What percentage of jobs actually require general intelligence though? Deep learning can only do things that are relatively simple and repetitive. But that covers quite a lot. Sure, they won't be taking the jobs of programmers or doctors any time soon. But what percentage of the population can be retrained as programmers or doctors?
I would turn the question around and ask you: what percentage of jobs are similar enough to repetitive, algorithmic tasks? We can joke around about it, but I think most things humans do are a lot more complicated than we realize (precisely because they seem so simple to us), which is exactly why the early optimism over AI turned out to be misguided. Even a "simple" office secretary would be hard to automate, because many of these sort of jobs are not specialized at all but consist of many ad hoc tasks that range from making phone calls to negotiating with other humans to collecting debts to getting the boss coffee, etc. Many trade jobs like plumbers, electricians, etc., require way too much "common sense" type reasoning and social interactions to be fully-automated at this point or any time in the near future.
