Google Deep Dream
Digital hallucinations and formless beasts
In July 2015, Google released Deep Dream, a piece of software that uses a neural network to find and enhance patterns in images. The network, ‘trained’ on Google’s vast storehouse of image-search results, tends to interpret shapes as animals, turning nearly any image into a hallucinatory, formless array of eyeballs, muzzles and beaks. The first thing the internet used it on was porn. When the handyman came, he knocked twice, hard, without hesitation. She kicked loose screws out from under the sink. He rang, leaning into her strident doorbell, and she opened only then, feigning having travelled some distance from a calamity across the apartment. He stood in the doorway in shirtsleeves, smelling of turpentine and musk. She shrugged, wrench in hand, unscrewed. Pointed at the water heater: a sound, an ominous clang. He peeled himself from the door jamb and poured into the hall, clocking her décolletage. He dropped his tool kit and then to his knees in the shifting paisley of the carpet. The muscles of his back ebbed as he worked. Standing behind him, she unbuttoned her blouse, hands softening into fleshy paws as they moved across her body, revealing a pair of natural, bouncing Akita puppies. She kicked off her shoes and tip-toed towards him on delicate beaks. She dropped the wrench and the premise of the water heater, never broken, having never once clanged. He turned his melting profile to her as she joined him on the floor and enveloped him in the purple curves of her thighs. He found himself hard and glistening as a scarab. Her every darkness was a spiralling eye, a panting mouth. He tore his shirt; the plane of his chest was rippled with rounded spirals of multicoloured muscle, each ghosting like a double-exposure. Her face fanned into the carpet in bestial ecstasy as the puppies whorled with each thrust of his throbbing, dog-nosed proboscis. They fucked in a feathery fractal in the time it took to render.
First published in Issue 176