Deep Dreaming Through Neural Networks

A trip to the grocery store turns trippy as everything morphs into a dog. (Link to YouTube).

Such hallucinogenic images are created using artificial neural networks, computers which are set up to resemble the complex web of nerve cells in the brain. 



Google and Facebook have used these systems to recognize and classify objects and to recognize faces, but here they're being used to generate images.

They're able to mimic the human tendency for pareidolia and apophenia—the recognition of patterns in what we see, and in particular our hard-wired penchant for seeing faces faces in things.

As the computer recognizes faces, dogs, or other traits in the target image, it re-renders it to bring out that enhancement. The system can thus reinforce the kind of visualization we do when we're daydreaming. As Google wrote in a blog post: “This creates a feedback loop: if a cloud looks a little bit like a bird, the network will make it look more like a bird.”

Via Glitterbat — Google Deep Dreams Dolls

What we're seeing so far definitely goes beyond the Photoshop filter look. The images strikes me as a weird combination of humorous, compelling, obsessive, hideous, and disturbing. 

But the style of the images doesn't exactly resemble a the way my human brain generates novel images. It's definitely a computer's way of dreaming. 

But what we're seeing so far is just the tip of the iceberg, and in our lifetimes we'll be surprised by completely different styles of images, some that resemble what we think of as humanlike, and others completely novel.

Related GJ posts:

Image Parsing

Pareidolia and Apophenia

Research on Machine Perception

Related articles:

Google blog post on "Inceptionism" (by which they mean being in someone else's dream)

Wired: Google’s Artificial Brain Is Pumping Out Trippy—And Pricey—Art

Via BoingBoing