Fuzzing can be used to find bugs. Could it be used to fix obscure bugs, bugs in complex data flows, by fuzzing over inputs? Could a fuzzing like approach be used to generate code (even code we have no chance of comprehending) that patches a n obscure bug? It seems possible. By extension could fuzzing generate code against arbitrary constraints?
Not completely off topic, but not really serious either : <a href="https://en.wikipedia.org/wiki/Infinite_monkey_theorem" rel="nofollow">https://en.wikipedia.org/wiki/Infinite_monkey_theorem</a>