Sam Littlefair

The Artifice

A new art app paints a picture of a bleak future.
  • writing

A new "artificial intelligence" tool called "Lensa" burst into social media this week. For $7.99, Lensa will upload a handful of photos of a person and output a collection of stylized avatars. Social media flooded with users sharing their new pop-art profile pictures, making them look like characters in comic books or movie posters. Critics quickly followed, attacking the software that powers Lensa.

"Artificial" comes from the latin ars, meaning art or skilled work — what we might call "craft" today. Artifex means craftsperson. Artificium means skilled work; the adjective form is artificialis, as in artificial. Ars comes full circle in the phrase "artificial intelligence art": crafting intelligence that itself can craft.

But I use scare quotes around "artificial intelligence," because AI is a buzzword, not a real thing. When people say "AI," usually they're referring to a computing practice more accurately called "machine learning." Machine learning is a type of computer program that self-corrects based on parameters defined by human engineers, which they commonly call "learning."

But machine learning never actually learns, because its metrics of success and calibration are always defined by human engineers. To demonstrate this, look at the mistakes that come through in machine learning models like the one that powers Lensa.

Women who use Lensa report that it outputs highly sexualized and even nude images. Women who input only headshots will get back full-body shots with huge breasts and a skinny waist. Some women find that they get more accurate and less sexual images by setting their gender to "male." Asian users say that the app makes them look more stereotypically Asian.

The old-school definition of AI is a technology that can think for itself. Obviously, Lensa has no power to think. It consistently makes the same absurd mistakes. These aren't thoughts: they're the recycled work of the immature white men behind the curtain.

Our definition of AI has shifted over time, largely due to its over-use. Like the word "literal" coming to mean "figurative," the phrase "artificial intelligence" has come to mean "pretend intelligence." AI today is technology that mimics human intelligence, which — compared to the promise of technology that can actually think for itself — is a much lower bar.

Is a barometer, which can tell whether or not its raining, an AI? Is a nine-digit calculator?

That's a pedantic argument, but there's a real point: we use words like "learning" and "intelligence" to humanize technology. But at its core, most machine learning programming is proprietary software that generates value through cost-cutting by replacing human labor.

When we talk about AI art, we often fall into the trap of whether a computer can produce art. Are the Lensa portraits art? That question misses the point. The rarified ideal of "art" as separate from (and higher than) craft is a relatively new idea — and a strange one.

Art is labor. The creation of art demands skill, time, and effort. And, while some art may seem to transcend the artist, most art is the product of mundane and repetitive work — photo portraits, upholstery patterns, birthday cards, social media avatars. And, work that is mundane and repetitive can be automated.

Computers automate repetitive tasks — even nuanced tasks like painting a portrait. Does that mean that the computer is intelligent or artful? No. A computer has no "artificial intelligence" just like a cotton gin has no "artificial skill." This isn't skill or intelligence — it's a machine that has been designed to mimic the skilled work of humans.

What makes humans different is that we can think for ourselves. And human artists spend thousands of hours studying artwork from around the world to develop their skill and style.

Maybe this is what Lensa has automated: the free and open tradition of artistic inspiration. After all, the greatest outrage over Lensa is how the company trains its model on billions of source artworks — many copyrighted, none credited. The company has sucked up the collective knowledge of mankind (with an apparent over-representation of white male fantasy comic book artwork) in an attempt to establish a sort of ownership.

Lensa claims to bring art to the masses. In truth, they're taking art from the masses. They offer nothing new or superior in the world of pop-art portraiture. Instead, they compete by undercutting the human craftspeople who make their livelihood as portrait artists — turning the production of art into a factory process.

It reminds me of the textile artisans who lost their livelihoods through the 18th and 19th centuries as new technologies arose to produce bolts of cotton cloth by the mile. The factories turned cloth into a cheap consumer good. Undercut by industry, the artisans lost the market for their goods, lost their livelihood, and lost their traditional craft. They protested in the streets, but the police and government responded with gunfire. Across Europe, rebellions failed and textile craft disappeared.

Though Lensa hasn't created a thinking machine, I think they have created something new. They might be the first company to sell "artificial intelligence" as a cheap consumer good. The consumers pays $7.99 in exchange for a few seconds of processing time on the Lensa servers, cutting out the millions of artists whose work built the software. Like the weavers centuries ago, artists today will lose the market for their intelligence, undercut by a pretend intelligence.

So, no, Lensa does not sell art, and they don't sell artificial intelligence. But they still sell ars in one sense — that of the artifice: a clever or cunning device used to trick others. You might think you're buying a computer's original thoughts. In fact, you're just cutting the artist out of the art. Call it the art of deception.

© Sam Littlefair 2024