Yes, AI makes “Art”

I’ve liked Ted Chiang’s editorials in the past about AI, but this latest one in the New Yorker, which loudly announces “Why AI Isn’t Going To Make Art,” is just plain old wrong.

It’s dizzying to figure out where to even start, so I’ll just go through in order. Ted starts out with a sort of spurious definition, I think, of art:

“…art is something that results from making a lot of choices…”

Pretty darn vague. A bit reminiscent of Scott McCloud’s definition of art from his 1993 landmark, Understanding Comics:

“Art, as I see it, is any human activity which doesn’t grow out of either of our species’ two basic instincts: survival and reproduction.”

I find McCloud’s version to be a bit more workable, but we’ll have to set that aside as we dig deeper into Chiang’s arguments… (Shopping for a winter coat online, for example – which I’ve been doing lately – requires tons of choices, and is absolutely ‘not art’ just on its own. But back to Ted:)

His basic premise, as I understand it, is that the act of writing text by hand is “choosier” than the act of… writing text… which results in AI generations, a.k.a. writing a “prompt.” Then, it seems that he’s making the value judgement that things which consist of more human choices result in end products that are “artier” and therefore better. Uh, okay… moving on.

He then launches into comparisons with the advent of photography, which gen AI is often compared to (and I think rightfully so):

When photography was first developed, I suspect it didn’t seem like an artistic medium because it wasn’t apparent that there were a lot of choices to be made; you just set up the camera and start the exposure. But over time people realized that there were a vast number of things you could do with cameras, and the artistry lies in the many choices that a photographer makes. It might not always be easy to articulate what the choices are, but when you compare an amateur’s photos to a professional’s, you can see the difference. So then the question becomes: Is there a similar opportunity to make a vast number of choices using a text-to-image generator? I think the answer is no. An artist—whether working digitally or with paint—implicitly makes far more decisions during the process of making a painting than would fit into a text prompt of a few hundred words.

This seems short-sighted to me. First we’re to go along for the ride that when photography first came out, people didn’t grasp all the choices that went into it. (I’m first off not so sure that was the reason it was disdained.) But over time and on closer examination, people got it. But then, we’re to believe (without any supporting evidence) that the same opportunity to more closely examine “generativist” AI art and gain new insights into all the choices that go into it on the part of the human artist simply won’t happen because… reasons? It’s not really clear to me why this exact same arc he’s describing won’t happen with AI – simply because he doesn’t want it to, I guess?

Also, I think this entire essay suffers from a fatal flaw, that it considers the “art” that is made by an artist using AI to simply be the final one image or one text that it ultimately yields. That is, one prompt = one image output, done deal. He is missing the critical conceptual innovation that I have termed as the “hypercanvas.”

What I mean when I say hypercanvas is something like, when you make a painting, it is composed of many individual brush strokes (each of which has its recognizable “choosiness” in Chiang’s thinking). But gen AI doesn’t work like that. Each time you do a prompt and get an output, each of those actions constitutes the equivalent of your “brush strokes” on the higher-dimensional space that the “artwork” inhabits, or as I’m calling it the hypercanvas.

If we think of it like this, Chiang’s argument falls apart:

An artist—whether working digitally or with paint—implicitly makes far more decisions during the process of making a painting than would fit into a text prompt of a few hundred words.

The most meaningful unit of comparison here is not between a finished painting vs. a prompt + image output, but between an individual brush stroke on a conventional canvas, and one on a hypercanvas. When we make that more accurate comparison, we can see that, hm, maybe the placement of a brush stroke on a conventional canvas might even have LESS “choosiness” than all the myriad possibilities and parameters open to us when composing prompts, or choosing & iterating image outputs.

Also, his own argument about AI art processes being incapable of “choosiness” is immediately after discredited by his own example:

The film director Bennett Miller has used DALL-E 2 to generate some very striking images that have been exhibited at the Gagosian gallery; to create them, he crafted detailed text prompts and then instructed DALL-E to revise and manipulate the generated images again and again. He generated more than a hundred thousand images to arrive at the twenty images in the exhibit.

I don’t know, that sounds like an awful lot of “choices” to me, Ted. It’s almost like this person is – gasp – using AI to make art??

It’s difficult to get past what I experience as something like willful blindness that crops up again and again in this piece, like in this apparently not tongue in cheek bit:

Generative A.I. appeals to people who think they can express themselves in a medium without actually working in that medium. But the creators of traditional novels, paintings, and films are drawn to those art forms because they see the unique expressive potential that each medium affords. It is their eagerness to take full advantage of those potentialities that makes their work satisfying, whether as entertainment or as art.

Hm, “the unique expressive potential that each medium affords” – um, you mean like in the medium of generative AI? Yes, I said it, this is an artistic medium, with forms, processes, conventions all of its own. It’s so blazingly obvious that I don’t even know why I have to fight strawmen on the internet just to be able to express it.

There’s a lot that I take exception to in the original piece, but I will have to be choosy for the sake of economy here. How about this one:

The point of writing essays is to strengthen students’ critical-thinking skills; in the same way that lifting weights is useful no matter what sport an athlete plays, writing essays develops skills necessary for whatever job a college student will eventually get. Using ChatGPT to complete assignments is like bringing a forklift into the weight room; you will never improve your cognitive fitness that way.

This is, in my experience, dead wrong. Like I wrote in the Register interview that was published over the weekend, using AI to help me write has taught me to write better. There’s no two ways around it.

AI has made me a vastly better writer. I’ve been writing for a few decades now, personally and sometimes professionally. But there are certain things I’ve always fallen short in, certain forms of structured writing and logical flow of arguments especially which have always eluded me. LLMs tend to excel at this kind of writing, even if their outputs can sometimes tend toward the vanilla. So the ability to have this tool, this writing partner, to bounce my ideas off of, and who can rapidly produce semi-usable results has been incredible. It’s not strictly a question of enhancing productivity or volume of work that I can create (though it’s that too), but this interrogative way of working has rubbed off on me, and the AI tools have taught me how to actually think more logically and clearly about problems, and then to more plainly organize those thoughts and communicate them with others.

Ted Chiang is wrong. He is also wrong about this:

It is currently impossible to write a computer program capable of learning even a simple task in only twenty-four trials, if the programmer is not given information about the task beforehand.

Dead wrong. This is, as I understand it, exactly what “reinforcement learning” (RL) is in the world of AI and robotics. This has been going on for years, but here’s a tweet from just a few days ago about an open-source DIY plan where you can teach robot arms to fold a shirt [click through for the video because it didn’t embed here properly]:

Again, this isn’t some kind of recent innovation. It seems to suggest this New Yorker piece wasn’t really fact-checked all that carefully before being published.

Lastly, Chiang concludes:

Whether you are creating a novel or a painting or a film, you are engaged in an act of communication between you and your audience. What you create doesn’t have to be utterly unlike every prior piece of art in human history to be valuable; the fact that you’re the one who is saying it, the fact that it derives from your unique life experience and arrives at a particular moment in the life of whoever is seeing your work, is what makes it new.

I mean, what else is there to say in response that isn’t simply repetition at this point? What he’s describing holds true regardless of the medium or technology used. Let’s not keep having these same old arguments again and again. It’s tired and doesn’t get us anywhere new. And ust because it’s published in The New Yorker doesn’t make it gospel.

Related writing & research