Say you were dumb, I mean really dumb, computer dumb, literally dumb as a rock, albeit one with some lines scratched into it. And say I showed you ten pictures of dogs–and I told you they were all dogs–and ten pictures of cats–and I told you those were all cats–and then I showed you one more…would be able to tell me whether the new one was a dog or a cat?
If you were truly computer dumb you wouldn’t have a clue. Your guesses would be little better than random, mental coin tosses and a hopeful expression.
But you aren’t computer dumb. The youngest child, ones too young to write, too young to read, even too young to speak in anything approximating a sentence, can tell a dog from a cat with almost perfect accuracy.
Computers though, really are computer dumb. They seem so smart sometimes–and so much work has been put into trying to make them smart–but really they are dumb rocks, just very fast dumb rocks with incredible memories.
A funny thing happens, though, when you show computers more pictures of cats. Show them ten pictures and they are bewildered. Show them a hundred and there is no improvement. Show them a thousand and maybe they are a little better. But show them a million dogs and cats, show then one hundred million dogs and cats and suddenly they know dogs and cats. They know them if seen from the front, they know them if seen from the rear, then know them up close, they know them from afar.
This big breakthrough–called Deep Learning or Machine Learning–the reason computers have gotten so much smarter in the past decade, able to recognize faces, able to drive cars all by themselves, able to read x-rays and detect cancers more accurately than trained doctors.
But to do any of this you need data, lots and lots of data. A crazy amount of data. For example, if you wanted to train a computer to recognize faces (so you could match up these faces later with real people) you would be smart to offer a free online photo service where people could share their phone images with friends and family. Without clearly telling people (because they might find it creepy and not use your service) you would mine this vast resource–Facebook alone has 350 million photo uploads a day–for all sorts of Deep Learning magic, all tied to the accounts of real people and to all their friends and associates.
And recognizing faces is oh so just the beginning.
In 2015, the Google AI Blog posted a research paper entitled “Inceptionism: Going Deeper into Neural Networks” which described an unexpected discovery made about the “neural networks” that undergird Deep Learning.
You could train a neural network to recognize cats vs dogs but you could also “turn the network upside down” and ask the network to find a dog or a cat in an image–force it to find a dog or a cat, even if there was no dog or cat–and then to display the thing it found in the image. In the extreme case you could start with a picture of colored static–random pixels–and the upside-down network would find stuff. And if you feed that resulting image back into the neural network as a new input you found magic.
It was useful no doubt as a diagnostic tool in building these neural networks but, my god, it produced such cool, alien, DeviantArt images, starting from almost anything. Each time you ran the program–later called Deep Dream–you got different results, and you could choose different “layers” of the Dream and let the Dream run for longer and shorter times, all of which resulted in different kinds of magic.
Last year I was working on a book project that veered from being about Donald Trump to being more about the polarization of society due to technologies like Deep Dream and it struck me that it might be fun to take this technology and apply it to Mark Zuckerberg himself.
And so I did. I ran three computers at once–my computers were slow and the Deep Dream program is computationally intensive–and I ran them for days for a single image, weeks to build up a collection of them.
The source image is Zuckerberg in his college dorm, dressed up for the photograph in a new shirt, looking at his computer screen, which is showing some sort of network analysis tool. That network is the first version of Facebook. Back then it was a fun little toy.
The image is filled with bizarre sci-fi organic shapes and in some cases sci-fi organic creatures. Look at the bloated, rainbow-colored slug-like things on the lower left, beneath the computer monitor and at the four-legged bird-thing above it.
Zooming in on Zuckerberg’s face we can get a taste of the dog/lizard/insect universe we have created. Those beetle/spiders that are hanging in the air are creepy.
Here is a close-up of the computer screen crowded with shiny, glowing orbs of color. It looks like it might mean something to a race of space aliens.
Here’s another image, this one from Zuckerberg’s evasive testimony before Congress.
Here is a new nightmare-ish menagerie, and most nightmare-ish of all is not the monster in the middle of the frame but the one to the right.
Yikes. I hope you are reading this by the light of day.
What about Zuckerberg? You can usually judge a person’s character by looking into their eyes, there’s something profoundly revealing in the eyes.
Whatever this reveals, it is most disturbing.
One of the fun things about the Deep Dream program is the nefarious characters sometimes lurking in the background. Check out this guy keeping a low profile below Zuckerberg’s left collar:
I think he must be one of those Facebook trackers we hear so much about.
A few years ago Zuck–his friends call him Zuck so maybe we can, too?–had this idea that he needed to improve himself with yearly challenges. You know, wear a tie, read more books, that sort of thing. One of those challenges was that he would not become a vegetarian but instead would face up to his meat-eating instead of hiding from it in grocery store, pre-packaged abstraction. Any meat he ate for that year would come from an animal that he had killed himself, closing the moral circle.
I’m not quite clear on what meats he chose to eat–a year of lobster and snails or did he throw himself upon a three hundred pound pig, bringing his blade up under its chin and slitting its throat, perhaps while wearing a loincloth and beating his chest with his free hand? For chickens I expect he just used his hands and teeth, spinning out the feathers with that Zucky-boy grin.
A few years later, instead of aiming to improve his inner self, he thought to improve his outer self, the public perception of him as an aloof Cesar-in-waiting (the Roman haircut doesn’t help). He held a Facebook Live event and got together with his bros to grill some meats in the backyard. Just like the common man, you see?
All manner of fantastical beasts, grotesque distortions of those found in the real world, live upon his head.
Another shot in the dorm room! We love pictures from the dorm room. He’s just a regular guy, hoodies and sweats. His mommy called him The Prince when he was growing up but I think he may have developed greater ambitions since then.
Zuck didn’t graduate from Harvard but he did get an honorary degree there. The source image here is of Zuck wearing the crimson robes, on stage, smiling and pointing. He is said to be red-green color blind so I don’t know what color he thought the robes were.
And lastly, on our tour of the Zuckerberg Dreams images, we have Zuck yet again testifying before Congress. In the software there is a setting that makes the algorithm find eyes, and you can have hundreds of disembodied eyeballs filling the frame. I didn’t go that crazy here, limiting myself to a few doggie eyes on his face and that strange cyborgian growth coming out of his neck.
So Hironemous Bosch-y, so Giuseppe Arcimboldo-y, so Salvador Dali-y. This is no chimpanzee making “art,” no sad Thai elephant with its handler’s hand tugging its ear, “drawing” a picture. There’s nothing guiding this stuff, no man behind the curtain.