The problem is you are applying the sense of "knowing" to a program. The algorithm will most likely answer any and all questions, but with varying accuracy. Of course, a threshold for the expected accuracy before initiating a button press could be set, but it is still a very different concept. Programs are a way off from AI, but I would love to see how this project fairs.
If we knew more about the workings of a brain and could electronically replicate it, there would be less of a distinction in my mind.
Now cue the arguments that humans don't "know" anything either...