clock menu more-arrow no yes mobile

Filed under:

Google Trained Computers to Make Their Own Trippy Art


Peter Kafka covers media and technology, and their intersection, at Vox. Many of his stories can be found in his Kafka on Media newsletter, and he also hosts the Recode Media podcast.

Remember when you had time to lie on a hill, gaze at the sky and imagine you could see shapes in the clouds?

Well, Google has trained computers to do that.

It has trained artificial neural networks, to be precise. So they can turn this image:

Into this one (you’ll want to click on all of these to get an expanded view):

It doesn’t have to be a sky. Google’s researchers working on image recognition have trained their computers to do it with any image. So this:

Becomes this:

Or this:

What Google is doing here is essentially reversing image recognition, and telling its computers to use the images they already know to augment new images. As Singularity Hub (via Engadget) explains: “Where the software was allowed to ‘free associate’ and then forced into feedback loops to reinforce these associations — it found images and patterns (often mash-ups of things it had already seen) where none existed previously.”

Google says it’s doing this to help it visualize how neural networks work. Its researchers are also well aware that this is full-on, lava lamp and bong water, “have you every really looked at your hands, man?” stuff. Which is why its blog post talking about all of this is titled “Inceptionism.”

There’s a gallery full of this stuff, and you should definitely check it out when you’re in the right time and place. Here are a few more images to let you know what you’re in for:

This article originally appeared on

Sign up for the newsletter Today, Explained

Understand the world with a daily explainer plus the most compelling stories of the day.