So to put it briefly google taught an AI to look for recognisable stuff within pictures. Fed it a whole lot of stuff in the form of animals, people, faces, cars, etc so that it might find them in patterns. Some tweaks of the code let it loop over and over, etc.
Essentially you can give it a starter image, or random fuzziness, and let it dream up whatever it can.
You can read more here. Worth a look.
ai.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html
The results are pretty stunning. Not just because it's an AI actually "dreaming up" images, but also some of them are just stunning outright.
(some of them are also really trippy, or pretty dark, or as close to acid visuals as you'lll ever find..)
I thought I'd share some.
This looks way interesting. Will be looking at this weekend and feeding in some, er, interesting images.
Select to expand quoteGavGav said..
I was about to say that it's a mandelbrot set with Johnny Depp in it 
My first implimentation of this was using Turbo Pascal on a x286 IBM AT PC at Uni. As a joke with a fellow student, I wrote one in COBOL and ran it on a VAX II minicomputer over night printing out the set with "*" characters on a 132 character carriage dot matrix printer. "Back in the day" 
(Hey, I was a bored and broke !)
