Klára Sedlo:
ARTIST VS AI - my retrospective reflection on the exhibition
"AI will replace the artist". "Aren't you afraid that AI will take your job?" I have to say, these questions leave me pretty cold. I'm of the opinion that AI can't replace an artist for several reasons.
But to be honest, while we're at it, one question concerning AI does get under my skin. Not only that, but I also feel an inner need to separate myself and my art from it. Every now and then someone asks me whether I use AI as a template. "What?!" I hear myself say. "What? Someone really thinks that?" The answer is simple. No. I don't. And I never have. I have no reason to - my mind is capable of generating much richer and much more interesting scenes. The question of whether my imagination must be aided by a machine is sensitive.
However, the more I think about AI, it dawns on me how my head works in a similar way. Sometimes it frightens me. It works similarly - and yet quite differently. AI actually mimics the process by which (at least) my imagination works. Although, having reflected further, the process of generating ideas is similar, we do have different input, motivation (if you can talk about motivation with a machine) ... And as a result, as this exhibition shows, also different output.
So how is my mind similar to AI, and how are we different? How is artificial intelligence different from human imagination?
My creative process looks something like this: I take in the visual sensations around me, whether it’s a beautifully colored sky, a moving yellow car, a gap between houses, a pattern on a dress, a painting in a gallery... Next, I perceive experiences, feelings, stories. As a synesthete, I translate all of this into visual codes in my head, where color codes exist for words, numbers, and sounds.
Everything I just listed is my input.
Input is both natural and moderated: Inputs enter my mind unconsciously, but also consciously. Sometimes a scene is so intriguing, that I store it in my memory. I call it "taking a picture in my head", and it basically means that I focus on the visual for a few seconds, scan it, and store it in my memory. (Sometimes I'm able to trace it back when it pops up in an idea). My input is constant.
How the process subsequently evolves in my imagination, I don't know. I have no ambition to understand it, I probably even can't. I also see, in the process of creating ideas and visions (which happens in my brain all the time), a certain spiritual connection. Which is of course subjective, but for my creative process so essential that I cannot omit it. Within this work I speak a different language - visual, symbolic, sensational.Output: output comes from my subconscious/imagination as a vision. I really see the ideas for my artwork as visions on the surface behind my eyes, in my head. I see them most often as complete finished images - they have a given composition, the colors, the form, as well as the content, and the mood. My amazing imagination spits out several of these visions per day. A few or a dozen. I must write them down, of course. I've discovered that my memory is limited to about 5-7 visions/ day, then I start to forget the older ones as the new ones arrive. My memory is the only limit, but as long as I have a sketchbook handy, there is no need to grieve over a low memory capacity.
I have just described the process of how my mind works. Isn't that actually similar to how AI works? AI, which of course, can generate many more images per day. So, what's the difference?
This brings us to my first item on my notional list of ‘How AI differs from an artist’.
1. Input.
As I described above, the input of an artist comes from the whole world. Experiences, physical sensations, childhood, emotions, and of course, visual inputs, which include various pieces of other artists. Works of other artists are only a part, a smaller segment of the input that the artist uses. In contrast, AI primarily uses other artists' works as input - possibly also textual works (descriptions of experiences or emotions and so on). Of course, it has more sources than the artist, but it still represents a limited segment of other people's work. AI can never use real life, experience or emotion as input. By its very nature it is incapable of doing so - it can only use descriptions of it. (Yes, of course, until we implement AI in the heads of living humans, I hear you say. And you're right. In fact - what if we ourselves are the AI experiencing itself in various situations in a simulation?)
But back down to earth from simulation theories.
Although AI has a huge amount of artwork in its database, it still doesn't make up for a fundamental lack of input. Input that the human mind uses - the experience of the real world through the physical body.
2. The artist.
I believe that AI cannot replace the role of the artist in society. It can generate "work" (and some of the many will be of aesthetic quality - more on that in point 4), but it cannot replace the personality of the artist.
The artist plays a vital role in society. Her/his task is much broader than the mechanical creation of images. Very often he/she plays the role of a rebel, visionary, mystic, inspiring role model. Great artists have personalities that change the world, and we look up to them. If we buy/listen to/perceive their work, we always connect, more or less, with their personality.
