It would be interesting, and has been tackled eg by http://www.cs.cmu.edu/~mheilman/papers/heilman-smith-qg-extr... . Nielsen specifically hypothesises that coming up with the questions yourself is part of learning and building understanding. Using software-generated q’s and a’s would be like using someone else’s deck, only the someone else is a program.

I haven't studied this issue much, but was surprised to hear on one of the learning scientists podcasts that generating cards may not be very useful for learning / consolidation.

(It was in one of their first 6 episodes, not sure what to think of the claim)

http://www.learningscientists.org/podcast-episodes

This would still be useful to guess-and-check against what questions you come up with versus all possible questions vis a vis the program.

Also, damn, I've been looking for something like this for ages.