The Etruscan Horse
Hearing more and more about the “incredibly realistic” abilities of AI art, I am reminded of one of the most important stories I know, the story of the Etruscan Horse.
In 1886, a museum in Britain procured a rare artifact from antiquity, an Etruscan statue of a horse. They staged an exhibit, with this new acquisition as its centerpiece.
One hundred years later, in 1986, the same museum said “Hey, you know what? It’s 100 years later, let’s stage a revival of that show from 1886 that was a big hit for us.”
So they set about prepping the revival of the show, and they sent curators into their basement and the curators found the crate that the Etruscan Horse was in and opened it up, and immediately saw, without any special tools of identification, that the sculpture they had put on display at the center of their exhibition in 1886 was an obvious forgery, with all the marks common to metalworkers in Victorian England.
The museum staff of 1886 had not been idiots, they were simply unable to see the marks of their own times in the sculpture.
I think about the Etruscan Horse on a regular basis, because no artifact can ever be anything but a product of its time. (I make forgeries of 20th-century commercial art for a living, so I think about this sort of thing probably more than other people.) In the same respect, no person can ever fully understand their times, because they themselves are a product of those times. You cannot be part of a vast system and also be able to describe that system from the outside.
I watch a lot of old movies, and I’m a student of special effects, and so when I watch an old movie there’s always a little soundtrack in my head that runs parallel to whatever the movie is doing. I’ll sit there thinking “Miniature, water tank, matte painting, miniature, transposition, over-printing, double-printing, enlargement, miniature” etc.
The thing is, none of the people who saw those old movies in theaters sat there thinking about any of that. If a movie opened with a shot of a mansion on a tree-filled estate, they accepted it for what it said it was, not a miniature built in a studio somewhere at 1/40 scale.
I remember, as a child, watching the peerless Titanic drama, A Night to Remember, and thinking, as the Titanic sank, “why do the waves look so weird?” Watching A Night to Remember reminded me of Godzilla, where Godzilla would be wading into Tokyo Bay and the waves, similarly, looked weird and blobby.
Of course, the water in those movies looked like that because the filmmakers were trying to make a small thing, a model ship or a man in a rubber suit, look gigantic. So they shot their subject at a faster speed, to give the action a sense of scale.
The definition of a special effect is a device that makes it possible to show something in a movie that would otherwise be impossible to shoot. Nowadays, of course, special effects are used for absolutely everything, from planets exploding to changing the pen that a character is holding. But the reason that so many special effects work in their time is that the audience, by definition, has never seen what the special effect is showing (a giant ape climbs a skyscraper, an ocean liner turns upside-down, dinosaurs eat people), so they have nothing to judge it by. I’d certainly never seen the Titanic sink, or a ghost, or a superhero fly through the air, as far as I knew that’s what those things looked like.
Whenever an effective new special effect comes along, there’s a moment when everyone in the audience says “that’s amazing, that looks completely real.” And then that special effect is used in everything, and then it’s no longer special, it just becomes part of the vocabulary of the form. When Disney+ first started showing their Star Wars shows, they looked amazing, with real actors in real locations interacting with real sets and props. A year later, I started to notice that there was a weird lassitude in those shows and a numbing sameness.
The innovation was that Lucasfilm figured out how to build 3D digital environments and then put those on 360° HD screens all around the actors, so that the actors were actually lit by whatever the illumination on the screens was. They appeared to be in a desert or on a spaceship or whatever because, in a way, they were, there were no green-screen lines or digital composition artifacts because it was being shot live.
But by the time the third or fourth show came along, I started to notice that, while the actors looked like they were in the environment they were supposed to be in, they never walked to the edge of it. The scenes were always in the center of the room, because that’s how the technology worked — if they were to walk over to fetch something from a table in the background they would bump into the LED screens. So they had to stand in the middle of the room and talk, in the middle of these extraordinary digital sets that looked great but could not be touched because they didn’t exist. And so the special effect that made the show possible also made it boring as hell.
AI is going to be that. Any special effect that is “cutting edge” will always — always — look dated fastest. That’s why special effects fans like me always prefer practical effects to digital effects: not because the practical effects look more convincing, but because the light in the scene is hitting everything equally, and the eye can tell that.
So when some AI hack says “This is amazing and it will only get better,” I always think of the Etruscan Horse, and how something may look amazing today but will, in just few year’s time, look obvious and clumsy.
The latest example touted shows a group of yellow lab puppies romping in snow. Why did the AI people choose that prompt? Because they knew we’d be looking at the puppies and feeling all the things we feel associated with puppies and not noticing all the marks common to Victorian London metalworkers.