Every time I hear the creation tales, I scoff a little. Who would bother making such ridiculous things as the blue-footed booby, or a Taonius squid? But maybe the creators were using image-to-image translation, bbroased on training an algorithm with something. Something that was… perhaps trained on 2,000 cat photos.
Using Google’s open-source machine-learning project called Tensorflow, Christopher Hesse created edges2cats, where you draw a design and it’s filled in with… cat. This doesn’t really work with my cat drawings, I notice:
I am not the only person making terrible cats, either:
“Some of the pictures look especially creepy, I think because it’s easier to notice when an animal looks wrong, especially around the eyes,” Hesse writes. “The auto-detected edges are not very good and in many cases didn’t detect the cat’s eyes, making it a bit worse for training the image translation model.”
So the attempted cats were horror monsters. Maybe some of that is problems with the machine learning algo, maybe some of that is that we’re not so good at drawing, especially edges. But we were definitely getting some creep-out vibes. I mean, haven’t you asked Siri where you should bury the bodies? So that’s when we decided to see what happened with… other inputs:
Anyway, it seems to me that machine learning is going to run very hard into human trolling, where people immediately try to see how far it will go, breaking it in interesting ways — much as one intrepid reporter has broken Madden. AI seems full of such promise for horror-beasts, and the apocalypse can always use more of those.