To teach the neural network how to create new shade-and-name combinations, the research scientist fed it 7,700 Sherwin-Williams paint colors along with their RGB values. Its first iteration created a bunch of dull, nearly identical shades — cranking up the creativity setting boosted the colors’ variety, but it paired them up with a bunch of gibberish, such as “Bylfgoam” and “Dondarf.”
The final iteration was much better and can recognize some colors, including white, red and grey. Like the neural network Shane trained to churn out pick-up lines, it’s still no Shakespeare, but it did come up with some amusing, slightly risqué names like “Gray Pubic,” “Bank Butt,” “Sindis Poop,” “Stanky Bean” and “Turdly.”
That said, even if you can think of something better than “Stoner Blue” and “Stoomy Brown,” you may still want to think of an alternative career for a world after the robot apocalypse. Shane admits that it could’ve been done better using different parameters, something another individual or a tech titan working on AI like IBM could do in the future.
