I contacted Williamson about this, and he wrote back:

If I remember rightly, Bacchus et al were suggesting that, if you all you learn is that 80% of all Scandinavians are Swedes then you should by default believe that Peterson is a Swede to degree 0.8 (with no assumption of random sampling). This seems right to me as a default principle, given that you fully believe that Peterson is Scandinavian. But the norms of subjectivism do not require you to set P(Peterson is a Swede|80% of all Scandinavians are Swedes) to be 0.8. Indeed this conditional probability can be more or less anything – it is subjective. So if it isn’t set to be 0.8 there is a mismatch between what Bayesian conditionalisation requires and the rational course of action.

I hope this makes more sense now, and apologise if the presentation in the book was a bit terse!