I think the thesis here is fundamentally correct. As automated systems become more proficient and more complex, the proportional number of “hard failures” relative to all system failures will increase. At the same time, though, those systems are also intended to reduce the likelihood of any system errors at all. So the question remains as to whether or not the actual “risk” (the probability of a failure times magnitude of the severity of that particular failure, summed across all failures) is greater in systems with high automation. I suppose this deserves to be inspected on a case-by-case basis.