The problem of working out what a program does given its input and output is a very general one. Maybe there could be a competition for programs that do this? (Like the Loebner Prize, but more sensible since the Loebner Prize doesn't really have anything to do with progress in AI). Or maybe such a competition already exists and I don't know about it.
Perhaps a generalisation of this could be used as an intelligence test for AI programs. (for criteria for such a test, see http://www.overcomingbias.com/2008/10/economic-defini.html )