"It can just query requests over network like maeon3 suggests."

Of course, you do realize that requests over a network need a server to reply to them? Millions of people making such requests would need quite a fews servers indeed. That is where the mass deployment would come in.

Still, I wonder... If someone was crazy enough, they could probably accomplish quite a bit with data scrapers, regular expression matching, and some distributed computing fairy dust.

Yes the hardware requirements are ridiculous and the business model even more so. There would be tricks to lessen the load, reverse proxies so that when a question is answered it doesn't have to be re-answered until the question requires a re-evaluation. Redevelop IBM's Watson to run on armies of cloud servers and computers. Make the service free, every human gets a key and can consume X minutes of Watson a day. Have a high-priced "premium service" where you can buy more of Watson's time and get real time re-evaluations to questions.

Yes it would be a more massive undertaking than Larry and Sergey took for google, maxing out their credit cards on hardware. But you would get your name written in the sands of time as mass deploying the first army of semi sentient machines to service the masses. This project brings us a step closer to a type II civilization.