As near as I could understand the comments Matt Cutts made about the new algo, they don't actually understand it, but they looking at the results it "made sense" to them.
The main thing I object in their use of human testers is the way they did it, giving them pages to judge as opposed to giving them problems to answer (anybody remember the old Google Answers?).
You'd need to have a pretty simple problem to find an answer on eHow, or Amazon, or the IRS site for that matter, since none of them provide expert advice or analysis. The IRS in particular thrives on not providing answers and sending everybody to their "accountant or tax specialist." That's what used to separate the small sites that drew organic linking from the rest, the fact that they offered something of value, the ability to explain complicated issues, experience, analysis.
The new system is just the mathematical expression of what people think of a website as they drive by. Not surprising that 500 million FaceBook users might think a FaceBook button is important, or that people would be comfortable giving their credit card to Amazon or Sears.
And when Google took these new metrics and looked at the results, they saw some ad-drenched article farms crawling with pop-ups went away, and they declared victory. The problem, with all respect to Mr. Panda, is they were asking the wrong question. So instead of quality search results they now produce comfortable search results.