Listening to a presentation on synthethic publics*, something we’re increasingly seeing discussed in the participation and democracy space.
Reassuringly, the research suggests that while synthetic publics are good at matching public opinion survey results on existing situations, they are not good at addressing new situations.
On one level, this is what you’d expect – the training data exists for the past, not the future. The AI is guessing what a group of people in the past might think, it’s difficult for it to take new situations into account.
I say it’s reassuring, though, because for me democracy is about defining the future, not commentary on the past. I don’t think an aggregate of past public opinion could ever be a reliable source for synthetic democratic participation, because the nature of being citizen is having a vision of the future, and working out in society how to deliver it.
That doesn’t mean we should ignore AI in our field, of course, there are plenty of organisations thinking hard about how to use it, and Democratic Society is one of them (I’m about to speak on a panel on this very topic). But this is further evidence that it should be used to enhance participation and deliberation, not to replace it.
* (where existing public opinion research is aggregated and the data is queried using AI to get a synthetic public opinion on an issue, or AI personas are created who can be interacted with as if in an interview).