The power of Artificial Intelligence (AI) is becoming more and more impressive each month. Take, for example, the AI that can diagnose certain diseases better than any mere human doctor can. That’s all well and good, but while us meat bags enjoy our last few decades of being the dominant intelligent force on this planet, it’s always good fun to revel in AI screwing up.
In this grand tradition of AI fails, a blogging research scientist has been playing around with a neural network to see if AI can invent fake but convincing- sounding ailments, and the results are ridiculous and fairly unsettling.
Janelle Shane plugged in a list of 3,765 actual disease names into an open-source machine learning algorithm. This algorithm, known as a recursive neural network, processes all these real diseases and eventually learns how humans tend to name and structure the names of their ailments.
Since this is very much a learning process for the AI, the first run of generated names was a bit off:
Much Esophageal Eneetems
Poison Poison Tishues
Eventually, the machine began to hone its skills and the names became (vaguely) more real-sounding, albeit no less hilarious:
Cancer of Cancer
Bull Cancer Spisease
Some other top favorites, which Janelle suggests could be good for you budding sci-fi writers, include:
Strecting Dissection of the Breath
Bacterial Fradular Syndrome
Excessive Woot Sweating
Oh, AI, your days of world domination are still a while off. At least you tried, though, at least you tried.
Fans of the computer fluff ups may remember the InspiroBot. This worked on the same basic premise by using a bank of data to try and create inspiration posters, complete with an inspiring quote and a whimsical accompanying image. The huge majority of the posters were either totally hilarious or terrifyingly dark.
Sleep tight and don’t let the “RIP Syndrome” bite.