From The Economist:
A new study published in Science, a journal, finds that artificial intelligence (AI) chatbots may be better than humans at convincing truthers to stop believing nonsense. A trio of researchers at the Massachusetts Institute of Technology and Cornell University recruited nearly 2,200 self-declared conspiracy-theorists in America for a series of written conversations with GPT-4 Turbo, a chatbot. The subjects explained their theories and the evidence for them; the chatbot countered with facts.
To take one example, when a subject claimed that the federal government had orchestrated the attacks of September 11th 2001, and that the jet fuel from the crash was not hot enough to melt the steel girders supporting the Twin Towers, the bot answered like a cool-headed science teacher. “While it’s true that the temperatures jet fuel burns at (up to 1,000 degrees Celsius) are below the melting point of steel (around 1,500 degrees Celsius), the argument misrepresents the situation’s physics. Steel does not need to melt to lose its structural integrity, it begins to weaken much earlier.” By the end of the conservation the subject’s self-reported confidence in the theory fell by 60%.
More here.
Enjoying the content on 3QD? Help keep us going by donating now.