DW: So AI can be wrong? I am NOT growing a second head?
A Swedish medical researcher invented a completely fake eye condition called bixonimania to test whether AI chatbots could spot obvious medical misinformation, and the results were not encouraging. Almira Osmanovic Thunström of the University of Gothenburg and her team uploaded two made-up studies about the non-existent condition to a preprint server in 2024 and watched what happened. Within weeks, chatbots including Microsoft's Copilot, Google's Gemini, and OpenAI's ChatGPT were describing bixonimania as real and even diagnosing users with it based on vague symptoms.
The fake studies were loaded with red flags: the fictional author worked at a made-up university in a non-existent California city, the funding sources included the "Professor Sideshow Bob Foundation" and "the University of Fellowship of the Ring," and one paper literally stated, "this entire paper is made up." Despite all that, the condition even began showing up cited in real peer-reviewed papers written by actual human researchers.
Source: Oddity Central