Either survival-enhancing mutations are the result of random processes or there is some (self?) organizing principle at work. (And I accept that biochemical reactions can, in some cases be self-ordering.)
However, the example used (the flagellum) is not merely an assembly of components, but an assembly of sub-assemblies.
Simply adding one more component to a successful sub-assembly (with, as postulated, a totally different function) is hardly likely to constitute a more-survivable assembly.
Unless each of the subassemblies constitutes a survival-enhancing trait in and of itself, the likelihood of arriving at the final assembly is remote. And, even then, there must be a demonstrated path via which the subassemblies could merge into a higher-complexity (more survivable) assembly with a distinctly different function.
Simply showing that removal of a sub-assembly can leave a survivable construct does not, to me, show a path to higher-level assemblies. In order to have "natural selection" at work, the higher level assemblies must be shown to be possible -- and the intervening component-at-a-time sub-assemblies-in-development must be shown to be survivable in and of themselves.
Let's just say that I don't find that component of Miller's argument to be convincing...
Follow the argument. He was not trying to show a path to higher-level assemblies. He was demonstrating the fallacy of 'irreducible complexity'. Nothing more, nothing less.