When we look at Van Gogh's Sunflowers, we reflect on his life. His work inspires us with regards to his personality, his life and his time. We see his art in context. The artist's personality and work together create his/her legacy. A furrow in the very fabric of reality. By watching the work, we connect with the artist (of course, the work itself can also have a story - a small segment of the artist's life, concentrated in one canvas.) An artificial neural network has no character traits and, above all, cannot experience real life. It generates images from available databases on demand. It can mimic some of the artist's work, but it cannot fully substitute the role of the artist in society. To become an artist - a visionary, a rebel... and create a life's work within the human society requires living a human life.
3.Motivation.
People like to create - bad and good images. Even AI generates images - bad and good.
But the question is: Why?
Let's look at us first. Our need to create something out of something else is one of the factors that emerges with the beginning of humanity. To transform matter, to become a creator – that was our then-original motivation.
Art can be seen as a tool through which we better understand ourselves and the world around us. From cave paintings in which we visually described hunting, to ritual statues, Turner's slave ship (which helped change the perception of slavery), to Hilma af Klint's work connecting with the higher spiritual realms.
To understand oneself, life, and (perhaps) to reshape the world. This is a fundamental motivation of perhaps every artist to this day. For most artists, the need to create is so intrinsic that they compare it to breathing (which I like to say, but David Bowie, for example, and many others say as well). Creating becomes a life priority, a life need. To make something out of "nothing" - a sculpture from a stone, a drawing from blank paper, a story from an idea.
In contrast, AI lacks motivation. It is a machine. It has an assignment. AI is a tool. AI itself cannot have impulses to create as we do. A motivation that comes from insecurity, fear and love for life - that is, from experiencing the world. At the same time, we can ask whether AI creates something out of nothing, or whether it creates new variations of something out of something. But of course, AI can faithfully imitate art, perhaps even faithfully imitate human emotions embedded in art, as it learns to understand how we perceive emotion. And that brings us to a point that I see as dangerous, and that is the humanization of AI.
AI has very different impulses to create images than humans - I use the word impulse deliberately instead of motivation. And it's questionable whether we can use the word "creates" with AI. In my opinion, since AI comes from already existing works, plus it doesn't have the actual motivation to create, it's just fulfilling an assignment, and therefore I use the word "generates".
AI can decently imitate an artist at first glance. (By the way, do we want to settle for imitation?) And that is the stumbling block. That's why I think it's crucial to use the right words, and not to humanize AI. It's important to remember that AI is not human. We need to distinguish between the generated images from created art. I am not suggesting the above to disparage AI, but simply because in both cases it is something completely different, and we should learn to see the difference.
4. Number of alterations - finally a word about the exhibition
In his book Art and Kitsch, Tomáš Kulka formulated a kind of mathematical equation through which to evaluate, in simplified terms, the aesthetic value of art and to distinguish art from kitsch. The equation addresses the aesthetic value of a work of art, through a number of alternations to formulate the terms unity, complexity, aesthetic intensity and their mutual relationship.
According to Kulka, within the framework of intensity, the degree of alternations that can be done to a painting is essential (i.e. changes that do not disturb the basic essence of the work). If the aesthetic effect of the work changes (for better or worse) with only one, two or three alterations, the work is aesthetically intense. The more changes done to a piece without compromising its effect, the less intense it is. The author describes that, "A certain degree of intensity is therefore not only a condition of high-quality art, but of art as such." (1) He continues in that the amount of amelioration + de-melioration + neutral alterations show the complexity of the work.
In terms of unity "The degree of unity of an art piece will be greater, the greater the number of damaging alterations and the smaller the number of ameliorating alterations." (2) Meaning, that for good art, a small number of improving alterations and a large number of damaging ones can be done (the opposite is true for aesthetically poor works). He further contemplates that specifically within kitsch, many alterations can be made, but mostly neutral ones. Only little can improve or fundamentally worsen kitsch (we can change the color, composition, paint a kitten with a ball of wool instead of a dog with a ball, the effect is the same).
AI-generated images do not automatically capture kitsch, a dog, or a cat and a cute baby in a daisy field, unless they're instructed to. But I think that Tomáš Kulka's ideas can be, in some way, applied here. AI images, that I have had the opportunity to see so far, and that didn’t have a notably corrected assignment, usually allowed for many changes without changing the aesthetic impact (which is naturally evidenced by the number of alternatives they generate per assignment). At the same time, they allow for many neutral alterations, which again can be inadvertently demonstrated by the number of image variants that AI generates.
