Google’s Artificial Brains Produce Trippy Artwork

Engineers at Google have been working with artificial neural networks which have now learned how to produce art, and even have dreams.

Google has released images of artwork that the machines have produced, and it is fairly creepy to look at. The neural networks are made up of between 10-30 stacks of artificial neurons, enough for them produce unique artwork, writes Quinten Plummer for Tech Times.

Google's Artificial Brains Produce Trippy Artwork

Artificial networks given a variety of tasks

The engineers constructed the artificial networks to function like a human brain, and as such the stacks can be separated into three layers differentiated by the complexity of the processing, which increases as information passes from the bottom stack to the top.

First of all the engineers asked the neural networks to recognize and classify images, but they later decided to invert the process and get them to produce their own images in order to improve their own understanding of how the neural networks learn.

“So here’s one surprise: neural networks that were trained to discriminate between different kinds of images have quite a bit of the information needed to generate images too,” says Google.

Google's Artificial Brains Produce Trippy Artwork

One experiment involved asking the artificial brains to visualize objects or animals when presented with unrelated pictures. The neural networks would then look for elements of the pictures that made them unique, before Google engineers emphasized these aspects by feeding the data through an information loop.

“If a cloud looks a little bit like a bird, the network will make it look more like a bird,” says Google. “This in turn will make the network recognize the bird even more strongly on the next pass and so forth, until a highly detailed bird appears, seemingly out of nowhere.”

Artificial brains dream and produce art

In a later experiment, the engineers allowed the neural networks to select which aspects of an image were important, before the information was also fed through an information loop in order to produce more and more complex results.

It was found that when the zoom was decreased on these over-interpreted images, the neural networks started to produce “an endless stream of new impressions,” a phenomenon which some have likened to dreaming.

“The techniques presented here help us understand and visualize how neural networks are able to carry out difficult classification tasks, improve network architecture, and check what the network has learned during training,” says Google.

As impressive as the technological advances are, it is the psychedelic images which stick in the mind.

Checkout! Michael Tyka’s gallery for more images.