What Google is extraordinarily good at is evolving the metadata of search. It knows how to search so many things so well, that it can begin to be thought of as an association machine. It's quite robust, and it is doing so on a completely unbounded problem - that of finding anything in the entire domain of the internet. But what if you added bounds to that problem and allowed Google to add additional and interactive metadata to that of some small bit of expertise? Would it scale semantically?
What I have in mind, of course, is the Lorite Interrogator, which is like one of Iain M Banks' drones connected to a ship mind and assigned to your protection. It can know human information - all of it. But specifically, the Lorite Interrogator will find your affinity toward philosophical systems of thought and help you to know the logical conclusions of all of your assumptions about the world. I'm not sure, but I think this is a sort of Kantian proposition.
Right off the top of my head, I think that some weighted graph between questions and answers might be the appropriate metadata. There would then need to be some self-evolution involved that helps such a system recognize relevant questions. So far, I think Google can do this, but is a bit more linear than a human being might be. If I asked the basic question, "Is prostitution good?" Where would we start? What we find is that people ask that question in any number of contexts. So it is the management of question context that is where our metadata needs bucking up. A philosophical framework is what we need, and thus some way to take all of the sentences and premises and conclusions and link them together. It is search plus metadata, and Google has shown that it can be scaled up. Computationally, there isn't a problem. It's just getting some way to get many eyeballs looking for those kinds of answers and diverting some slice of all of Googality in that semantic direction.