Is AI Just a Tool For Making Art?

Is AI just a tool when it comes to making art? Where does the real art happen? Is the idea the most important element, or the execution? What kinds of ideas are present in visual art? What does everyone miss when talking about ideas in AI art?

There is a common argument in debates about AI art that goes like this: humans provide the idea, and AI is just a tool that does the busy work of executing the idea in visual form. The real art is the idea. We find a similar belief in discourse around conceptual art, which maintains that the idea is paramount in art. This same sentiment appears in the familiar notion that the purpose of art is to “start conversations”, or that the function of art is to propose an idea. All of these notions have their genesis in the anti-art movement of the early 20th century, and especially in the pronouncements of Marcel Duchamp. He positioned the artist philosopher as superseding artist painters and making them redundant. The argument that the execution of the image is mere grunge work allocated to a worker-bee ultimately brands both AI and painters as mere tools. The making of art, and the resultant image are both considered as mere vehicles for communicating a thought in linguistics [spoken and written language]. Here we see the privileging of linguistic language as innately superior to visual language, and within the realm of visual art.

Something fundamental to a genuine understanding of visual art is tragically absent from such a viewpoint: there is no recognition that visual art is rife with its own language and ideas that are entirely different from those expressed in words and sentences. We used to recognize that imagery could communicate independently of linguistics when we said “a picture is worth a thousand words”. Today we say that a prompt is worth a thousand pictures.

AI art clearly illustrates the contest between linguistic ideas and visual language ideas. Humans forumulate instructions in words and sentences in the form of a prompt, and then AI creates a corresponding image using visual language. The breezy assumption is that there are no competing ideas embedded in the imagery itself. It also ignores that artists come up with their own ideas in linguistics.

People who are pro-AI art will frequently argue that AI has allowed anyone with creativity and insight to become an artist. One no longer has to subject oneself to the endless drudgery of learning to render imagery, whether it’s through drawing, painting, sculpting or digital mediums. There is some truth in that, and AI could allow people with disabilities to explore making visual art that otherwise prevented them from doing so. But let’s not make some truth into an absolute truth and ignore 90% of the equation. A large part of the appeal of art is that someone was able to do something that we can’t do ourselves. Nobody is impressed if a kid using a laptop wins a spelling B at the local elementary school.

The accomplishment of the feat

Consider the digital image above, conceptualized by someone who calls themself The Unclean, and rendered by AI. [Note: I’m not criticizing the artist, and I reposted this creation on my X account because it so impressed me. I maintain that good prompting takes practice and people with backgrounds in art generally get better results, or at least that used to be the case.] Here, from the “AI is just a tool” perspective, we are to understand that the idea of an angel slumped over in the netherworld is the brilliant element.

On the other hand, I’ve seen a lot of depictions of angels, and this piece is reminiscent of the cover art of Sad Wings of Destiny by Judas Priest [incidentally their best album], painted by Patrick Woodroffe.

The concept of an angel who has fallen into hell, or somewhere similarly gruelling, is not entirely new. Both images share a demonic sort of symbol, and in the case of the Priest cover, it’s the devil’s three-pronged cross.

Making such images can be extraordinarily difficult, and pulling them off is a large part of the art, just like landing a triple axel is in figure skating. And while Boston Dynamics’ robots can already do choreographed dancing and backflips, it would raise eyebrows if the programmers credited themselves with executing the physical feats. We humans don’t always succeed at the ten-point landings, or even a three-point landing, if we can ice-skate at all.

Despite Patrick Woodroffe being one of the greats of contemporary illustration, when I listened to Sad Wings of Destiny in high school, with the album cover propped in my lap, I couldn’t help but notice problems with the angel’s anatomy. It bothered me that the figure appeared to be sitting on the ground, but his legs bent at the knee as if he were sitting on a chair. There was no place for his lower legs to go. The structure of his right knee looked too bulbous, and the folds of flesh connecting his legs to his waist were too conspicuous and peculiar. Realistically rendering a human body in perspective is anything but easy, especially because anyone looking at it carefully can spot if anything is wrong. The challenging part is fixing a problem or getting it right in the first place. The task of rendering both the Judas Priest cover and the somewhat similar fallen angel image is likely astronomically more arduous than coming up with the concept. It’s not entirely unlike the thought of doing a triple axel requires thousands of hours of hard training less than actually doing it.

