For advertising purposes it really only depends as a whole how truthful a picture the likes tell. If the likes are on average 70% truthful it is still going to be more effective targeting than not using them at all.
It's a reference to a study where they had an algorithm fed with a certain number of facebook likes compete against personal acquaintances (including spouses) in predictig personality traits. The algorithm won:
But what's the supposed ground truth that spouses and algorithms were tested against? Self reporting? Some other glorified coin flip algorithm that maybe just did the same mistakes as the "thumbs" algorithms? An export panel populated with people that, unlike the benchmarked spouses and acquaintances, used the same jargon as the algorithm authors?
(Glancing over the footnotes it seems to be (b), some other algorithm)
The beauty of adtech: it's perfectly fine to be wrong as long as advertisers think you are right.
On meta level, this "better than your lover" meme/study is surprisingly enlightening.
The parent's argument assumes though that this kind of accuracy is achieved regularly and consistently today, which is way different than the report of one study.
It doesn't need to be consistent in order to be horrifying.
The fear is that big data tech will become radioactive.
Imagine a bus full of school kids crashes because the driver was a recovering alcoholic who fell off the wagon.
Some smart SV engineer realizes their tech spotted the driver visited AA groups regularly & his wife just left him. The algorithm knows this data makes him an excellent target audience for _new alco-energy drink!_.
It doesn't really matter if the technology is even capable of that yet, what matters is that this is the sort of outcome that adtech engineers are trying to create.
