The great Philip Kindred Dick posed the question in the 1960s with the title he chose for one of his most famous novels. Dreaming, loving, and hating are natural for human beings, but the big question is whether these will one day extend to a sufficiently advanced artificial intelligence. To that short list we can also add artistic expression, and on that point Google offers an answer through its neural networks, which were given free rein to “create images.”
A woman loses her loved one, turns her pain into inspiration, and writes a song. A man with strange visions of monsters and disturbing places decides to put those images on a canvas. An elderly man from a remote tribe finds a piece of wood and sees in it a statuette that he begins to “reveal” with his knife. These are classic examples, but they serve to prove that art reaches us through different paths. Now, is art exclusive to human beings? Although there are several references to animals painting and drawing, what I want to ask is whether at some point an artificial intelligence will be able to make art on its own.
How Google's Neural Networks Work
As we have known for some time, Google uses neural networks for its image classification and voice recognition platforms. A network has between ten and thirty layers of artificial neurons, and the content to be analyzed and interpreted passes through each of them until it reaches the last response layer, which presents the result. This time, Google modified the process so that the neural network generate images from the material it already has at its disposal.
The first requests were relatively simple: bananas, screws, parachutes, and starfish were some of the things the neural network “created” by manipulating images. But another much more interesting aspect is that the network can not only work with noise, but also with more complex images. In those cases, it was allowed to optimize and amplify what it found, giving rise to very strange results, such as a sky full of animals, a knight with a hippopotamus face and his right hand replaced with a fish, or entire structures hidden in trees.
The Dreams of a Neural Network
Google used a very powerful word to describe the images from its neural network: Dreams. Each image was generated based on the network's previous experiences and available knowledge. Of course, no love, pain, hate, loneliness or anything like it here... but each of those neural layers saw something that none of us could see. Perhaps some will dismiss these images as the erratic product “of a machine”, however, who hasn't seen a cloud shaped like an animal...?