To compare human versus AI achievement in visual art, let’s have a closer look at Patrick Woodroffe. Woodroffe is a consummate artist and illustrator, and his cover for Dave Greenslade’s The Pentateuch of Cosmogony knocked my socks off when I was a teen, and it still does today.

I remember being drawn to the ripples in the sand, and the technique of rendering the droplets of blood (or whatever that fluid is) within them. It was the creature’s skeleton itself, however, with its antlers, talons, and mouth frozen in a death howl that I was particularly struck by. 

As it happens, the double album is a book filled with his illustrations.

One of his paintings that most impressed me was of a group of whales floating in the sea, their flesh eaten away and their organs spilling into the water. The culprit was undoubtedly the row of nuclear cooling towers billowing smoke in the distant background. Here, the artist most likely came up with his own concept and certainly every aspect of the execution. In the end, it’s the details of the whale in the foreground—just the way that it looks—that most captivated me. The pinks, yellows, oranges, and reds of the whale’s innards were as colorful as a Monet haystack at sunset.

The intricacy of the permutations of exposed flesh making up what remained of the head, and the bones protruding through the gaping holes in the fins, affected me as pure visual experience. While I can find words to refer to what I’m talking about, in no way do they conjure that same brilliantly crystalline image. Notice the calm interaction of the intersecting concentric ripples in the water between the heads of the adult and juvenile whale in the foreground. That is not something my imagination would deliver based on reading a description. Those kinds of details develop in time while the artist is in the process of making the image, as one thing suggests another. Artists who make such complex images using the imagination do not know what the final image will look like, and then painstakingly reproduce it in physical mediums. The image comes to fruition in and through the process of creation.

If you don’t mind terribly, and even if you do since you can’t protest in advance, some of my own art is ideal for illustrating how the image develops through the process. Often I will start a work with no preconceived idea. I’ll make marks on the canvas [physical or digital], and then look at it for ideas, similar to taking a Rorschach test. Imagery will slowly emerge, and only when the drawing or painting process is finished is the image realized. All the ideas manifest themselves in the process. I’ll give just one example.

At this stage of a drawing the imagery has started to coalesce and I decided to run with it:

Here you can see me working on the drawing in Photoshop:

And here’s the completed drawing:

It is entirely from my imagination with no initial “idea”, and is completely done by hand using my drawing skills. This whole process and the result is not accounted for when one assumes that art is the execution of an idea in linguistics.

And now back to Patrick Woodroffe.

Damned if Woodroffe didn’t also paint the cover for Bandolier by Aussie hard-rockers, Budgie.

There are myriad visual ideas here that a novice artist wouldn’t even think to include. Why are the legs of the horses furthest away blue and solid shapes of color? Part of it has to do with there being more atmosphere, which is blue, between us and further objects. You may be familiar with seeing this in photos of mountain ranges. But here, the artist chose to make the blue bold in places where it buts up against the orange-brown of the horse’s hide.

It would not occur to the amateur to incorporate, let alone amplify such an effect. When the artist or AI supplies these expert ideas in visual art, they are likely visualizations—or ideas in visual form—that would never occur to someone commissioning or prompting the art. Nevertheless, just as a judicious sprinkling of spices from a master chef can kick a recipe to the next level, so the intricate components of image-making can be among the most complex and interesting ideas in a painting.

Woodroffe knocked out three of my favorite album covers from my own music collection, and I bought all those albums because of the cover art. The reason Patrick Woodroffe was called upon to create so much cover art was that there weren’t that many artists who ever reached his level of technical virtuosity. Few have the patience or love of the medium necessary to get THAT good.

Until now!

With the help of AI, there are tens of thousands of overnight virtuoso artists and illustrators flooding social media with whole portfolios, even ouvres, of their fledgling but masterful creations.

Worse for artists who make their own art by hand using their own hard-earned skill: no artist is safe from their style being appropriated by AI. On top of that, AI artists whose creations are trained on the styles of established artists will claim the results as their own art. The following pieces were made by Midjourney’s AI after it was trained on the work of Patrick Woodroffe.

When I first saw the image above, I thought it was a painting by Woodroffe himself that I hadn’t seen before, and an outstanding one at that. 

