And the way we prevent abuse is with anti-discrimination laws, and AI won't change the deals here. If, for example one notices that a company doesn't hire black people even though there are plenty of them applying with suitable qualifications. Regulators could step in and ask what's wrong, and I doubt "that's my AI" will be a satisfactory answer, but neither is "that's Bob".
Have you not noticed how heavily those are contested in many jurisdictions. If you discover some way to draft legislation on morality such that it's effective, consistent, and widely accepted in a disparate population, I look forward to hearing about it. I can't wait to read about the solution to spam, scams, and discrimination deployed at industrial scale. In the meantime, it's wise to take the amplifying aspect of technology into account and consider its potential for abuse as well as for good.
the amplification factor works for the good and for the bad, so you need to focus just on the bad
That's just hand-waving. You have no plan for dealing with bad things, but you are worried about the loss of good things. This is not so different from saying you want to cash in on opportunity, the wholly predictable downsides of the opportunity are just someone else's problem. Guess what, nobody wants that problem so it just festers in proportion to the enthusiasm with which people chase the upside.