I love how the only actual example they give is one in which the problem was effectively created by humans interfering. :/

This is always the danger when you give humans autonomy.

I think there is a big gap between autonomous technology that 'works' and autonomous technology that works in all circumstances and won't fail in a way that causes harm.

Re: the example of caring for the elderly. The harm here is going to be psychological and cultural. We are a pretty selfish generation in general when it comes to how we view our responsibility to the elderly. I'm not sure being able to shovel off our responsibility to past generations on a few robotic servants will better us as people.

Modern treatment of the elderly is a matter of economics and demographics, not one of moral decay. There are simply too many elderly and not enough workers for the elderly to be afforded the same treatment as a generation ago.

On the other hand if morbidity gets compressed [1] we may need fewer caretakers.

[1] People suffer from age-related illness for a shorter time. They stay healthier for longer. Some statistics point in this direction.

Obviously the real concern is that "such machines that do the sort of jobs people find dull, dirty or dangerous." will get angry and unionize.

We must prevent this or all is lost.

Judging from how often software breaks, not very.