Rohin Shah comments on [AN #59] How arguments for AI risk have changed over time

[Rohin Shah]

I see, so the argument is mostly that jobs are performed more stably and so you can learn better how to deal with the principal-agent problems that arise. This seems plausible.