C. S. Lewis on AI art

(well, sort of)

In Scott Alexander’s Friendly and Hostile Analogies for Taste, a follow-up to his AI Art Turing Test post, he expresses puzzlement about what people who appreciate “sophisticated” art are getting out of it, compared to “kitsch” art:

This is a bit mysterious. Many (most?) uneducated people like certain art which seems “obviously” pretty. But a small group of people who have studied the issue in depth say that in some deep sense, that art is actually bad (“kitsch”), and other art which normal people don’t appreciate is better. They can usually point to criteria which the “sophisticated” art follows and the “kitsch” art doesn’t, but to normal people these just seem like lists of pointless rules.

Coincidentally, as part of some internet rabbit hole last week I ended up reading C.S. Lewis’s An Experiment in Criticism, which discusses exactly this topic. I don’t find his answer totally satisfying, but I do find it interesting, so I wrote it up.

(By the way, I recommend trying the test, even if you’re like me and lost the original test invite somewhere in your emails, and also didn’t have the self-control to wait to read the discussion post until after you’d taken it. Scott talks about five pictures in the main text of that post, so those are spoiled for you, but there are still 45 left, which is going to feel like plenty once you’ve been staring at zoomed-in details for a while. I got 38 right and 7 wrong, for a total success rate of 84%, which seems not amazing, particularly as I might have picked up some tricks from the post, but not bad either. Hopefully enough to qualify me to spout off about it in this blog post at least!)

The main focus of Lewis’s book is on what separates “literary” readers, who love books and read their favourites over and over, from “unliterary” readers, who will sometimes read a book to pass the time if no more interesting activity is available, but are insensitive to writing quality and rarely read the same book twice. The “experiment” of the title is an attempt to see what this focus on readers can tell us about books, rather than following the traditional critical approach of studying the books themselves:

Literary criticism is traditionally employed in judging books. Any judgement it implies about men’s reading of books is a corollary from its judgement on the books themselves. Bad taste is, as it were by definition, a taste for bad books. I want to find out what sort of picture we shall get by reversing the process. Let us make our distinction between readers or types of reading the basis, and our distinction between books the corollary. Let us try to discover how far it might be plausible to define a good book as a book which is read in one way, and a bad book as a book which is read in another.

As a warm up, he first considers the same question for visual art. Lewis’s own tastes changed at a later age for pictures than for books, so he remembers more about the transition.

As a child he enjoyed book illustrations like those of Beatrix Potter and Arthur Rackham, but it was a different kind of enjoyment to the way he appreciates art now:

When I now turn their pages I by no means say ‘How did I ever enjoy such bad work?’ What surprises me is that I drew no distinctions in a collection where the work varied so vastly in merit. It now stares me in the face that in some of Beatrix Potter’s plates you find witty drawing and pure colour, while others are ugly, ill-composed, and even perfunctory. (The classic economy and finality of her writing is far more evenly maintained.) In Rackham I now see admirable skies, trees, and grotesques, but observe that the human figures are often like dummies. How could I ever have failed to see this?

His answer is that a child he was mainly interested in what was depicted, rather than the details of the pictures themselves:

I liked Beatrix Potter’s illustrations at a time when the idea of humanised animals fascinated me perhaps even more than it fascinates most children; and I liked Rackham’s at a time when Norse mythology was the chief interest of my life. Clearly, the pictures of both artists appealed to me because of what was represented. They were substitutes. If (at one age) I could really have seen humanised animals or (at another) could really have seen Valkyries, I should greatly have preferred it. Similarly, I admired the picture of a landscape only if, and only because, it represented country such as I would have liked to walk through in reality. A little later I admired a picture of a woman only if, and only because, it represented a woman who would have attracted me if she were really present.

