When an AI Invents Diseases, You End Up With 'Penis Arthritis'
Research scientist Janelle Shane trained a neural network to make up fake diseases.
Scary-sounding, futuristic diseases are a time-honored trope of science fiction. But if you're short on ideas, a friendly neural network can help you invent a condition that's adequately unnerving—as long as you give it enough time to learn what sounds disease-y.
Janelle Shane, a research scientist and blogger, recently worked with one of her readers to train a neural network to come up with fake ailments. First, her reader Kate Leroux, a cartographer, scraped 3,765 common names for conditions from an online medical database. Then, Shane fed those into an open-source machine-learning algorithm. The names the algorithm first generated weren't very realistic-sounding:
But after a few rounds of training, the algorithm started to get smarter, and learned the difference between disease names, body parts, and symptoms. Eventually, it came up with some pretty convincing, sci-fi-sounding conditions:
It also had a bit of a predilection for reproductive health conditions, and concocted diseases like "penis arthritis," "vaginal diffection," and "orgy fever." (Shane sent these particular examples to me via email.)
As often happens with neural networks, the project wasn't a slam dunk. I don't think the writers of the new Star Trek series need to be worried about their jobs. But there's something earnestly delightful about tasking computers with creative tasks like this. Machines, even learning machines, approach them in such a different way than a human would, and it's sweet when a computer says "like this? Did I do good?"
Sure, computer. Let's pretend Cancer of the Cancer is a thing.
Get six of our favorite Motherboard stories every day by signing up for our newsletter.