How people and machines are smarter together
Nick Polson, James Scott 2018 Beaverton 620.82 POL
p03 AI is an algorithm, not a droid.
Mentions of robot cars; the usual conflation of all drivers in all circumstances compared to a few million miles driven in limited circumstances, and the usual either/or rather than "both". Most drivers are sometimes incompetent, some drivers are mostly incompetent. Should mostly-competent robot cars replace the most competent drivers? Who pays for this expensive imposition?
I'm a cyclist. Will self-driving (robot) cars include the extra expense of detecting cyclists, or will municipalities prohibit cyclists on roads?
Other than that, most of the book is useful and interesting. Much of it is about answering questions with "big" data, and common errors in mathematical reasoning. Bayesian analysis is lauded, but not described well enough for John Q. Average to calculate it. Florence Nightingale is featured as an data-driven analyst, not "merely" the founder of modern nursing and hospital design.
p209: two giant big-data studies on the effects of bisphosphonate osteoporosis drugs on esophageal cancer. Cardwell@Belfast ("retrospective cohort") says no, Green@Oxford ("case-control") says yes, using the same data but different assumptions. I guess it is OK to start using bisphosphonate drugs, but not OK to have used them. Or it is possible that we have a lot more to learn about the very very complicated human body. Or that there are much better ways to avoid both osteoporosis and esophageal cancer; my guess is that they share causes.
Dimaggio hot hand effects, birth control, traffic accidents. This should temper optimism about the specifics of robot cars, and the errors inherent in overgeneralization.