One issue with data-driven design is that it can be incredibly inefficient for developing new behaviors. If you're changing hues, sure, it works. When it comes to building a novel interaction that might really push the needle, users typically won't learn it overnight. You can't test even two or three significant UI changes without incredible patience and inefficiency, because the data effects might not materialize for weeks, and the user re-education costs at each step are very high.

This effect is multiplied when you're trying to test design changes in products where users need to interact with one another (makes testing new design on only a subset of users incredibly difficult). Data-driven design in these contexts will likely just tell you to stick with what already seems to work.

At the moment, it's unclear if Google's really in the business of developing new user behaviors. They're operating a cash machine with reasonably well-understood dynamics, and will probably do pretty well just tuning it. I think this may turn out to be a strategic issue for them in the long run, but for now it seems like the answer is simple: if you're a talented visual or interaction designer, don't expect to have much fun working at Google.