To be fair, I think the article (and the original paper) do more than bug sniffing; the code that fails test cases is being automatically modified to eliminate the bug.
The "Do what I mean" programming idea is certainly a good one. One of the pioneers of genetic programming, John Koza, once described the possibility of using genetic programming as a companion for human innovation. Intead of the algorithm completely replacing human written code/design components, it could produce potential solutions to problems posed by an operator which then go through tweaking or analysis by humans and can be sent back for more evolution, turning it into a sort of partnership (I can't find the exact source, but I used it in a project presentation a few years back). Kind of cool to imagine: pair programming with a computer partner!