Wed09202017

LAST_UPDATEWed, 20 Sep 2017 8pm

An AI Invented A Bunch Of New Paint Colors That Are Hilariously Wrong

At some point, we've all wondered about the incredibly strange names for paint colors. Research scientist and neural network goofball Janelle Shane took the wondering a step further. Shane decided to train a neural network to generate new paint colors, complete with appropriate names. The results are possibly the greatest work of artificial intelligence I've seen to date.

Writes Shane on her Tumblr, "For this experiment, I gave the neural network a list of about 7,700 Sherwin-Williams paint colors along with their RGB values. (RGB = red, green, and blue color values.) Could the neural network learn to invent new paint colors and give them attractive names?"

Shane notes that, at first, the algorithm seemed to be forming words that are a combination of brown, blue, and gray.

Shane told Ars that she chose a neural network algorithm called char-rnn, which predicts the next character in a sequence. So basically the algorithm was working on two tasks: coming up with sequences of letters to form color names, and coming up with sequences of numbers that map to an RGB value. As she checked in on the algorithm's progress, she found that it was able to create colors long before it could actually name them reliably.

The longer it processed the dataset, the closer the algorithm got to making legit color names, though they were still mostly surreal: "Soreer Gray" is a kind of greenish color, and "Sane Green" is a purplish blue. When Shane cranked up "creativity" on the algorithm's output, it gave her a violet color called "Dondarf" and a kelly green called "Bylfgoam Glosd." After churning through several more iterations of this process, Shane was able to get the algorithm to recognize some basic colors like red and gray, "though not reliably," because she also gets a sky blue called "Gray Pubic" and a dark green called "Stoomy Brown."

In the end, she concludes: "1. The neural network really likes brown, beige, and grey; 2. The neural network has really, really bad ideas for paint names." Possibly the neural network needed better parameters, but really, who can argue with results like these?

I cannot wait to live in a world painted by robots. Thanks to Shane's work, we are one step closer to knowing what that will be like.

-Arstechnica