Just imagine a world where they revolutionized healthcare as they did with advertising!
After wasting many millions on failures like Calico, Google hasn't learned that healthcare is a service that's continuous, custom, and complex. And since the "customer's" health and life are at stake, maintaining health is a damned sight more important and difficult to do sufficiently well than Google's current business: spam.
Tech companies like Google simply don't have the DNA or the persistence to succeed in science-based markets driven by R&D and heavy regulation. They want faster profits driven by higher multipliers and lower standards for accuracy than you can get in healthcare, where accuracy really counts and every case is a little bit different. In medicine it really matters if you get the right answer for your A/B test, otherwise someone may die or they may sue you. That's why so little of healthcare has been automated, and why that's likely to continue.
Never, or literally tomorrow if they take over. "Personalised medicine" decisions made with "AI" will provide a perfectly unaccountable black box to blame when (if ever) the general public realises that vulnerable and poor people have been getting the B end of the A/B testing.
I agree with the rest of the post but god have mercy on us if these companies take over healthcare. I never thought I'd be living in such a totalitarian state in my life.
This is a problem for dataminers like me within pharmas too. Most of our clinical trial data is locked up contractually within contract research organizations (CROs) who are responsible for taking specific measurements of it as part of a trial. From what I've seen, dataminers within the company can gain access to only selective bits of that data (one or two features, perhaps) from all the patients in a trial, but never to all the data on each patient, and never data from the trial that spans multiple CROs.
This 'federation' of clinical trial data collection serves the needs of the trial design (and satisfies HIPPA regs), but it makes mining the data or asking new questions of it almost impossible. Too little data is available (and never longitudinal longer-term outcomes, since trial durations are short) to allow mining of more than short-term signals, like immediate response to treatment. Pharmas are unlikely to change that practice, in part because it's easier to federate the analyses and allow the raw data to remain in distributed silos. But federation is also a good way to discourage asking creative yet inessential questions that might later prove awkward or invite lawsuits, since litigation is an omnipresent worry in US healthcare.
How companies like Calico/Google hope to navigate such minefields is a question they have yet to answer successfully.
(note: I worked at google over a decade and work for pharma now. I helped establish Google's move into health care and even launched products intended for pharma users and set up the ethical guidelines to work with private data like clinical records. Unfortunately, Google's Research leadership chose to take their health efforts in a completely different direction so now I focus on stuff that will actually make a difference for pharma).