| Walmart is "manipulating" the placement of products on the shelf so that it's more likely for you to engage in bulk buying when you visit their stores. Both Facebook and Walmart have a fiduciary duty to their shareholders to create value for them. The difference is that, with user generated content, the idea of black and white "bounds" of the law is no longer applicable and you have to devise a system of checks and balances based on probabilities. You can consider 10'000 posts for offline analysis: give them to some human raters and decide retrospectively what engagement and thoughts (positive/negative) are they generating in teens, which should enable you to draw some statistics about the expected average outcome. This doesn't mean it's either scalable or economically feasible to do so in real time for every post (so you cannot take decisions based on something that doesn't exist at the individual post level). You can have multiple algorithms, send all of them to human raters and get for each algorithm some aggregated behaviour, but then we're back to the book question above -- what ratio of positive vs negative outcome in outliers is acceptable, and how do you define a "legal"/"allowed" algorithm? |
Facebook doesn’t have a duty to manipulate content. This is a very weak excuse that works mostly for people directly benefiting from the situation. Didn’t cigarette companies have a duty to maximize profits? Pharma companies pushing accessible opioids? Is that a more apt analogy?