AI and scientists fell for a humorous study – details of the blunder

A Swedish researcher created a non-existent disease called "bixomania," leaving obvious clues in the text about its fictitiousness. Top neural networks, including Copilot, Gemini, and ChatGPT, accepted the fake as a scientific fact. Indian scientists cited the parody in a peer-reviewed journal.

A Swedish researcher conducted an experiment by creating a non-existent disease called "bixomania," supposedly developing from long hours of work at a monitor and manifesting as reddening of the eyelids. Obvious clues were left in the text: the acknowledgments thanked a professor from the Starfleet Academy of Star Trek, the sponsor was the fictional Sideshow Bob Foundation from The Simpsons, and the methodology mentioned the participation of "50 fictional people." Despite this, AI systems, including Copilot, Gemini, Perplexity, and ChatGPT, took the fake seriously. Perplexity even generated non-existent statistics. Indian researchers cited the parody in the peer-reviewed journal Cureus, exposing problems with content filtering.