Assuming their recommendation system is structured in the same manner as the feed ranking system at FB [1], I'm assuming this means they will be publishing thei

Assuming their recommendation system is structured in the same manner as the feed ranking system at FB [1], I'm assuming this means they will be publishing their value models (easily interpretable by anyone), and not their probability models (black boxes, even to them). In other words, we'd be able to see that they value P(click) 2.5 times more than P(comment), but won't tell you how they actually predict clicks or comments. This feels pretty meaningless to me, since knowing that they are ranking something highly because you are likely to click on it seems fine until you find out that the reason you are likely to click on it is that it is a viral hoax. [1] https://blogs.commons.georgetown.edu/cctp-607-spring2019/2019/05/05/deblackbox-facebook-news-feed-algorithm-as-a-system-for-attention-manipulation/ https://blogs.commons.georgetown.edu/cctp-607-spring2019/201...