On this blog we discuss, programming, art, and anything I find interesting on Tuesday and dare to write about on Wednesday. This publication used to be called Generation Machine. Why is it here now? I dunno. Why did Gandalf the grey become white? Why is it that c = 299,792,458 meters/second? Why is the north wind cold and bitter? I don't have the answers. You are probably getting this because I wrote a post about Perlin noise archipelagos, and you found that interesting and compelling enough to drop your email in a box. There will be more interesting work like that. Thank you, kind soul, for your vote of confidence.
Thank you for Sharing.
No, not the ones the CIA keeps hinting at on cable news.
My recent online coursework in robotics got my little monkey mind stewing about the intersection between robotics and art. Humans have always used tools and primitive machines to make objects. Training machines to create, however, is different from the vast history of human crafting. We imagine that an intangible spark of consciousness undergirds the creation of art. Many have explained to me over the years that robots don't have this spark and that they cannot create. That's wrong. Robots can create.
Every year, the RobotArt competition hosts a smattering of metallic arms, algorithms, and glorified printers that create new art. Here’s reArt (an arm-based painting robot), the third-place winner, which painted a reproduction of Wheat Field with Cypresses by Vincent van Gogh.
Every reproduction in RobotArt differs from the original in some respect. Little touches get added as the paint's consistency, the texture of the canvas, and climate alter the constraints each robot feels. This creates adaptation and originality in the works.
There is original art in RobotArt too, which usually is a result of combinations of algorithms that have been tweaked to produce abstracts. This piece, titled Miserable, was made by measuring human theta, alpha, and beta brainwaves (presumably unhappy ones).
When a human paints something new, the inception of that art comes from inside the mind (well, mostly). This piece is a robot's original interpretation of that mind.
Ok, but what about something completely novel? Something not based on another piece, photo, or my own brainwaves?
For where is the artist without the critic?
I'm glad you asked about this kind reader. A Generative Adversarial Network (GAN) is a way for a machine to learn how to create original art. A GAN is a system of two dueling neural networks competing against each other. One network generates pieces of art, and the other network criticizes each developed work, comparing it to real artworks and providing feedback to the generator. In the end, we hope the generator can create novel, realistic pieces. I ran my own experiment with GANs generating monsters by teaching neural nets about pokemon. I had to do a lot of tweaking to get the generator to create unique monsters that don't look like rainbow blobs. What I will say about each generated output (even the blobs) is that they were original. A human could conceive the spark of life somewhere inside it.
Here are some examples I picked from my GAN experiment.
Ok, if you want to see some exquisite GAN art, you should check out GANGogh. It was trained on famous artworks from the masters and produces incredible pieces like these paintings of flowers.
GANs learn features about flowers from actual paintings. These features can be things like petal shape, texture, stem length, or vase orientation. Then it takes what it learned and creates new versions of those features to generate novel art. If you were a human, this is precisely what you might do, observe existing features of flowers, try and create an archetype in your mind of those features, and then paint them.
While the internal functioning of a Robot differs from a human brain, the goal is the same. Both feel feedback from the environment and adapt the art. Both can understand aspects of their subjects and try to create novel art from them. Both are trying to express what we can see as the spark of consciousness. That is to say, when they kick toward the universe, the verse kicks back, and we get art.
I hope you enjoyed this dive into the verse. If you like it, share it with a like-minded friend.
If you're interested in the robotics course I was doing, you can check out modern robotics on Coursera.
Art produced by GANs cannot be distinguished by human critics in many instances. Check out this paper (I recommend scrolling through to see the art).