There does not exist a single methodology that can always avoid local minima/maxima, short of somehow brute forcing the entire design space. (If evaluating the design space and locating the best design is somehow not an NP-hard problem, it is close enough for human purposes.) Accusing one methodology of producing local minima/maxima is therefore a noop. One must demonstrate that it is somehow more prone to such a result.
Good luck with that. I would despair of even getting the definitions correct, let alone demonstrating a result with even minimal rigor.
My intuitive feel is that if you feel like testing is getting you stuck in a local optima, this is more likely to be because your testing is actually giving enough sense of how your design fares that you can actually feel you are in a local optima, rather than the usual situation where you are just clueless, happy about it, and not even in a local optima, let alone the global one. You can't blame the equipment that allows you to detect that your are in a local optimum for getting you stuck in it; that's textbook observation bias.
There doesn't exist a single data-driven methodology that can avoid local maxima without brute force the entire design space. Conceptualizing things and being creative is what avoids local maxima.
Nope. Being a human is not the magic algorithm for finding global optima. We aren't anywhere near that smart. There's no methodology for finding the global maxima, period.
I don't think the facebook case studies in the article are against AB testing per se, but more against the fact that AB testing is often used to focus on minutae. It seems to be a psychological temptation to use it to focus on minutiae, rather than something that's inherent in the research method itself.
Good luck with that. I would despair of even getting the definitions correct, let alone demonstrating a result with even minimal rigor.
My intuitive feel is that if you feel like testing is getting you stuck in a local optima, this is more likely to be because your testing is actually giving enough sense of how your design fares that you can actually feel you are in a local optima, rather than the usual situation where you are just clueless, happy about it, and not even in a local optima, let alone the global one. You can't blame the equipment that allows you to detect that your are in a local optimum for getting you stuck in it; that's textbook observation bias.