The problem with not having a math background is that you will succumb to all the people who throw around buzzwords but don't know what the hell they're talking

The problem with not having a math background is that you will succumb to all the people who throw around buzzwords but don't know what the hell they're talking about. You need to be able to recognize (and assert authority over) someone whose first thought is that "support vector regression and a bayesian net" is the right way to go. Or, to put it nicer, refine their thought into something useful. The problem with not having a software and "grungy coding skills" background is that you aren't able to efficiently reproduce/check/verify claims being made which are often wrong or misleading. A recent example I had of this is someone studying how an algorithm behaved on a subset of interest of a particular population, and they didn't even bother to check that the subset of their data was statistically large enough to draw any conclusions (only 17 records out of thousands). Needless to say this example also failed on the reproducibility and reuse fronts. The problem with both in academia (even at polytechnic non research schools) is that they don't know enough about each other.