Google’s Deep Dream neural network watched Fear and Loathing, promptly trips balls
There is an increasingly good chance that Google will eventually doom humanity through its experiments in artificial intelligence, but in the meantime it’s content with having some fun with it.
Google recently unveiled its new neural network (nicknamed “Deep Dream,”), a program designed to recognize and interpret images using 30 layers of artificial neurons. The first few layers are for the basics, things like identifying the shape of an object. As the image goes through the layers, the program goes into more detail.
The Google researchers discovered that with a few additional commands though, the AI could try to find images within the images. For example, it could be shown an image of clouds and asked to identify other objects. The program would then analyze the photo and create animal images, for instance, based on lines that somewhat resembled a specific animal.
It isn’t that far removed from when we look at clouds and see shapes, but the neural interface could then show that image to others. Google then set the program into a feedback loop, and the results are images that look like surrealist art.
Google named the process “inceptionalism,” but there it gave it another name too: dreaming. So now you have a program that can essentially daydream.
Google went ahead and released the program to the public, and so naturally people began to feed the neural interface some bizarre stuff, possibly in the hopes of either raising a Cthulu-like monster in a digital landscape, or just to see how close to insane they can get.
One of the most bizarre and interesting results came after a user fed the interface a short section of Fear and Loathing in Las Vegas, specifically the scene in which Thompson (Johnny Depp) reminisces about tripping in a club in San Francisco.
The results were appropriately weird. Apparently, the program has a thing for dogs.
Doctor Thompson would be impressed. Probably so much so that he would use one of his good guns to shoot at the program.
Check it out.