Here’s another selection:

Today’s keyboard virtuosos may even turn their noses up at the same artists whose work they feed into the AI. They may feel superior because they weren’t fool enough to have squandered years of their lives alone in the studio, mastering techniques that would soon become obsolete. It’s a bit like being able to become Mr. Universe without setting foot in the gym. Not only did they not have to pump unlimited iron like Sisyphus endlessly rolling the stone up the hill, they got better results in minutes without ever doing a single squat. Some may even perceive the artists who take umbrage at losing their careers to AI as an army of reactionary, redundant, and ultimately talentless drones who can’t keep up with the times and didn’t make the evolutionary cut. All of this is enough to make digital artists hang up their styluses and tablets in a lethal bath of disgust and despair. I will argue why there may still be hope in another article. Stay tuned.

Recent examples of what AI art bots are capable of

Initially, I shared the piece by The Unclean because it reminded me of Woodroffe’s cover for Judas Priest, but I could have used any of tens or hundreds of thousands of other images. Here’s a bunch of remarkable AI generated images I just nabbed from a search on X.

Just between you and me and everyone who reads this, OH MY GOD, AI IS ABSOLUTELY PHENOMENAL! People don’t like when I say this, especially if they want to take full credit for what AI generates for them, but AI is a super genius virtuoso artist [er, in which case people sharing AI art as their own productions are sometimes naively posing as uber genius artists themselves]. AI being an artist in its own right is a controversial statement for another article. 

Those last 3 amazing images in particular help illustrate the point of idea vs. execution because they are all based on the same idea.

I don’t know what the full prompts were, but the “idea” which the artist shared in each post is “like taking candy away from a baby”. Don’t get me wrong, there may be much more behind-the-scenes work involved in creating these images than just the idea. This AI artist could have spent weeks or months obsessively tinkering, practicing, integrating, and manipulating images with AI before being able to realize anything of this caliber. Further, these few stunning works may be the cream of the crop, selected from hundreds or thousands of failed attempts.

On the other hand, AI is now training on its own creations in a feedback loop of exponential flowering of unbridled genius and has the power to create such images only based on a one-sentence prompt. It’s easy to get magnificent results from AI, but harder to get it to do explicitly what you want. 

An image is itself a vision made up of myriad ideas in visual language

Making art on the level we see AI cranking out is an extraordinarily difficult feat we can’t take credit for. But that is not my main point. The critical point is that the rendering of an image is itself an idea, or rather tens, hundreds, or thousands of them that come together to form a vision. They are ideas in visual language. Don’t think this is just flowery stuff we nod along with but has no hard basis in reality. Stop and consider how many decisions Patrick Woodroffe made in creating the Planet of the Apes-inspired Bandolier cover.

Even the most passionate believer in “The idea is the art” would admit you could easily generate hundreds of images using the identical prompt. We’ve reached the point where the almost flippant idea “Planet of the Apes, but with budgie heads” is more brilliant than the extraordinary effort and mental gymnastics necessary to realize the painting. Worse, such efforts are considered on par with painting a picket fence, while the real artist is like Huckleberry Finn tricking gullible drones into doing the drudge work for him.

You could give the Planet of the Budgies idea to a hundred artists, and each would come up with something different, most of which would be fairly derivative and uninteresting. The vaunted “idea” is frequently an uninspired thought within linguistics. Truth be told, AI can make a masterpiece out of an insipid idea. The person who is responsible for the concept, especially if it is not the artist himself, may have no image in his head to go along with the idea. Conjuring an image in your head IS an idea. All ideas do not arise from, are not made up of, and are not transferable to words and sentences. The ability to visualize is another kind of thinking that is outside and beyond linguistics, just like composing instrumental music requires all kinds of thought processes that have nothing to do with arranging words in sentences. This is one reason that some of our best artists and writers deliver up milquetoast on a platter when they try to express themselves in another genre.

Music and visual art are modes of communication outside of linguistic thought which allow artists to manifest their internal vision. In fact, these modes provide an essential role as checks and balances against the potential tyranny of verbal and written language. While linguistic thought unfolds in a linear fashion, in an orderly succession of words, like a train track, the whole of a painting can be seen in an instant. But do not think, incidentally, that the whole thing can be absorbed in the time it takes to flip to the next image on your smartphone. It’s worth exploring why this is the case in order to better appreciate how visual language differs from spoken and written language. 

