The Artonomous Project

artonomous is an artificially intelligent machine that paints with a brush on canvas. While it can already paint portraits at human parity, photographer Kitty Simpson and A.I. Artist Pindar Van Arman have teamed up to train it on the subtleties of fine art. Kitty is sharing her evocative photography with the robot and Pindar is reprogramming its A.I. to learn as much as possible from her. The three are collaborating on a limited series of paintings as they attempt to hone the robot’s technique to create portraits with more emotional significance.

upatrobotwide_2000.jpg

Pindar and Kitty are instructing artonomous using traditional artistic methods.

Pindar and Kitty have it practice by painting large sets of representational portraits which are taken by Kitty Simpson with careful attention to both aesthetics and ethics.


"What artonomous will learn will be dependent on the input given by us. With that comes responsibility. The input is, for all practical purposes, “the truth” for the robot. We, the artists, are teaching the robot what is “beautiful”, “evocative”, what the characteristics of a “desirable” face are. Based on these inputs, the robot creates its own version of evocative portraiture, and this gets wired into its bias. It is easy to assume that our direct environment is a mini version of the world at large, but in reality, it is not. It is our own bubble. We tend to live around people similar to us, similar in race, socio-economic, religion, ability. As artists, we have a responsibility to train our AI responsibly, as we would raise a child. We need to feed it diversity, and not just our own limited view of beauty, to the best of our ability." 

Kitty Simpson

After each set, they train a neural network with the completed paintings, and have the robot autonomously paint an original composition using Feedback Loops and more than 24 creative AI algorithms, each competing for control of the brush.

Pindar and Kitty then critique the work and adjust artonomous’ algorithms before repeating the cycle on another set of representational portraits.


A more in depth look at their technical process can be seen below.


Each cycle begins with a set of curated photographs either taken or directed by Kitty. The robot then studies them and paints a representational portrait using its A.I. library.

kittytakingphoto_200.jpg
corrinnepaintnewpics.gif

Every several portraits, the robot reflects on both the final image and each individual brushstroke in an analysis that it uses to improve its neural networks.

training_anim3_b.gif

deepanim_blur_200.gif

artonomous then uses the improved neural networks to imagine and paint an original abstract portrait. (See this write up if you would like to go really deep into how its creative mind works.)


Kitty and Pindar then critique the finished painting looking for ways to improve its emotional impact. When they agree on an approach, Pindar implements it by making modifications to artonomous’ code, neural networks, and hardware.

coding_website_all.jpg

Repeat…

The trio then repeat these steps in a creative feedback loop to paint another set of portraits. Each cycle refines artonomous’ process bringing it closer to complete creative autonomy, and its work closer to Fine Art.