NSFW: Gaze Upon the Horrid Melting Flesh of Neural Network Porn

This week, we’re pleased to bring you a different version of Futurism, containing stories from the horizon of hedonism. Welcome to The Science of Pleasure. In collaboration with our friends over at MEL Magazine, this week, we’ll be bringing you stories from both publications about the pleasures of tomorrow, today. 

It’s a nonstop cacophony of pure flesh. Breasts melt into arms, twist and darken into matted hair, stretch into a beautiful face — or, wait, was that the yawning grimace of a misshapen skull? — before liquifying into a beak, the eyes impossible pits, the entire dreadful visage writhing and mutating in an undifferentiated, ever-dissolving barrage of vaguely feminine buttocks, swimsuits, midriffs, stumps, and blobs of flesh.

This horrific, provocative spectacle is the work of digital artist Eric “Shardcore” Drass, who produced this particular video — “Looking Glass” (2019) — as well as a greater body of work he calls “The Machine Gaze,” using machine learning tools.

“By turning the AI on the domain of porn you make imagery that's not pornography but it's fleshy — you can see the roots that it's come from,” Drass told Futurism. “It starts to trigger the similar parts of your brain but in slightly the wrong way.” 

Drass isn’t exaggerating. The material is uncomfortable and often grotesque. It forces us to ask uneasy questions about our own thirstiness.

“The overriding feature is, of course, flesh — and the melting and conjoining of fleshy forms is perhaps the very definition of pornography, is it not? Yet machines have no genitals or bodies,” Drass said. “So what do these images mean?”

From his studio in a boho bastion of Brighton on the English south coast, Drass says he’s spent the last several years training models, built with machine learning tools including Nvidia’s StyleGAN and Google’s DeepDream, using a vast array of imagery gleaned from vintage erotica and PornHub.

“The underlying question,” he said, “is what do machines — with no concept of sexuality or sexual behaviour — see when they look at this content?” 

If his AI images and videos are anything to go by, the answer is unsettling and at times ghoulish.  

The videos, like sex on an acid trip, veer between titillating and vaguely nauseating, with everything happening too fast to necessarily get a grip on the experience. Drass layers his own musical compositions over the footage, like a dreamy porn riff playing over a dark parody of a striptease, or a searing drone soundtrack on a Picassoesque sexual nightmare. 

Perhaps the strangest is a video that seems to be suggesting a sexual encounter between numerous figures of ghastly, indeterminate skin. They’re broken up visually with raw negative space, in which they engage in a morbid pantomime of rubbing, kissing, sucking and fucking in which their bodies dissolve and fuse into one another in an endless venereal hallucination.

The overall effect is to make you wonder if you might be going insane.

If anything, Drass’s still images are even more intriguing, because you can stop to behold them as a complete work without watching them constantly liquify into new representations of something so strange that it’s difficult to put into words.

The results can be haunting, spectral, or just plain disturbing. Drass has a knack for catching the faces he generates at the most appallingly liminal moments — awful masks of madness that seem to form as these dreamscape beings warp between forms. The bodies, if anything, are even more macabre — yearning neural succubi which, frankly, make us very uncomfortable.  

Is any of it hot? It’s difficult to say, and that ambiguity is the core of the work.

Sometimes Drass leaves his algorithms churning all night, he said, sifting through the results the next day. 

Of course, that approach raises the question of whether he might be essentially generating vast numbers of mostly-uninteresting permutations on a concept, and then using his all-human sense of aesthetic taste to select the most evocative. In other words, his algorithms could be the modern equivalent of Clever Hans, a performing horse in the late 1800s which was renowned for its ability to spell and do arithmetic — until it was discovered that the horse had been picking up on what were likely unintentional cues from its handler.

Drass acknowledged that concern, saying that it “feels like a bit of a trick when I've got a machine that can make an infinite number of images.”

“I need to ask 'what am I contributing'?” he mused.

Drass, who says he’s swerved between an unfinished PhD in cognitive psycholinguistics, modeling, software, music, and art, started to conceptualize the project as he thought about porn — or, rather, pondered how porn might be perceived by a non-human intelligence. 

“A huge proportion of internet traffic is porn but people don't talk about it,” he said. “But what about the machines that are observing our data, observing our behavior? To a machine watching the internet, it will see 30 percent of it is these monkeys looking at other naked monkeys. It's fucking crazy! But the machines are never asked to look at that — they're just asked to look at your banking behaviour.”

It’s also difficult to separate the topic of neural network-generated pornography from the phenomenon of “deepfakes,” which often use similar tech to insert the likenesses of celebrities or other people into existing porn, almost always without their consent.

Basically, deepfake porn is a nasty phenomenon that uses cutting edge AI tech to target and harass women. For his part, Drass says that he sees it as a waste of extraordinary new creative tools. 

“The vast majority of deepfake work involves transferring the face of a Hollywood star onto the body of a porn actress,” he said. “This is not just morally objectionable, but also shows a startling lack of imagination.”

He’s also interested in using machine learning to probe the way we might already be observed by AI in our online lives.

“People are up in arms about the whole hidden algorithmic culture that's going on, but they don't ever think about porn — something I'm doing furtively with myself,” he said. “Because it's so base and animalistic an activity, it's somehow not part of the greater landscape of the web.”

Illustrating that point, Drass posted a promo video about “The Machine Gaze” on various social platforms to see if their algorithms would shut it down. 

“LinkedIn was the only place that didn't block it!” he said with a laugh. “Facebook closed it straight away, Twitter was later when it got flagged, YouTube would have none of it. But there is a weird tension, because this stuff is not naked people, just what machines are judging to be naked humans. So who deems it filthy? It's just a picture. That's fascinating, because the machine doesn't know anything.”

After spending all this time training neural networks to produce a simulacrum of human desire, you might wonder whether Drass believes machines will ever feel pleasure themselves. But he demures, shaking his head. 

“Of course that's an interesting philosophical question,” he said. “But pleasure is about being in a body. The fact that we're made out of biology and have a direct relationship with the world, machines cannot currently grasp.”

More on neural networks: Physicist: The Entire Universe Might Be a Neural Network