It seems solid. But I would like to have the same study with an human conversation instead of an AI one.
It's always "well, my cousin's friend's kid got autism right after getting a vaccine" and before you can look up the necessary studies to refute it, they're already talking about hydroxychloroquine.
An AI *that the user trusts* is going to be much more effective at that sort of whack-a-mole.
It is also about improving human speech too, what can we do to match AI performance ? I think, that are some very interesting questions that it may helps to answer.
How can we make AI even better at « brainwashing » ? Does the value of information matter ? I mean is it easier to AI to « nullifies conspiracy beliefs » or to « nullifies scientific ones » ? We may have opinions on these issues, but the scientific reality behind them is not obvious. It's so exciting !
From this study, it doesn't seems that speed mater that much here since it was a short conversation with a fixed subject.
You introduce a new variable in the equation. Trust. I think that is justified and its posing a whole set of new questions. I would like to answer them but I can't.