Twilight Sparkle. Rainbow Dash. Pinkie Pie. If you have kids–or grew up in the 1980s or ’90s–you’re probably already familiar with the big names of the My Little Pony series, which has been around since 1983. But you’ve probably never heard of Big Blue, Barp Moon, and Princess Sweat.
That’s because these new potential characters were generated by a neural network trained by the scientist Janelle C. Shane. You might remember Shane from her previous experiments, which included training neural networks to generate everything from new Star Wars characters to fake craft beer names–that were often difficult to tell apart from their real counterparts. Shane is at it again, this time using the ridiculously cheery names from the children’s franchise My Little Pony to generate entirely new ponies.
Shane trained the neural net on a data set of 1,500 pony names from the My Little Pony Friendship Is Magic Wiki (yes, that is its full name). Perhaps due to the fact that real My Little Pony characters have names like Sunset Shimmer, Apple Bloom, and Diamond Tiara, some of the results are “plausible,” she notes. There are gems like Glowberry, Bright Seas, and Berry Spy. And the neural network also came up with some incredibly evil-sounding pony names: Dark Candy, Flint Sting, and Sunsrot. Generally speaking, the system works really well: It’s easy to imagine all of these names making it into the My Little Pony universe.
But the most hilarious results are the pony names that aren’t quite so realistic. Can you imagine Cheese Breeze, Pearlicket, Tracklewock Packin, or Mice Full trotting across the bright, friendship-filled fields of Equestria? I don’t think so.