Google's artificial neuron networks, or ANNs, have been having some strange dreams - and have even learned to draw bizarre, beautiful pictures of them.
Google has been delivering the best information directly to your fingertips for years, but they’ve been working on some other interesting projects as well. According to a report from Popular Science, their artificial intelligence networks have been having some bizarre dreams.
Google’s artificial neural networks, or ANNs, are made up of layered artificial neurons that can process images on their servers. Programmers can teach an ANN to recognize a spoon by feeding it millions of spoon pictures. The ANN scans all of these images and remembers that they’re depicting spoons – the computer learns what a spoon looks like.
Every layer inside the network pulls a different aspect of the images’ characteristics, like the rounded edge at the tip of a spoon or the metallic sheen. After a while, it can recognize any spoon it comes across.
Then, something magical happened between the Google team and the artificial intelligence. Programmers wondered if the ANNs might be able to use their capabilities to make their own original images. After all, if you know what a spoon looks like, you can most likely make a picture one.
The computer is no Da Vinci. Some of the images it has produced are extremely bizarre – for instance, a picture of an exercise weight with an arm attached to it. Since the ANNs had seen so many people lifting weights in photos, it thought that they might have arms themselves.
But the ANNs got better. Eventually, the computers learned how to repeat patterns of images, creating beautiful, sprawling collages based off of one simple concept, like a bird. The Google engineers then removed all of the inputs and made the computer start with a blank canvas. Starting with nothing, the computer produced a series of psychedelic landscapes that are truly a sight to behold.
The resulting images, referred to as “dreams,” are completely a product of the computer’s inner workings. Most of them speak for themselves.