This fits with the pattern of popular pictures mostly being of something that would also be popular in real life:

Nearly all those pictures which, in reproduction, are widely popular are of things which in one way or another would in reality please or amuse or excite or move those who admire them—The Monarch of the Glen, The Old Shepherd’s Chief Mourner, Bubbles; hunting scenes and battles; death-beds and dinner parties; children, dogs, cats, and kittens; pensive young women (draped) to arouse sentiment, and cheerful young women (less draped) to arouse appetite.

I feel like Lewis could do a lot more at this point to explain why, if the topic is the important part, people want to look at pictures at all, rather than, say, a piece of paper with RIDE OF THE VALKYRIES or SMALL CHILD BLOWING BUBBLES written on it. Presumably something about the visual image itself is important… but then, what separates the experience of the image that the unsophisticated viewer is having from that of the sophisticated one?

Anyway, Lewis’s claim is that, whatever the unsophisticated viewer is doing, it doesn’t involve much careful looking at the details of a picture. The viewer is more “looking through” the picture, using it as a way to evoke a cloud of associations:

The result, as I now see, was that I attended very inadequately to what was actually before me. It mattered intensely what the picture was ‘of’; hardly at all what the picture was. It acted almost as a hieroglyph. Once it had set my emotions and imagination to work on the things depicted, it had done what I wanted. Prolonged and careful observation of the picture itself was not necessary. It might even have hindered the subjective activity.

In Friendly and Hostile Analogies for Taste, Scott Alexander considers more abstract, sociological analogies. Maybe taste is like fashion? Or tastemakers are like the priesthood? But, interestingly, Lewis’s argument is there in the original AI Art Turing Test post. Scott quotes a friend, “Ilzo”, who does digital art, and who took a particular dislike to this AI-generated image:

“Ancient Gate”, AI art by Piotr Binkowski, from Scott Alexander’s post

Here’s a cut down quote of Ilzo’s objections:

When real pictures have details, the details have logic to them. I think of Ancient Gate being in the genre "superficially detailed, but all the details are bad and incoherent". The red and blue paint and blank stone feel like they're supposed to evoke worn-ness, but it's not clear what style this is supposed to be a worn-down version of... There are matchy disks in the left, center, and right, except they're different sizes, different colors, and have neither "detail which parses as anything" nor stark smoothness. It has stuff that's vaguely evocative of Egyptian paintings if you didn't look carefully at all. The left column has a sort of door with a massive top-of-doorway-thingy over it. Why? Who knows? The right column doesn't, and you'd expect it to. Instead, the right column has 2.5 arches embossed into it that just kind of halfheartedly trail off.

Ancient Gate is an example of exactly the kind of AI art that I also find really, really annoying, so I was definitely the target audience for this rant. I really wanted to quote the whole thing, but I had a feeling that “looking at the details of images” and “reading the details of quotes” might be correlated and I’d lose exactly the people I’m trying to reach, so I shortened it. But the rest of it is great, go and read it.

I’m not especially into art, and I’m not knowledgeable about it, but I also like to look at the details of things, and this kind of image is very frustrating if you care about details. AI images like the pretty landscape and evening street scene are fine. There’s not much point looking at either for long, but that’s OK, because there isn’t much detail there in the first place to trick you into trying. I happen to like the style and colours of the landscape, whereas the street scene does nothing for me, but neither one is going to bother me if it’s at the top of a blog post or something. A lot of the AI art I see is in this category and I don’t mind it. It’s in a similar category to stock images, which are also mostly uninspired space fillers.

Ancient Gate is much worse, because it looks like it has details. It looks like it could plausibly be worth looking at. So I start paying attention, but it turns out to just be a load of incoherent greebles. It’s an invitation to waste my time.

I don’t have as good an eye as Ilzo’s, but I mainly used a similar technique in the AI art test. I don’t know much about good technical tricks for spotting AI art, other than the ones everyone knows like too many fingers, so my main strategy was to actually look at the picture, like properly look at it, and think about the details I saw. The sort of questions I thought about were things like … does this picture make any sort of global sense? Does the action in the foreground relate to what I can see in the background? If there are people in the scene, are they doing things with some discernible purpose, or are they just kind of dotted around to fill up space? How do they relate to other nearby people in the scene? Do they have a shared purpose or are they just standing next to each other? If an object in the scene is covered in decoration, is there some sort of overall harmony and sense to the decoration, or is random bits and pieces?

