Artificial Intelligence Convinces More Targeted than Humans

Large language models such as GPT-4 show a surprising strength in online debates: they argue more convincingly than humans – at least when they can use information about their counterparts. This is the result of a recent study conducted by Francesco Salvi of the Swiss Federal Institute of Technology Zurich (via press release). In a controlled experiment Salvi and his research team analysed the argumentative behaviour of 900 US participants. They…