Artificial intelligence can do amazing things, but never let it mix and name new paint colors.
Not that you were going to do that. Research scientist Janelle Shane, on the other hand, decided to go for it. She gave a neural network a list of about 7,700 Sherwin-Williams paint colors, along with their red, green and blue color values.
"Could the neural network learn to invent new paint colors and give them attractive names?" Shane asked on her site.
The answer is a big fat "don't bet your not-so-artificial-life on it." The AI could make colors, all right, but at least at first, they were weird melds of the grayish-green you'd use to paint your walls only if there was literally no other color on Earth available, or you really, really wanted to live feeling like you were surrounded by mold.
The names, though, were where the AI really fell off the cliff. It began with randomness such as "Saae Ble," and then progressed to Klingon-looking nonsense like "Rererete Green" and "Gorlpateehecd." But when it finally advanced to what Shane admitted was "about as well-trained as it's going to be," it was still inventing names that sounded more like losing Scrabble hands, or The Onion's parody of a box of Crayolas.
"Ghasty Pink"? "Rose Hork"? "Snowbonk"? These sound like characters from a "Game of Thrones" parody novel.
"Dorkwood?" "Sindis Poop"? "Stanky Bean?" Did we just wander into a game of Mad Libs?
"Stoner Blue?" "Turdly?" Oh, now someone's just letting the 11-year-old start making things up.
Shane summed it up nicely in the end, noting:
"In fact, looking at the neural network's output as a whole, it is evident that:
- The neural network really likes brown, beige and grey.
- The neural network has really really bad ideas for paint names."
(Via Ars Technica)
Logging Out: A look at death in the digital age.
Star Wars at 40: Join us in celebrating the many ways the sci-fi saga has impacted our lives.