[ylmm]

Ah, I think you've misunderstood entirely.

I meant the teams in those companies as opposed to the companies more broadly (e.g., Core Data Science at Facebook, not Facebook in general). I mention those companies together because they're well-known for investing a lot in research (e.g., by hiring PhDs). And in these cases, they're hiring PhDs for reasons that are totally different from the reasons for which they hire engineers (who may also have doctorates). For example, there is indeed a difference between the institution-level goals of Facebook and Microsoft Research, but that difference is less substantial between researchers at Core Data Science at Facebook and researchers on the Computational Social Science team at Microsoft Research.

I'm making the point that there is a difference in the value of a PhD depending on where in the company you work. For the research-oriented teams, the value of a PhD lies in the fact that you've ostensibly been trained to contribute to what we know, rather than just applying it.

Going along with your ML example, the difference would be like comparing Athey, Tibshirani, and Wager's work on generalizing random forests against building a random forest using scikit-learn. I'm not saying that someone without a PhD can't write the paper that they did, but it's for sure not at all just a matter of who's better at writing code.