Vincent Van Gogh is one of my favorite artists, and I never look at one of his best paintings and think, “I understand this completely”. In order to do so, I’d have to match his depth, breadth, and scope. My understanding and appreciation of reality would have to encompass his, and that’s never going to happen. I think I get his art to a large degree, but not having made it myself, one brush stroke at a time, out in the wheat field, I can never quite fathom his own art to the same degree he himself did. We may think that we have looked at an image enough to judge it merely because we saw the whole thing, but we haven’t even started to really absorb it until we’ve gazed at it without distraction long enough for it to make a distinct imprint in our inner mind.

The speed at which we can scan a visual image is similar to how you can read a poem much faster than a novel, but to really get the meaning can take a lifetime. Meanings in art aren’t just the correct answers that we memorize to fill in a blank on a test. There are no correct answers, no answers, and no words. Meanings can exist outside of linguistic language. We have the word ineffable just for describing this kind of experience:

You could even say that a goal of art is to conjure or capture the ineffable. More often, the word “transcendent” is employed. To conjure such a sensation, that is also the whole being greater than the sum of its parts, requires a sort of alchemy where the artist or musician transfers a kind of awe into a work of art in order to share it with others. In this way, an artist is also an explorer seeking new impressions to introduce into the collective consciousness. To make all of that subordinate to a declarative paragraph or a set of instructions, merely because those happen in linguistic thought and satisfy a reductionist and exclusionary definition of “idea,” can miss or worse eradicate one of the greatest capacities of art: transmitting knowledge beyond words.

In the case of AI, the cognitive power required to make the image, at least if it were performed by a human, could easily dwarf the neuronal firings necessary to produce whatever idea someone harbored while typing the prompt. This is the most likely scenario because the AI produces images that would take human artists days, weeks, months, or even years to produce. All sides of the AI debate should be willing to concede that the amount of sheer mental energy it would take a human to produce the equivalent of a high-quality AI image by hand is vastly more than is expended in typing the prompt.

It is a very curious phenomenon that in much of contemporary art, especially conceptual art, and in current art theory and criticism, visual language is not only sidelined but can be completely omitted. In today’s art discourse, the word has triumphed over pictures. AI artists who glibly state that the idea is the real art are merely echoing an idea which originated in conceptual art going back to Duchamp.

To further illustrate that the rendering of an image is the core of visual art let’s compare two paintings of the same subject by two different artists. In 1869, Monet and Renoir painted the same subject in the same context and with broadly the same underlying paradigm. Both were painting on easels outdoors, and capturing the light and colors they saw in quick impasto brushstrokes. Both shared the underlying conviction of depicting nature and the leisure class, and as opposed to the grandiose history paintings painstakingly produced in the studio that were the hallmark of the academy and the Salon. The two artists were working practically side-by-side, looking at the same subject, and competing to capture it. The subject of the afternoon was a popular restaurant on the bank of the River Seine called La Grenouille [The Frog Pond]. The results are markedly different on close examination.

Monet’s painting seems thicker and more assertive, with broad horizontal strokes evoking the reflections in the water, which he better captured, particularly in the lower right of the canvas. Renoir’s production is more fussy, soft, sketchy, delicate, and wispy. Monet depicted a wider range of light with the trees in the distance brightly illuminated by the sun. Renoir’s canvas shares an all-over, shady, ambient light. Renoir’s people are better realized; they have more features, and their clothes are easily discernible. In Monet’s painting the people are comparatively blocked in as almost abstract patches of shape and color.

Neither artist could have painted the other’s version. Each painting subtly and indelibly reflects the underlying psychology, personality, training, experience, and history of its creator. To tease out the differences for yourself, you can imagine which painting you’d rather have on your wall and which you’d rather visit if you could walk into the art. While in linguistics the idea behind both paintings is nearly identical, the real ideas are wrought in the painting and the painting process itself.

Monet would never have presumed to take credit for Renoir’s painting, and vice versa. Nevertheless, people today will share images rendered by AI on social media as, in their own words, “my art”. I like to compare this to using a chess computer to make your moves while playing a human online and believing that because you were savvy enough to employ such a tactic, you actually understood the strategy of the chess computer’s moves, or have evolved beyond the need to even bother. In the same way that someone using a chess computer to make their moves for them may not have any idea why the computer made the moves, or someone who uses Google Translate to translate English into Chinese can’t read the Chinese, someone who uses AI to make an art image may not be able to fathom what their own ostensible art conveys. If third-grader were to type “my parents on their honeymoon” into a prompt, for example, she might end up with a highly inappropriate bedroom scene [even if not explicit] far beyond her years of comprehension. Undoubtedly elementary schoolchildren can use AI to produce art a highly trained adult artist could not.

