An AI was entrusted to create “attractive” new names for thousands of paint colours, replacing jobs currently held by these people. Instead of churning out fabulous, desirable monikers like “Tunsgate Green”, “Cinder Rose” and “Calamine” (which is a nice, dusty pink), it instead spat out a bunch of hilarious gibberish like “Gray Pubic” for a robin’s egg blue, and in some cases, unintelligible names like “Caae Brae 228 128 116”. Yikes.

Research scientist Janelle Shane had planned the experiment using an algorithm called “char-rnn”, which guesses the next character in a sequence. “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 wrote on her Tumblr.

Despite some tweaking, which did improve the machine’s ability to creatively name and recognize colours, it just didn’t work as initially hoped. Shane ended her experiment concluding, “1. The neural network really likes brown, beige, and grey; 2. The neural network has really, really bad ideas for paint names.”

This is what the AI came up with: