Janelle Shane, a research scientist with a penchant for silliness, decided to train a neural network to generate new paint colors, complete with appropriate names. The results are hilariously bad, and range from the impenetrable (“Dondarf” for a lovely shade of cornflower blue) to the almost-there (“Ghasty Pink”, the color of Pepto-Bismol). Ars Technica has the details on how she did the training.
Her conclusions: “1. The neural network really likes brown, beige, and grey; 2. The neural network has really, really bad ideas for paint names.” Professional namers need not worry about losing business to AI just yet.
(She has also done a similar experiment with superhero names, and it went about as well as you would expect.)