Popper distinguished science from non-science by falsifiability: if a prediction could theoretically be disproved, it's a scientific one. New information is often useless, because new information can never prove a claim—the only useful information is disproof. On a larger scale, models that turn out to make false predictions are either altered or abandoned. It's all well and good to look at past results and make predictions, but Popper's point was that there can be plenty of models that are consistent with past results and there's no way of choosing between them until they're winnowed down through falsification.
Bayesianism acknowledges that you can't necessarily pick one right explanation, but it tries to account for that by statistically weighting models, and with new information the distribution/weighting is updated.
(This explanation is a gloss, of course, and so necessarily an incorrect model. But hopefully a helpful one.)