With April Fools' Day upon us, an artificial intelligence (AI) expert sought to determine if AI could create some uproarious, sidesplitting pranks.

Computer researcher Janelle Shane tasked a language-modeling software to do gags humans would normally pull on each other.

The neural net came up with such hilarious ideas sawing a mattress in half, or standing in queue for a film for around 30 minutes, and just walk away. Another suggestion was generating a language only you and your cat can comprehend.

ALSO READ: Artificial Intelligence Can Map How a Brain's Complex Network Works to Understand Sentences


An alternative was taking a bunch of jellybeans, place them in a Zip-Lock bag, and place it around your neck.

"AI Weirdness" Delivers Funny, Nonsense Pranks

Among the suggestions from the neural net came the most dangerous yet still funny-pouring hot coffee in your lap. It quipped that the prank was an "oldie but still a favorite."

Shane, who works as an optics research scientist at Boulder Nonlinear Systems, finds delight in toying with what she calls "AI weirdness," or experiments with the technology that yields riotous results.

(Photo: Possessed Photography on Unsplash )

She also tasked the AI program to do the funniest outputs, such as create pick-up lines or knock-knock jokes. In time for April Fool's Day, Shane pushed the technology to create pranks.

Those AI-created pranks, she said, are amusingly similar to those people do on themselves. Due to social distancing, Shane thought it was the best time to come up with some DIY April Fools ideas.

Shane used Generative Pre-trained Transformer-3 or GPT-3, a complex algorithm that utilizes deep learning to generate human-like text.

Four Variations of GPT-3

Part of GPT-3's Internet training had a lot of April Fools' gags. As such, Shane provided four variations of the AI program a new list to fulfill.

DaVinci, the most powerful version of the GPT-3, recommended a language you and your cat could only understand. Other pranks include keeping car keys in the freezer and forgetting where you placed them, and queuing up at a cinema for half an hour, and just leave.

If you do not consider them as good pranks, DaVinci suggested a gag that might take the cake: Dedicating a day to clean your mess and go out for a run, and not do them at all.

As the program gives out more pranks, DaVinci might have not produced what is a really good representation of pandemic quarantine life. This is because DaVinci's Internet training ceased in October 2019.

Curie, which is considered the second-largest GPT-3 variant, meanwhile proposed to install a fake microwave that you will never use, or take a ZipLock of jellybeans and tie it around your neck. It then came up with a clincher: Setting up a freezer within a freezer and then place a sign, "freezer burn."

The prank suggestions only got worse, with Babbage, a less-powerful GPT-3 variant telling Shane to eat a banana inside her mouth or put her name on a balloon for a surprise. But Babbage gave the most dangerous suggestion of pouring hot coffee on your lap, which it said was an "oldie but a favorite."

Meanwhile, Ada, the least powerful among the GPT-3 variants seemed not to understand what a prank meant, saying that its cat slept through a Florida blizzard in December this year.

Ada's best suggestion of an April Fools prank was walking into a toy store to buy bathmats, an inflatable pill bottle, and a video game, only if you are a cat.

AI is guided to repeat patterns but so much of the laughs comes from tweaking those patterns in ways that are clever and unexpected.

AI Finds Difficulty to Grasp Creative Humor

Creative humor, experts note, is among the most difficult areas for AI to understand. While humans cracking a joke requires real-world knowledge with background information and common sense, AI only processes what you tell it and what it draws from. Previous tries had also been downright funny.

AI cannot do any better with romance, as Shane tasked DaVinci to come up with pick-up lines. The reader, it said, had a "lovely face" and then requested to "put on an air freshener" saying it wanted to keep that scent close to it always.

RELATED ARTICLE: OpenAI Deceived? Handwritten Notes Fool Computer Vision in Typographic Attack


Check out more news and information on Artificial Intelligence on Science Times