Everything We Know About Facebook’s Secret Mood-Manipulation Experiment

@Cpl. Kirk S.:
I am a Facebook coach. I have mentioned this before; on Rally Point.

From the testimony this is how we learned this works:

The primary model shows content to users that is most likely to create engagement and generate and maximize revenue. As it turns out the type of things that keep us engaged are things that generate negative emotions. This is why political posts that get us fired up show up on our feeds. The more time we spend looking at them or replying to them the more of them we get and thus we go in a downward spiral. It's also why kids get content that makes them feel bad, not because they are targeting that content at them per se, but it's content that the model has determined will keep them engaged.

This is where where the Facebook official replies are disingenuous. Sure it's true that there is no conscious targeting of objectionable content to minors but they also know that the net effect of the engagement maximizing model is that kids will get that type content because of how it exploits human nature.

The second part is the "safety" models where they (before showing things to the user) filter objectionable content that the first model is determining should be shown. But as she stated this second part is unreliable, underfunded etc. For example, how are they going to remove objectionable content for minors when they are not even doing a good job of knowing who the minors are? Another example she gave is many of these safety models don't work in foreign languages so people in many countries are getting the raw results of the engagement model.

Her main point is that the secondary "safety" models are not funded well enough and will never be accurate enough to filter out the damaging results that Facebook knows the engagement model is generating. But the engagement model is really, really profitable so they don't want to change it nor I suspect do they want their safety models to be all that good.