>It makes Bayesian statistics seem like a matter of letting your expectations influence your results.

The smart approach is to have a set of a prior expectations, run through the data, and see what comes out the other side. Essentially, through a large amount of experimentations, you can narrow down what assumptions were indeed correct.