Technology

What's become really disturbing in the past ten or so years is how they've applied ML to the targeting. Used to be it was just basic keywords and demographic stuff. Now the big platforms put your entire last decades' worth of history (often both web browsing and social media) through a bunch of filters and spot that people who are like you are more likely to buy this product or join this website.

The reason why it's fucked up is that "people who are like you" could mean things like anorexia, or addiction problems, or the kind of relationship trouble that makes you a soft target for incel indoctrination, or a bunch of other protected vulnerabilities that would get a company sued through the floor if they actually did it up front. But because it's all just a bunch of untagged probability distributions in a black box, it's impossible to "prove" that you deliberately and knowingly targeted a gambling addict to push a high interest credit card, or a recovering alcoholic with booze, even though that's exactly what happened inside the bundle of weights.