivalent of art. But that would still not be human art. It would be robot art, addressing matters of importance to robots, and intended to be viewed and contemplated not by humans but by robots.
We think that computers are free of our flaws and weaknesses. But how can something we have created be better than we are? We are the result of millions of years of evolution. Even if we build a computer that is conscious and can think for itself, as opposed to mindlessly executing a set of instructions, who is to say that it’s not going to suffer from its own version of denial, depression, anxiety, self-doubt, self-hate, and self-sabotage? Who is to say that such a computer will not lie in order to further its own purposes? Why do we have so much more faith in a software program developed by a for-profit company than in our fellow humans, who share this earth with us, trying to be good, working hard, loving, feeling, suffering, facing their mortality, and looking for meaning in life? Isn’t art a contemplation of that human condition? How can a machine, conscious or unconscious, perform that feat as well as or better than a human?
If you are an artist, and this article makes you angry, please think about what you are aiming to achieve with your art. What makes you so certain that one can make great art without first learning to draw and paint? Is that not like expecting to make great scientific discoveries without first putting in the effort to learn the techniques and methods? If art is worth doing, can it really be much easier than science? You only have one life to live. Do you want to spend that life trying to fool yourself? As the world cheerfully drives off a cliff into an abyss of superficiality, do you want to jump along with it? Or do you want to do everything you can to maximise your chances of creating great art? What do you want to leave behind when you die? Something true and meaningful, or a lie?
- What has been mentioned is Model Autophagy Disorder (MAD), in which the use of AI output as input to train AI results in distorted output. That problem has a solution, at least in principle: one can tag or identify AI output in order to avoid accidentally using it as input for AI. The problem I am referring to is a different one and has no solution. ↩︎