Polling as Witchcraft

One thing that's interesting to me about the way our politics works is that pollsters have this almost witchdoctor-like role in political campaigns. They're extremely important, influential advisers and they're wielding all this survey data and entrails and so forth, but at the end of the day they really seem to just be winging it every bit as much as anyone else. Take this from yesterday's NYT article on disagreements between Mark Penn and Geoff Garin:

Inside the Clinton team, Mr. Penn advocated increasingly sharp attacks on Mr. Obama as Mrs. Clinton’s best option. Long before he joined the campaign, Mr. Garin argued that her route to success lay more in presenting her strengths than in assailing her opponent.

This is obviously a big question, and yet two well-regarded pollsters have no kind of consensus over it, and it's clear enough that you're not going to resolve the dispute by surveying more people or staring longer at existing surveys. But so what, exactly, is the special authority of the pollster supposed to be?