T. Kulka finally suggests, that kitsch "is not a continuation of mediocre or below-average art" and that "It must be understood as a completely different phenomenon." (3) Doesn't the same apply to AI? "Although kitsch has all the formal characteristics of an artistic image, in terms of its function it is more akin to a pictogram."(4) Now, how about we boldly borrow and slightly modify this sentence? Let's try: Although AI images have all the formal qualities of an artistic image, in terms of their function they are more like a pictogram. I'm not trying to imply here that everything AI creates is kitsch, not at all. But that AI in the business of "making art" belongs in a separate category, as well as kitsch.
And now I am (finally!) getting to the exhibition itself. During our experiment, when AI generated my images "reversed" (according to the text-based assignment also created by AI), I was able to compare AI's "creation", i.e. generation, and my creation. Although AI has a database of many works, it was often not able to capture the subtle details that are essential to the meaning of the work. Thus, it has created alterations that fundamentally changed the essence of the work. I attribute this to the only textually mediated experience and the absence of image perception. Take two works as examples - Sounds and Abandoned Luncheonette.
In the case of Abandoned Luncheonette, AI has completely omitted the central motive of the painting - the hand of the human figure merging with the body of the luminous being. The original image depicts not only an encounter with something out of this world or with one's soul, but above all, a fusion with this luminous entity. AI has managed to capture this solely as an encounter with a „spiritual figure “. In the original painting, the fusion is depicted very physically - the fingers turning into something luminous. The image of the melting fingers can also evoke fear or revulsion since a part of the body is drawn into a kind of energy - the figure loses its fingers. The merging of the hand is essential, but it is also depicted so subtly that it was overlooked by AI. Without this detail, however, the essence of the work changes completely. AI reverses generated images into a luminous figure and a human figure, however the essential moment of the physical, intimate, even terrifying fusion is missing. To understand this moment, the experience of inhabiting the human body is vital.
We can find small hidden pop culture references in the piece, which are not only overlooked by AI, but I must reveal them myself to the audience while telling the story of the piece. The title itself and the caravan in the back of the landscape is a reference to Daryl Hall and John Oates' album Abandoned Luncheonette. The album cover features just such a trailer. The combination of a random album, an encounter with God and an abandoned trailer is a bizarre combination that comes from lived input. From a situation I've lived through.
The next image is Sounds. In this case, I was particularly struck by the intensity with which AI focuses on expressive representation - the sounds, the screams, the terrifying snakes. All much more garish and expressive than the original image. Sounds come from my painting series Bizarre Vacation, in which I captured a fictional holiday resort with something out of place. Similarly, in the painting Sounds, everything is quiet, straightforward, elegant, static - the attitude of the ghost, the sky, the implied arch and the hands. The only thing that disturbs the static composition are the levitating dinosaur heads, which are relatively small in relation to the whole picture. An inappropriate sound, perhaps heard only in your head, disturbing the calm atmosphere and the overall picture. This contrast was completely omitted by AI. It concentrated only on the expressive representation of the dinosaurs, which it conceived as snakes, and to which it devoted the large image area and the terror of the figure.
The concept of the dinosaur heads as snakes is also quite interesting. AI was probably trying to match the element to something from its database, so it chose snakes. But the truth is that the dinosaur heads are based on a snapping plastic dinosaur head-shaped toy that I bought randomly years ago in a shop in Prague, just because it caught my eye. It has appeared in many of my paintings since then. I smile and think to myself that the absurdity of the subject far exceeds the current "thinking" capabilities of AI.
On the other hand, for the Balance image, AI generated the image quite accurately (essentially creating an AI copy), and an interesting result was produced by AI variations on "What's Happening," where the network recognized the brushstroke input as a realistic painting. Still, the result is more poster-pop art and, again, doesn't quite accurately reflect the essence of the work - i.e. the entry of a child-like brushstroke into a realistic painting.
