Mostly speculation, but they do make a good case. Current Google PageRank incarnations are most likely long past "random surfer" models. Google simply has too much data available to be able to resist using it to improve its performance (and bottom line).
The problem, of course, is that the only way a google-killer will ever arise will be with a completely different approach and not just a data-driven refinement on the current one.
They use all the data they get from every source they can get and they link it all together any way they can. I have no inside knowledge, but that seems pretty clear cut. What's to speculate?
Do they really use all available data? Or does the "don't be evil" mantra prevents them from using specific datasets?
I suspect Google believes using more data is always "good", so it's impossible for "do no evil" to mean "don't use data we legally collected". Does not compute.
The problem, of course, is that the only way a google-killer will ever arise will be with a completely different approach and not just a data-driven refinement on the current one.