Because in real life, there are plenty of laws that prevent invasions into privacy. Also, the people capable of doing the same in real life don't have the tools

Because in real life, there are plenty of laws that prevent invasions into privacy. Also, the people capable of doing the same in real life don't have the tools to create truly personal advertisements. For example, a supermarket chain might use an aggregate of purchase histories (i.e., products most often purchased together) to influence product placement in the store. They won't re-position products for each customer as they walk in the door, though... Online, this can, and does, happen. It might not seem like a big deal when it's useful. However, there will be many instances of people trying, and failing, at creating a useful product. I think that's a valid point of distinction: when people fail at adopting new technologies right now, the results are mostly harmless. If they fail with machine learning, there could be some major privacy concerns.