by Pindar Van Arman
We are working to make a generative art system that achieves human parity. If this A.I. system is successful, it will force us to consider the possibility that all art is generative.
Kitty Simpson and I, Pindar Van Arman, are artists teaching our creative processes to a machine. The machine is artonomous, an artificially creative painting robot. Kitty is teaching it how to take photographic portraits and I am teaching it how to paint them. The goal of our project is to teach it everything we know then let it begin to create art completely autonomously.
artonomous is fifteen years in the making and has already completed thousands of “hand-painted” canvases. It has a paint brush in its gripper, a camera on its arm, and a collection of more than two dozen A.I. algorithms fighting for control of the next mark it makes. Some of its algorithms are procedural, others are feedback loops, and many are neural networks attempting to imitate how the human brain creates. While each of these algorithms has an important role in completing its artwork, we will be concentrating on improving the neural networks responsible for its creativity and imagination.
Our approach will follow traditional artistic training using vigorous practice and reflection.
1: Kitty begins the process by providing the robot with a curated set of portrait photographs.
2: artonomous then paints a study of the portraits using its array of A.I.
3: Once the painting is complete, the robot analyzes each brush stroke and final painting by incorporating it as training data for its neural networks.
4: Every eighth study, artonomous attempts to create a unique portrait from its imagination.
5: We critique the imagined portrait and decide on how to improve artonomous’ creative process.
6: I then make adjustments to artonomous’ hardware, code, and neural networks to better actualize it as an artist.
The process then begins anew with another set of photo shoots, in a creative feedback loop.
We do not know how long it will take us to improve artonomous’ creative process to the point it achieves human parity. We expect it will take several years and hundreds if not thousands of cycles. Part of this project is recording the progress being made, how its is being made, and exactly what improvements, if any, are occuring.
As part of this record, we will be tokenizing the imagined portrait from each cycle on superrare. Expect new tokenized artwork every 8-10 days as this is typically how long each learning cycle takes. The form of these artworks will vary. Some will be animated gifs of the neural network thinking, others will be still frames of the finished painting, and some will be timelapses. Regardless of form, the one thing each will have in common is that it will be based on the original artwork created by artonomous in that cycle of its learning.
Look for the first tokenized artwork by artonomous titled “x00000000“. It is the first of what we believe will be an exciting and interesting exploration of the creative process and the emergence of an artificial artist.