Earlier quoted context omitted.
Because a machine has much more capabilities than one salesperson. The salesperson doesn't know in what shops you have been before. The salesperson might also not know you talked to his colleague the day before. This is about trust and privacy. You can't trust what they do with your data.
So if the salesman has an eidetic memory it becomes unethical? How about just an above-average one?
1) These people are rare in the general population, and demand for their time is likely to be incredibly high. Therefore, they cannot be deployed everywhere, unlike machines.
2) When confronted with a human being in a sales scenario, people have a chance to be on guard against potential manipulative behavior. When the sales scenario becomes ubiquitous and invisible, it is much harder for people to avoid being taken advantage of.
3) Ethics are not so absolute. Something that is only mildly bad at an individual level can have terrible results when thousands are doing it. (Littering, for instance, or illegal hunting/fishing.) This is known as a social trap, and it leads to negative outcomes for everyone involved.