Leaked internal Facebook docs reveal that top company execs were told that the company's "engagement algorithm" was sowing division among its users, by feeding people ever-more-fringe versions of their own beliefs.

https://www.wsj.com/articles/facebook-knows-it-encourages-division-top-executives-nixed-solutions-11590507499

1/

The slides warned that the algorithm served "more and more divisive content in an effort to gain user attention & increase time on the platform" and "64% of people who joined an extremist group on Facebook only did so because the company’s algorithm recommended it to them."

2/

Writing in @TheVerge, @nickstatt attributes the emphasis on controversy to FB VP of Global Public Policy Joel Kaplan, an ex-GW Bush official who attended the Kavanaugh confirmation hearings to give moral support to the accused rapist.

https://www.theverge.com/2020/5/26/21270659/facebook-division-news-feed-algorithms

3/

Kaplan is Facebook's sop to the right; and he championed Facebook's political ad policy of allowing provable falsehoods to appear in paid political ads. He also sidelined proposals to weaken the influence of hyperpartisan "supersharers," who are disproportionately rightist.

4/

Facebook's marketing material promises advertisers that it can use machine learning and surveillance data to convince potential customers to buy whatever they're selling, but, as with all marketing material, these self-serving claims should be viewed skeptically.

5/

That said, I'm willing to believe that machine learning systems CAN "sort by controversial," making rough judgments about which material is likely to evince anger from readers, just by looking at simple signals like vocabulary choices (e.g. "pro-life" vs "pro-choice").

6/

Sort by controversial is indeed cause for concern - but it's a far cry from the feats mind-control that Facebook claims it can perform when it is trying to sell its advertising platform to marketers.

https://slatestarcodex.com/2018/10/30/sort-by-controversial/

Image: Rob Beschizza, modified

eof/