My theory is, to build an AI that just helps with 'cumbersome things', and does not come bundled with the faculties to do a whole lot of other things, is impossible. Jagged intelligence and all of that, but intelligence is about patterns on all levels of abstractions, so if you have something that can do your expense report for you, you probably have something that can also evaluate whether its a good expense report and displace someone who does it by hand. To call this unethical is, well, an opinion. The whole history of humanity can be summarized to "learn useful things and automate them".

As for tools in the wrong hands and all, sure. Obviously. My point was that humanity never abandons a technology because of risks, or real issues.