Overall this got me pretty far. There were pictures by human artists where I was confused by exactly what was happening, either because they were old and I was missing context or because they were plain weird. But still, I could normally tell after looking for a while that some coherent thing is happening. The difficult cases were the scenes that were either so simple or so conventional that it seemed plausible that either a human or an AI could have made it.

I’ve mainly covered Lewis’s argument for what sophisticated viewers are doing. In the Analogies for Taste post, Scott also asks whether there is a useful purpose to developing your taste. Is taste Like A Priesthood, But Good And Important, or is just a bunch of Semi-Fake Justifications? Is it valuable to see paintings in this way, or just an annoying skill that makes your life worse by surrounding you with bad art you hate?

An Experiment in Criticism does sort of cover this. As you would expect for a book by C. S. Lewis, Christianity is not far below the surface, even if it isn’t discussed directly in this chapter. You can hear the echoes in this paragraph on surrender:

Real appreciation demands the opposite process. We must not let loose our own subjectivity upon the pictures and make them its vehicles. We must begin by laying aside as completely as we can all our own preconceptions, interests, and associations. We must make room for Botticelli’s Mars and Venus, or Cimabue’s Crucifixion, by emptying out our own. After the negative effort, the positive. We must use our eyes. We must look, and go on looking till we have certainly seen exactly what is there. We sit down before the picture in order to have something done to us, not that we may do things with it. The first demand any work of any art makes upon us is surrender. Look. Listen. Receive. Get yourself out of the way. (There is no good asking first whether the work before you deserves such a surrender, for until you have surrendered you cannot possibly find out.)

Even without the religious angle, though, it seems obvious to me that this orientation to art gets you something valuable. If I turn off my expectation of what I think is there, and open up to what the artist has put there, I get to see something new. As Lewis puts it when he talks about literature later in the book:

We are not content to be Leibnitzian monads. We demand windows. Literature as Logos is a series of windows, even of doors. One of the things we feel after reading a great work is ‘I have got out’. Or from another point of view, ‘I have got in’; pierced the shell of some other monad and discovered what it is like inside.

It’s hard for me to imagine really looking at that Ancient Gate picture, paying attention to the actual details that are there in the arrangement of pixels in front of you, and still liking it. If Scott genuinely likes it, I assume he’s doing something else. Maybe, as Lewis suggests, he’s looking through it to some more compelling associations that it evokes in him. Scott’s way of engaging with the world is clearly insanely generative, he’s produced this vast corpus of distinctive writing, so I could imagine that these could be interesting in themselves.

Still, even if he’s seeing something good there, and I could learn to see it as well, I’d still want to be able to drop both his associations and my associations and try to see what the artist saw. Even if this ability mostly means that I have to keep wasting my time surrendering my attention to yet another piece of meaningless slop.

---

Edit: I just discovered that this isn’t even the first post relating An Experiment in Criticism to Scott Alexander’s post on taste this week! You should also read Contra Scott Alexander on Taste by sympathetic opposition.

This post is a bit of an experiment in writing more blog-post-like things here, rather than my normal more newsletter-y monthly ramblings (which I’ve skipped for a few months while I explore various rabbit holes). I finally accepted recently that I was done with my old blog, Drossbucket, because WordPress had got too painful to use, and it made me realise that I’d like to do more of that type of writing again here. I like both formats, so maybe I’ll do a mix, I don’t know.

The cover image is this Wikimedia Commons photo by Olaf Tausch of an ancient gate at the Temple of Hathor in Dendera, Egypt, which I added to save having to look at the AI gate any more.