Google's Deep Dream for Dummies

This article appears in the August Issue of VICE Magazine Fat and fluid, with lazy curves like unrefrigerated sausages, their bodies pass in and out of one another like wet cotton balls in a plastic sack. They're definitely alive, but a spark lies behind their palm-size glass eyes. They're most like the animals you see in clouds.

Meet the Puppy Slug, the internet's newest cyborganic species: the dog of Google's Deep Dream, or the result of a thing called an artificial neural network over-interpreting an image. It's image-classification software being given a picture or video file and then being asked to re-create it from what that network knows images like it look like—a.k.a. "Inceptionism."

Unleashed on the public on July 1 via the Google Research Blog, the creatures are the result of a process that involves running a deep-learning framework called Caffe through a scientific Python-distribution software such as Anaconda. If it sounds complicated, it is, but given its open sourcing, #DeepDreaming is already as easy as a "click to upload" online tool.

It's not the first time an artificial neural network has been used to generate images, but it's certainly the most vivid. In this case, as Google's visualization tool was "trained mostly on images of animals," from porn GIFs to the Mona Lisa, many of the results are puppy-headed slugs.

Up until now, we've witnessed many a species's slow and steady march toward extinction. Perhaps this is the way the internet—arguably the last final frontier—responds to the corporeal: As we dive deeper into the web, we'd be remiss not to expect encounters with new kinds of wild animals. The head of a doge, say, on the amorphous body of a troll. They're not a newborn litter but a pack young enough to still look excited about sliding out of the techo-surf on their bellies to say hello, shit all over your campsite, and then bite your ankles.