Marissa Mayer, Google, Vice President, Search Products & User Experience at Google (my emphasis):
"Google is painted as the algorithmic purist. That's not our view. The algorithmic approach is important. That said, once you had the basic algorithm, you can layer human elements into it. We have properties like Google Co-op where people can label items and Google Notebook which has human interaction. But you need to layer the two together – algorithms and human elements to achieve relevance."
Given the number of Google employees (10,000 and probably hiring at a rate of 200/week) contrasted with the number of leaks of Really Important Stuff (virtually nill) I think most people at Google work on the tank-principle: they have no clue what they're really doing.
And thus everything that anyone with some real knowledge at Google says about Google, is interesting to me.
I find the above statement by Marrisa Mayer, made during a keynote conversation with Danny Sullivan, quite interesting, to say the least.
After years of claiming no human, manual intervention, basically playing a semantic game, GoogleGuy, a Google representative, finally admitted in 2004 that, yes, although "we put our efforts much more into improving search by writing better algorithms instead of trying to fix a smaller set of searches by hand. We have a quite small set of circumstances that can result in taking manual action". Such circumstances would include especially spam and security.
In 2005 we learn via Henk van Ess that human involvement in Google's search quality is much higher than the occasional spam report. He writes about Rater Hub Google, a series of posts that all started with a job opening announcement on MonsterBoard asking for "part-time, temporary, home-based workers to help with work on a search quality evaluation" who "will evaluate search results and rate their relevance".
The look inside the labs shows that Google is using paid human quality control not just for spam but to assess the relevance of its results when it comes to things such as affiliate sites or payment or registration required content pages.
Matt Cutts, too, in 2007 both opens that human, manual intervention door ("we're open to semi-automatic approaches to ignore paid links, which could include the best of algorithmic and manual approaches") and confirms the Rater Hub Google impact ("our manual spamfighters had detected these links as well").
So, if all this goes back to 2004, what's the deal, Ruud?
The deal is after years of claiming no human, manual intervention at all, Google has gone from admitting such intervention in special cases where needed to broad-blanketing human, manual intervention when applied to search quality… while now we have someone very high up at Google saying, paraphrasing, "yes, algorithmic search is important but you need the human element to achieve relevance".
That's major, especially when combined with Marissa's pointing towards specific human Google properties.
Here's the Google of 2008 you'll be playing against. It ranks documents based on 200 factors, uses temp workers paid to assess search results relevance and quality, has manual spamfighters apart from individual spam reports, feeds back query complaints into its algorithm, uses personalized results, and is preparing to layer mass-human feedback and interaction onto its search results.
That "The ranking process is completely automated and takes into account more than 100 factors to determine the relevance of each result" is a semantically correct statement at best. Now … and for the future.
Recommended related reading: "Is Google Objective? Manual Edits in Search Results"