The argument goes something like this: Given the right data set, a data scientist with no domain expertise can out-perform experts that have been working in the field for decades. For example, providing weather insurance or marketing strategy.
One of the panel (I didn't get their name, sorry) at the media meeting I attended called domain expertise "quaint." The idea being that if you have enough data, you can move beyond hypothesis testing initiated by domain experts. That is, instead of saying "I think this may be true, let's gather data and see if that's right," moving to "let's gather data and then we'll be able to see what patterns emerge."
Though there was a fair amount of pushback against the "domain expertise is quaint" argument, I heard enough of this during Strata to wonder if there's a grain of truth there.
The audience at Strata ultimately voted (barely) in favor of machine learning over domain expertise. I'm curious what the ReadWriteWeb audience thinks. Given a large enough data set and a sharp (but generalist) data scientist, can you replace domain expertise?