Once the AI has rendered the image, the would-be artist may only know if they like the result or not, and may be incapable of intelligibly analyzing or discussing it’s merits.

You cannot say that the artist provides the idea and the AI merely executes it in visual form, because the execution is in another language altogether and involves potentially vastly more, much greater, and even more profound ideas than the concept in words and sentences.

To go back to the image Midjourney’s AI rendered, following the instructions of the AI artist, what impressed me when I first saw it was not the idea but almost entirely the execution.

Despite a lot of irregularities a trained eye will see on close inspection, such as that the angel’s head is far too elongated from front to back, or that the wings hover above the back rather than integrate into the musculature, the details and overall aesthetics are exceptional. An artist would need to have spent nearly as much time and effort as Patrick Woodroffe did to create the Judas Priest album cover to paint or digitally paint this image.

What’s missing in the simple stance that the artist provides the idea is any understanding at all that there are visual ideas in visual language in visual art. If this is still at all opaque, I always find musical analogies much more accessible than talking about visual art. What is the idea behind a Beethoven piano sonata? My favorite is possibly his 23rd, the Appassionata. It’s a tour de force mind-blower for me. There’s not a single lyric, and there’s no accompanying writing, other than the title, to tell us what the musical composition is ostensibly about. Would you say that the sonata has no ideas in it and is only menial craft? If you think that, then you have not yet had the pleasure of thoroughly imbibing this musical masterpiece. The notes, arpeggios, cadences, chords, octaves, key, tone, timbre, melody, and how they intersect are all ideas, and there is no idea in linguistic thought that is a part of it or that can touch it.

It is the same with the paintings of La Grenouille by Renoir or Monet. They use color, composition, movement, texture, modeling, lighting, shading, and so on, not nouns, verbs, adjectives, subjects and predicates. You cannot give an instruction in linguistic thought and take credit for a cornucopia of thoughts conceived in response, and in another language that you are not fluent in.

Most people, I think, would get past this phase of thinking they were a sleeping Picasso, but some will boldly insist that prompting is like being an architect. The architect doesn’t physically build the buildings, they tell me. This, again, arrogantly relegates the production of art to brute labor, and fails to even recognize that visual language exists. Most humorously, they miss that, in their own analogy, they aren’t really the architects because they don’t design the building. They are the customers who order it.

Lastly, this is not an attack on AI artists or artworks produced with AI. I sometimes use it myself, including in my most recent work. An artist can make art with milk cartons. It is never a question of medium, but what an artist does with it. You can potentially make totally legit art by exclusively prompting AI, and you can be equally relevant and interesting while completely rejecting AI and painting plein air landscapes oil paints. I know I could really get into either one, as well as the milk carton art.

I’m just pointing out that visual language is utterly packed with its own complex, nuanced, transcendent, and ineffable ideas, and that they are the core of visual art, just as musical ideas are the foundation of musical language. In this case, humans can collaborate with AI, but cannot dismiss AI’s contribution as merely following instructions, because typically the making of the image requires vastly greater ideas in visual art than any instigating idea in words. That said, the range of approaches to art-making, and various contexts, are so vast that each work needs to be judged individually and perhaps by its own native criterion. In the same way that my own drawing from my imagination is outside of the paradigm that says the idea is the art, there is certainly AI art outside of the scope of this article.

Stay tuned for the video version of this post.

---

UPDATE: The video is out. I’ve subsequently updated the article quite a lot, so the video is a bit different regarding the core text, but also includes a lot of extemporaneous commentary, including about the AI images I used as examples:

If you want to watch the video on YouTube rather than embedded here, us this link: https://youtu.be/OCB33sq_Wvg

---

And if you like my art or criticism, please consider chipping in so I can keep working until I drop. Through Patreon, you can give $1 (or more) per month to help keep me going (y’know, so I don’t have to put art on the back-burner while I slog away at a full-time job). See how it works here.

Or go directly to my account.

---

Or you can make a one time donation to help me keep on making art and blogging (and restore my faith in humanity simultaneously).