My Five-Year-Old (Computer) Could Make That!

Supply, demand, and how the Steam Store helped me feel less anxious about the threat of machine learning

Featured image is a weird portrait of me as a barbarian or something, generated by one of those Midjourney-style AI toys

There’s a style of character design that I used to like a lot, until recently. It’s that exaggerated-but-still-kind-of-heroic cartoony style that I most associated with games like the Torchlight series. It managed to be appealing but not too simplistic, somewhat painterly but still approachable, bringing a novel and intentional, artistic, interpretative design to a medium that for too long seemed to prize photorealism above all else.

Now, though, I associate it with cheap, exploitative, pay-to-play games. It more or less became the de facto art style for every one of those mobile- or Facebook games that were advertised incessantly, but still managed to do absolutely nothing that I could recognize to distinguish themselves. That style of character design went from something that I aspired to, to something that I viscerally disliked, just because the market became flooded with it.1Ironic, since so many of these games seem to involve trying to save a character from a flooded, trap-filled room.

Lately there’s been a ton of talk online about machine-learning-disguised-as-artificial-intelligence, and how it’s ruining search engines, and threatening the jobs of artists and writers. The apprehension about ML taking over search engines is completely justified, since these are already tools that are relied on to deliver deterministic, predictable results.2More or less. I’m aware that Google’s algorithm for recommending search results is byzantine, and that an entire industry has developed around improving SEO. Replacing them with bullshit-generating algorithms is categorically a terrible idea.

In terms of ML replacing, devaluing, or stealing creative work, I’m not so sure. In the short term, absolutely. These systems are being trained on artists’ original work, and not only is it impossible to credit or compensate the original artists, it’s increasingly becoming the argument of ML apologists that the original artists shouldn’t be credited or compensated, because it’s somehow no longer their work. But in the longer term, I feel like it’s going to pretty quickly reach a state of equilibrium, where the art that’s being devalued is the art that already has the least value.

What convinced me: I follow a musician and artist on Mastodon who frequently posts links to indie games that have just become available on the Steam store. Several times now, I’ve seen a piece of game art that immediately gets my attention: technically proficient, perfectly rendered, instantly compelling. I always think, “wow, an indie game has that level of art? The bar’s been raised a lot!”

And almost every single time, I’ve gone to the Steam’s game page and discovered that the image I saw was a title screen that looked absolutely nothing like the in-game art. It’s quickly clear that the developer(s) generated an image using DALL-E or Midjourney or some such, put it up as the title image to get attention and lend some “professionalism” to the project, and then hoped for the best.

And since it tends to get my attention, you could conclude that it works exactly as intended. But there’s diminishing returns. I’ve only been aware of these algorithms for about six or seven months now, and I’ve already learned to look for the telltale signs of ML-generated “art” and become dismissive of it. It’s already been commodified.

The inescapable fact about ML systems — at least in our lifetime — is that the more sophisticated they get, the more boring they get. Janelle Shane has already lamented that she’s having a harder and harder time getting anything interesting out of ChatGPT. And on the art side, the “surreal daydream” period in generative art was the last time it was actually novel enough to be genuinely interesting.

In fact, calling it “generative” is itself a little too generous, since it gives the connotation that these systems are creating something new. But they’re only capable of generating anything based on the input they’ve been given. No matter what the “everything is a remix!” “all art is interpretive!” crowd tries to tell you, that’s not the same as real art. It’s like the old, old story about the curmudgeon in the modern art museum, griping “My five-year-old could make that!” The response: “But she didn’t.”

When ChatGPT isn’t just outright lying, it’s generating boring, completely predictable writing. I hate that companies are choosing to use these systems instead of hiring writers, but at the same time, they kind of writing they’re replacing isn’t rewarding or particularly valuable. It’s for content farms like BuzzFeed, or sites that just regurgitate press releases for products. I’m optimistic there’s always going to be a market for original, entertaining, insightful writing.

For code, I should probably be a bit more concerned, since my most bankable skill is being a thoroughly adequate-but-not-exceptional programmer. But here again, I’ve seen what happens when companies hire programmers who know how to use Stack Overflow but lack the ability to understand how the code snippets actually work, and can’t diagnose problems with it. ChatGPT-generated code is likely going to settle into a comparable equilibrium.

And I’m not an artist, but I feel even less concerned about the long-term impact of the ML-generated visual art generators than the text ones. It’s going to be useful for vanity projects (“I want a self-portrait of me as a bad-ass cyborg!”) and weird meme generators (I confess that I kind of love the whole Some Property as an 80s Dark Fantasy Film” meme going around on YouTube).

The attention around these things isn’t going to last very long, because the thing that makes them so easy to generate is the exact same thing that makes them not valuable. Stuff only gets traction if already-rich people can make even more money off of it. A market flooded with ML-generated images is going to do nothing to increase the value of ML-generated images; it’s only going to increase the value of real works of art.

The hype around these systems — which extends to calling it “AI” in a I-suspect-deliberate attempt to make them seem more powerful than they actually are — is the same kind of gold-rush that’s happened over and over again. They are remarkable, and there are undoubtedly tons of interesting applications for them, for people who think of them as artists’ tools, instead of as artists.

  • 1
    Ironic, since so many of these games seem to involve trying to save a character from a flooded, trap-filled room.
  • 2
    More or less. I’m aware that Google’s algorithm for recommending search results is byzantine, and that an entire industry has developed around improving SEO.