AI invented new paint colors. They’re Not Great.

Neural network lover Janelle Shane conducted an experiment, asking computer to generate some colors and names for them. It didn’t end well.

Ok, so there’s this scientist Janelle Shane, who’s running a pretty interesting blog. Although she has background in imaging and optics, she also enjoys tinkering with modern AI algorithms. Most recently she decided to use neural networks to invent new paint colors and give them attractive names. For this experiment, she gave the neural network a list of about 7,700 Sherwin-Williams paint colors along with their RGB values. The system was given the task to come up with new colors and appropriate names for them. Unfortunately, it didn’t really work out.

ArsTechnica reports that Janelle chose the algorithm called char-rnn, which predicts the next character in a sequence. Neural network was coming up with sequences of letters to form color names, and coming up with sequences of numbers that map to an RGB value. The algorithm was able to create colors long before it could actually name them. Unfortunately. the names were incredibly dreary and the colors themselves appeared incredibly bleak.

You can read more about this experiment in Janelle’s blog.