If you give the computer enough examples of what is right and what is wrong—and in the age of the supernova you can do that to an almost limitless degree—the computer will figure out how to properly weight answers, and learn by doing. And it never has to really learn grammar or Urdu or Chinese—only statistics!
