A significant quantity of the ad space is pernicious: The ads seeking to be displayed are explicitly intending to manipulate the viewer to change their cognitive orientation to the benefit of the buyer, with anywhere from callous disregard to outright contempt for the viewer's outcomes.
Big tech has significantly aided in the delivery of these harmful messages, and seeks to use personalized information to find the channels, times, and context that are _most_ effective at undermining the will and agency of the viewers.
Are there models where an information / recommendation broker could justifiably collect what would otherwise be concerning amounts of personal information? Potentially, if it were using that information to compute and facilitate the maximal individually beneficial outcomes of each individual user, but that's not what's happening here. Right now, we're building a system that would happily have people die painfully and preventably at 40 after living miserable lives because the economic returns of that model maximize the returns on advertisers / delivery networks.