There are countless ailments already in existence that threaten to cause us pain and discomfort, or speed up our mortality – but, if it were up to one AI, our fate could be far worse.
A Colorado-based research scientist has trained a neural network to invent ‘serious-sounding diseases’ based on the names of nearly 4,000 common medical conditions.
While the AI managed to generate several ‘reasonably convincing’ afflictions, the experiment predominantly led to made-up diseases that sound not only far-fetched, but in some cases, likely to leave the sufferer in agony.
Scroll down to see the full list
Shane’s neural network spit out dozens of nonsensical medical conditions, ranging from comical (but perhaps uncomfortable) to extremely serious. The researcher fed the names 3,765 common conditions to the algorithm. Stock image
From Cancer of the Cancer to Eye Stools, the AI invented all sorts of bizarre conditions.
The idea was to use artificial intelligence to create plausible-sounding diseases that could work in science fiction narratives or medical dramas, research scientist Janelle Shane explains in a blog post about the experiment.
In the past, Shane has trained neural networks to come up with guinea pig names and Dungeons & Dragons spells, among other scenarios – with similarly hilarious outcomes.
The researcher fed the names 3,765 common conditions to an open-source machine learning algorithm called a recursive neural network.
This type of AI is able to learn to imitate its training data.
As a result, Shane’s neural network spit out dozens of nonsensical medical conditions, ranging from comical (but perhaps uncomfortable) to extremely serious.
You might be able to manage a case of the Sleepys, for example, and while embarrassing, it’s unlikely Seal Breath would kill you.
But, a diagnosis like ‘RIP Syndrome’ may not bode so well for your future.
‘Early on in the training, the neural network was producing what were identifiably diseases, but probably wouldn’t fly in a medical drama,’ she wrote.
Notably, this group includes the AI-generated afflictions: Insect Sculls, Poison Poison Tishues, and Sexurasoma Ear.
Or, the more vague, ‘Complex Disease.’
As the AI improved over time, it began to generate some conditions that sounded more plausible to the untrained ear.
‘Early on in the training, the neural network was producing what were identifiably diseases, but probably wouldn’t fly in a medical drama,’ she wrote. This group included the vague ‘Complex Disease.’ A stock image is pictured
Shane split the results into four groups, with those considered to be ‘reasonably convincing’ sorted into the second.
The other groups include the diseases that are essentially just generic words or body parts regurgitated by the AI, and those that sound ‘both highly implausible but also pretty darn serious.’
And, in the unpublished fourth group, Shane has sorted the ‘reproductive-related diseases,’ and ‘those that contain unprintable four-letter words.’
This category, according to Motherboard, includes ‘penis arthritis,’ ‘vaginal diffection,’ and ‘orgy fever.’