#185: The Painted Word

When you write about technology—or most subjects, really—there are certain topics that the internet seems to hand down like homework, demanding that Twitter’s vast volunteer army and the semi-professional Substack corps weigh in with a full spectrum of hot takes. The latest such topic has been AI art, specifically DALL-E (a welcome break from previous topics like crypto and the metaverse); as the thinkpieces flooded my inboxes and feeds over the last week, they began to feel like the products of DALL-E prompts themselves, churned out by the human hive mind as reliably as the software churns out its uncanny images. Anyway, you’re currently reading my contribution to the collective effort: A few days ago, Sachin Benny tweeted that “AI art is overproduction of elite overproduction,” a perfect rebuttal to the widely-articulated concern that tools like DALL-E and GPT-3 have finally grown advanced enough to render entire swaths of human creative effort obsolete. Peter Turchin coined the term elite overproduction to describe societies that generate a surplus of potential elites who cannot be absorbed into the power structure or given roles commensurate with their entitlement. As you might expect, elite overproduction is a source of social instability; Twitter is one effective (and probably necessary) release valve for the pressure it creates, a simulacrum where status is gamified and quantified, and where the illusion of inflated importance is manufactured for millions of well-educated “overproduced” users (I’m not denying my own participation in this, by the way).

If elite overproduction is a real problem, then AI art is a curse as well as a blessing. Content creation is one of the main activities left for humans to do! Machines should be outsourcing it to us, not the other way around (and they still do, of course). Many of us depend on content to keep us busy, if not to make a living. On the other hand, every major step forward for AI further erodes whatever remains of the pretense that so much online behavior hasn’t already become bot-like, ever responsive to the cues and affordances that our digital environments have carefully refined over time. Max Read describes how “Twitter users offer up their own strange, distributed language model en masse, producing and then voting on hundreds of thousands of possible completions for prompts like ‘She’s a ten, but,’.” Social media virality increasingly positions the human user as a relay station for memes rather than a source of original material—a disembodied avatar with a countable pair of eyeballs and a hand for clicking, for whom sharing is less and less distinguishable from creating. We could stop producing new content altogether and still have more than enough to keep remixing for decades. Every major advance in AI’s ability to produce visual art and literature seems like a red herring, a momentary passing of a meaningless torch and a distraction from the real problem: There is already an unimaginable surfeit of content, which humans needed no help overproducing. Automating its creation doesn’t solve the more urgent problems of people needing to keep themselves busy and feel important (it obviously makes those problems worse).

The current wave of AI art is surprisingly entertaining, even if the novelty is already starting to fade. But it serves another valuable purpose, as artificial intelligence always does: mirroring our own reality back to us, enabling us to see with objective detachment what would otherwise be too familiar to notice. The artistic output of DALL-E unintentionally reveals an increasingly pervasive quality of online communication—the demand for legibility and verbal shorthand. Several observers have noted that the text prompts used to generate DALL-E imagery are inseparable from the imagery itself. Twitter accounts like @weirddalle derive much of their humor from the cleverness of the prompt (the only part of the process with direct human involvement, interestingly). Avery Trufelman compares this to the “random” humor of the ‘00s internet and suggested to me that the written prompts could eventually stand alone as shorthand for the whole process—like tweets, basically—with no need for the AI-generated imagery as long as the text evokes it. More broadly, today’s meme-saturated internet depends upon such shorthand—tags, headlines, metadata—despite the prevailing myth that it has taken a visual or post-literate turn. Molly Fischer has written that the essential characteristic of “Instagrammability” is readability: “The viewer could scroll past an image and still grasp its meaning, e.g., ‘I saw fireworks,’ ‘I am on vacation,’ or ‘I have friends.’” Just as an article’s headline often overpowers the piece itself, becoming its entire message, much visual content must be reducible to a brief unambiguous statement, whether the text is actually displayed or not. As streams of such content endlessly flow by, we rarely have time to ponder them long enough to absorb their more complex or nuanced implications. Memes rely upon text because it’s the most efficient way to convey a specific message. What would all that DALL-E art amount to if we cropped out the written prompts? Most of it would suddenly become so abstruse that we’d never even notice it scrolling by.

This newsletter is supported by paid subscriptions, which give you access to an extra issue (in addition to the free Friday issue). Sign up to get the full experience.

Last week’s paid issue was about the Chesa Boudin recall and San Francisco’s “fourth world” conditions.

Reads:

• When Baking and Real Estate Collide. Anna Wiener on Bay Area bakery Tartine and its partnership with a major developer in pursuit of growth.

• Robin Sloan’s excellent interpretation of AI art. “The pleasure, it seems, is not in the image; rather, it’s in the spectacle of the computer’s interpretation.”

• I wrote an essay for Dirt about how Netflix and other streaming services block screenshots, and why doing so is counterproductive.