Under the AI Kimono

I asked an AI to write a digest of this article: "Japan has always blurred the lines between what is real and reproduction, making it an interesting place to observe the impact of generative AI.”

Something interesting happened last week — my wife and collaborator on many projects and books, Hiroko, tweeted a critique of an illustration of a woman in a kimono, and it went viral. (The person she’s quote-tweeting deleted their post amid the deluge of attention, but you can see all of the imagery in Hiroko’s thread.)

The image is an example of what is known as “generative AI,” a type of artificial intelligence that can generate text or images based on simple text based prompts from humans. The app used to create this particular image is called Midjourney; it was prompted by someone typing in the text “a full body photograph of a maiko princess” and letting the program dream up the rest. No human hand was directly involved in its creation beyond that prompt.

Midjourney isn’t the first application of this kind, but it is the best of the moment. It produces startling imagery based off very scant requests, filling in the blanks with… we’re not sure what. Midjourney, also the name of the company that created the AI, is understandably tight-lipped about what’s going on under the hood. There is informed speculation that it relies on “latent diffusion,” which involves “learning the underlying structure of a dataset by mapping it to a lower-dimensional latent space.” (Right, right, he nods sagely, not understanding any of this.) Essentially it is an algorithm that has been trained to create new things through exposure to a huge library of stock images.

On the one hand it is fascinating how far AI has come. A simple text prompt can generate something of such convincing quality. Then again, is it quality? Perhaps not, as you’ll see if you read Hiroko’s analysis. It looks startlingly real at first glance but isn't based in reality. It feels factual but can't be used as a factual reference. And only a specialist in the field, such as Hiroko, could tell you why. 

The image of a kimono-clad maiko – an apprentice geisha – has long been one of the most iconic symbols of traditional Japanese culture. It’s ironic that this style, seemingly frozen in time, manages to somehow capture everything that is disconcerting about such cutting-edge technology. This virtual maiko reveals how AI of the moment feels a lot like the guy in the room bloviating loudly and authoritatively about a subject, only to be deflated when someone more knowledgeable pops up and starts asking pointed questions.

Generative AI is something that didn’t exist in any publicly usable form when the revised paperback edition of Pure Invention came out in 2021, but the core tenet of that book was that the only way through our seemingly dystopian socio-economic techno-hellscape (i.e., the current moment) is to imagine the tools we need to help us navigate it.

Most of the gadgets and services chronicled in the book were initially greeted with derision, scorn, even outright venom. Everyone thought they were an expert. Anime was no more than “moving paper cutouts.” The karaoke machine would destroy the art (!) of bar conversation. The Walkman would never succeed because people hate wearing headphones. Video games would turn children into delinquents, or zombies, or both, somehow. But what happened? None of that, and far more than any of the critics could have imagined. All of these devices and more were incorporated into the fabric of daily life by consumers who used them in ways their creators could never imagine – particularly in the post-Bubble doldrums that so presaged our current post-Lehman shock, post (?) Corona moment. You can see it in how anime morphed from entertainment into an identity, how the Walkman became a sonic virtual-reality for brightening our humdrum lives, or how video games transformed from quarter-munching fads into immersive and even (sometimes) nourishing escapes.

Like all powerful and transformative new technologies, generative AI is incredibly divisive. Artists hate it because they feel it is illegally appropriating their work without permission or compensation. Analysts have raised troubling examples about how AI amplifies entrenched racial and gender biases. And almost everyone worries that the rise of easily-generated, high-quality AI imagery is going to spawn misinformation on a scale never before seen. They’re all right, and there is no question that the ability to generate seemingly realistic art and photos at the tap of a keyboard is going to prove disruptive. How are we going to chart these murky new waters? It’s scary stuff — but as the virtual maiko teases, Japan might have some answers.

In his classic essay “The Work of Art in the Age of Mechanical Reproduction,” the philosopher Walter Benjamin proposed the “aura,” an essence of original works that is missing from copies. In short, originals — whether natural (such as a beautiful vista) or man-made (such as a painting) possess an inherent authenticity that copies will inevitably lack. It makes intuitive sense; this is why we value the actual sight of Mount Fuji over a photo of it, or why the Mona Lisa is one of the most valuable paintings in the world, while posters of it sell for only a few dollars. But Japan has always thrown a monkey-wrench into this thinking, blurring the lines between what is real and reproduction, authentic and ersatz, forcing us to question our assumptions about the values we place on either. This is the land where the Grand Shrine of Ise, a cultural treasure with two millennia of unbroken history, is torn down and rebuilt every twenty years. Where some of the most beloved artistic treasures aren’t paintings but mass-produced woodblock prints. And the world loves them for it: a print of Hokusai’s Under the Great Wave of Kanagawa just sold at Christie’s for 2.7 million dollars.

Benjamin would have had a field day with the idea of generative AI, which extrapolates limitless “originals” that never existed from a sea of copies. But that’s the world we live in now, where even mass-produced objects possess an “aura.” That Hokusai may sound like a lot of money, but it’s only about half the $5.2 million a Pokémon card sold for in 2021 — tellingly, another sort of “print” from Japan. I suspect there is going to be far less resistance to AI in Japan than in the West, where we tend to frame new technologies in more adversarial, zero-sum ways. But that is a longer discussion for a later newsletter.

In the meantime, let me leave you with this thought. If writing Pure Invention taught me anything, it is that everyone thinks they understand how new technologies are going to disrupt things, but they’re almost always wrong. Frightening though the implications of AI may be, it’s important to remember that it will be our choices that chart a path for it – the individual choices of countless users and consumers swirling and merging, taming AI from a monster under the bed into a tool we can’t imagine living without. Just as has happened so many times before.