Great post, and I agree.
Working in healthcare BI for a few years I saw a lot of 'pie in the sky' dreams when it came to ML\AI. Lots of talk about possibilities and little action. When ML\AI was applied at my organization it was first used in the finance, despite all the talk about improving clinical outcomes. It made sense to focus on finance first. No patient impact or requirement to involve clinical staff, as well as a large large amount of well understood data to work with.
In order to build working clinical models it will likely be necessary to pool data sets from multiple organizations. Something our organizations EMR was working on(or at least appeared to be). This is exciting and concerning at the same time. The possibilities for improving patient care with a vastly larger data set need to be weighed against concerns of misuse and privacy.