And all the more disturbing given the rapidity for which it is moving into the medical field.
Remember MYCIN?
Back in the day, AI meant expert systems, with inference engines driven by rule-based systems using knowledge derived from subject matter experts. AI moved from LISP to Prolog.
Fuzzy logic was a big thing, because it dealt more robustly with edge cases and gaps. But systems were still build from a knowlege engineering perspective.
Now we're at the point where neural networks are fed tons of data, and we don't care about the inner workings. As long the error measure is below some given level, the system is fit for use.
Strange days.