Of course, AI can visually simulate the craft side of art very well - i.e. generate something that will look like a realistic painting, photo, illustration or poster according to the brief. It can also generate an image, that looks like another painting or mimics the handwriting of the artist. It can generate images that mimic real art (which in the age of social media, where attention span is reduced to tenths of a second, is enough). But it can't (at least not yet) naturally capture the more subtle emotions at work, and it doesn't work with the image as a whole. In my experience with AI, it mostly sort of spasmodically focuses on the display that will be as "impressive" as possible, at the expense of everything else - i.e., it tries to target simplicity and emotion. Careful here, the difference between portraying an emotion, and trying to affect an emotion are two very different things. Both have an effect, but the representation of emotion comes from the inner need of the artist, whereas trying to affect emotion is flat and calculating. (Logically, AI is not capable of the former.)
AI can of course generate random complex and high-quality works, but it does not generate them on purpose (if unless it's working with a very specific, corrected assignment). Unwittingly - and perhaps a little mischievously - a popular theory about the chimpanzee and the typewriter comes to mind. However, the shortcomings described above can of course, I suppose, be corrected by a specific and corrected assignment, or by drawing on a particular segment of the database - but at that point AI serves as a tool, which is corrected by the human hand (to which I will return in the conclusion). At the same time, the question is whether and why to make AI capable of creating complex works of art. See points 2, 3 and 5.
5. Ethical question
I personally encounter AI most often in the artistic field - I see the use of AI-generated images within graphics, illustrations (even in books), etc. Of course, this raises the ethical question: Is it ethical to cut artists out of illustration and give their jobs to a machine that will do them cheaper while draw on the work of living artists from its database (often without the consent of those artists)? Is it ethical to abolish positions of existing illustrators and "give the work to the machines"? Do we want to use the cheaper alternative at all costs? And what is more important - life or money?
I don't want to go too deep; I think this topic has been described by others, and not just today. We have asked similar questions during technological progress in various permutations. And that is why I would like to conclude my list with just one sentence, a quote. It's - you'll probably laugh - from the musical Hello Dolly. It goes something like this: 'Money is like manure. It's not worth anything by itself, but when it's spread over a field, it helps little plants grow."
Therefore: let's not overestimate the savings of money. Human existence itself is the most precious thing we can encounter that far exceeds any other capital. If we have money, let us exchange it for the opportunity to develop other people's lives - we'll save ourselves a nice pile of dung at home by spending a few thousand on a budding illustrator, but nothing will grow in the field.
So, what is the conclusion? Is AI useless to artists, an enemy we must destroy? Not at all. AI can serve as a great tool. If we task the artificial network with purposeful correction, modification and use it to turn our particular vision into reality, it can be beneficial. AI can be a tool of creation, much like a camera, or when foundrymen cast our sculpture in bronze, and bring our model to life. The implementation of AI in certain areas is inevitable, and ultimately it can be helpful. But at the same time, we should distinguish AI-generated images from human-created works (which is a problem on social media, which is another debate). And above all, we should not aspire to replace the person of the artist by artificial intelligence, and artworks by generated images. Let's look at the rich inputs and processes used by artists. Their input is literally their whole life, lived in a physical body. To try to replace this unique and rich experience with an artificial neural network is, in my opinion, an unnecessary step backwards. And for another, creativity is part of our human nature, our way of grasping the world around us, ourselves, the way we evolve. We should preserve the ability to create. AI can and should expand those possibilities, but not replace them.
Klára Sedlo
Made AI so we can do the fun stuff and it could do the dumb
stuff. But now it does the fun stuff. And we became the dumb
stuff. – Ryan D. Anderson (5)
1 Tomáš Kulka: Art and Kitsch, Torst, Prague 2000, p. 99 (loosely translated)
2 Tomáš Kulka: Art and Kitsch, Torst, Prague 2000, p.93 (loosely translated)
3 Tomáš Kulka: Art and Kitsch, Torst, Prague 2000, p.103 (loosely translated)
4 Tomáš Kulka: Art and Kitsch, Torst, Prague 2000, p.105 (loosely translated)
5 Ryan D. Anderson is an American comedian, musician, animator, the quote comes from his short animation "We Wanna Do the Fun Stuff", 2023
Note: In this reflection, I consider art as a complex discipline. That is to say, the artistic process of creation, which may include the so-called craft (the ability to paint, model, etc.). But as we understand the word art in the text, craft is only one component of it.
Note II: I have outlined the theory of art and kitsch very roughly and simplistically. But I think that a more detailed examination might yield more interesting ideas and I certainly recommend everyone to read Mr. Kulka‘s book.
