For a very real example of facebook bias a group created two Facebook groups full of similar hatred but directed at two different ethnic groups.
One was shut down almost immediately, one was still up when the video was posted.
Won't name them as I don't want to drag politics into HN but I am fairly convinced it is repeatable (and if anyone is interested in giving it a try I can try to dig out a reference and post.)
Interesting experiment, but two groups seems like a too small a sample.
If moderation is done by humans, you could simply prove that one moderator had a worse morning than the other, or that through luck of the draw Group A went to a sympathetic moderator while group B did not. That could demonstrate individual bias, but not institutional.
If moderation is automated, how do you control for the weight of keywords, phrases, etc? Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged.
If both groups were designed to be unpleasant, but various "blacklist" words and phrases are weighted differently, Group A might have achieved "flagged" weight while group Group B was "flagged -1." One might have scored 99 pts vs 100 for the other, but the visible difference is that one remains visible while the other does not.
Thinking out loud, now... let's say that Facebook does have a political bias. How do you prove that Facebook's financial success is despite that political bias, and not because of it?
Is it possible to have a monopoly on digital socialization? If you can't declare Facebook a monopoly, and therefore subject to greater regulation, then your only lever over Facebook is social pressure, presumably in the form of bad press and boycotts, which will reduce their advertising revenue. How do you prove that the existing political bias is leading them to sub-optimal revenues?
> Maybe Group A was full of statements that almost threw a flag but not quite, while Group B contained mostly innocuous wording but (only) one phrase that actually got flagged.
Couple of months since I saw this, so can't say for sure but I think a major point of the experiment was that the two "hate groups" were equal except which ethnic group they were directed at.
One was shut down almost immediately, one was still up when the video was posted.
Won't name them as I don't want to drag politics into HN but I am fairly convinced it is repeatable (and if anyone is interested in giving it a try I can try to dig out a reference and post.)