Creativity, in this discussion, is the act of creating (generating) something new. In this case, as in most creative processes, is made up of everything stored in the memory bank. All the known (old) images coming together to form a new one. Blaise Agüera y Arcas, principal scientist at Google, has taken this concept to the world of computers and made A.I. that can produce artwork.
The story begins a few years ago when a way for computers to perceive things was invented. From this moment on it became possible to take a photo of something and the computer would tell you what it is. Whether they knew it or not, this was the first step towards the artist-AI.
How this works is similar to the way a humans mind works; something they call machine intelligence – the engineering discipline of making computers and devices have the ability to do some of the things that brains do. If you could see the way a human brain (cortex) works when it processes imagery that comes from the eye, you’d notice it looks exactly like a circuit diagram. In other words, the visual cortex works like a series of computational elements that pass information from one to the next in a cascade.
This brings us to perception. Perception is the process by which things out there in the world — sounds and images — can turn into concepts in the mind. The inverse of perception is creativity; taking a concept and putting an image or sound out into the world. “Perception and creativity are very intimately connected,” Agüera y Arcas says. “Any creature, any being that is able to do perceptual acts is also able to create.”
When an image in the computer is perceived, what it really is is a grid of pixels. You can think of the pixels like neurons. This data feeds into one layer after another of synapses that, when narrowed down, result in a connection to the answer. The answer is the concept. There could essentially be many of them. If this process is reversed, you have creativity.
What if these deep neural networks for machine perception that are trained to recognize images were run in reverse? That is what Alex Mordvintsev from the Google team was wondering when he decided to experiment and see what would happen if they had the computer find the image from the answer. It worked, but since there are many answers, what the computer came up with was ambiguous…a morph of all the answers.
Then they tried another experiment. This one is very interesting. Instead of starting with a blank canvas (no image, or white noise) they tried beginning with an image, then asking the computer to find the answer within the image. For example, having a picture of clouds and asking the computer to find images within the clouds.
What they have managed to create are neural networks that are entirely trained to recognize different things in the world, then able to be run in reverse and generate something new. Decades ago, computing began as an exercise in designing intelligent machinery modeled after our minds. Now, this machinery is capable of processing information the way our minds do, and then even create something with the information. In the not too distant future, a computer will be an extension to our minds.