I find a software parallel here. When building new systems in poorly understood domains, you can't proceed as if you already know the paradigm. Instead you have to poke around and try things until you find a working paradigm.