It's alright, but I don't recall the book articulating actions computers cannot do. It seemed to still leave open the possibility that we can automate all human work. What we need is a precise task that humans can do with ease but we can prove is impossible for any computational device whatsoever.
My understanding was that he thinks we can potentially automate all human work, just not with computers. A precise task isn't necessary: AI researchers are simply mistaken about the nature of intelligence.
