STUDY: ROBOT DOCTORS WRONG HALF THE TIME, STILL ACCEPTING PATIENTS

What actually happened: BMJ Open study finds about half of AI chatbot answers to health questions are problematic

IDIOCRACY LEVEL: 7/10

Researchers from the US, Canada and UK asked ChatGPT, Gemini, Meta AI, Grok and DeepSeek ten questions in each of five health categories. About half the answers were rated problematic and nearly 20% highly problematic, worst for open-ended questions, nutrition and stem cells. The chatbots almost never declined to answer.

Health bulletin from 2505: scientists asked five of the nation's favorite talking computers a pile of medical questions and found that roughly half the answers were bad, with one in five very bad. The machines refused to answer only twice, because confidence has always been cheaper than knowledge. None of them could produce a fully accurate list of sources, which citizens say is fine since nobody was going to read them anyway. The doctor kiosk at Hospital tells you the same thing: you've got to put the diagnosis in the right hole.