> Why wouldn't we expect these systems to learn that among other things, information that increases polarization results in the desired goal of increasing reven

> Why wouldn't we expect these systems to learn that among other things, information that increases polarization results in the desired goal of increasing revenue, absent of any deliberate intervention by human beings? This is all well and good, but revenue is not an input to most machine learning algorithms at Google. First off, there is a pretty serious firewall between search and ads, so any organic results will be completely oblivious to revenue. Search systems literally do not have access to that kind of information. Even within ads, the concept makes sense only if you don't think about it too much. The big money-maker ads are when someone is trying to buy something. Ads around viral content are just not something that people can pay much for, because the margins in the news business are very thin. Now if you want to make an article that the media has learned that spectacle and outrage drive clicks, I am here for it. (But this has always been true. Yellow journalism was a thing more than a century ago.) And I guess you can make an argument that Google facilitates this by pointing you to the media, and by participating in the feedback loop. But I don't think Google revenue is as causally linked to outrage as you seem to believe. After all, correlation usually doesn't indicate causation.