It's an interesting idea and can make for educational games, but I don't think it will see much traction as a means of doing useful work. The vast majority of bugs exist in the gulf between a programmer's characterization of a problem (or a problem's requirements) and the actual use case. And of course, if your puzzle generator can understand what is meant by a piece of code sufficiently well to identify a bug, it can fix the problem without human involvement.