> The empirical way to do it is more along the lines of neuropsych -- taking a look at the physical processes involved. I think human psychology and behavior is a lot like the plumage of a peacock; pretty, loud, but ultimately an abstraction above what is really going on.
There is no denying it is an abstraction. I would in fact claim the converse: all human scientific study to date has dealt with abstractions of varying degrees.
To offer a counterargument, we treat a great deal of systems like (semi-)black boxes and ultimately manage to derive useful statements about those systems. Examples include economics, black-box software analysis (like fuzzing), biology (e.g. we've learned many useful facts about the human body before we even knew of the existence of the cell), even basic physics (like the physics of gasses, as an epitome of something randomized, but still following broad rules). There is no reason to assume human psychology is different in this particular aspect.
In fact, we cannot assume so, as a lot of psychological knowledge we've gleaned demonstrably works, despite other results proving to be irreproducible due to the usual reasons[1]. As an example, it's hard to deny the existence and predictive power of modern human personality trait models.
It should also be telling that many modern, advanced statistical tools were invented by none other than psychologists, for use in psychology.
Neuropsych is definitely a worthwhile approach and I wouldn't separate it from the rest of psychology. Ultimately, we cannot really hope to derive everything using the bottom-up approach any time soon so a variety of approaches are needed. Also, to continue the above example, many studies of personality traits have shown connection to underlying genetic and environmental factors, such as this one: http://doi.org/10.1016/S0191-8869(01)00137-4
[1]: Science is hard, humans like and many times need to cut corners, the black-box system is extremely complex (as you noted).