4 ms·
Not exactly. It's the other way around. Just like herbstein said earlier theres no boolean flag that tells facebook that something is "inflammatory and tribal".
by kevinsundar 5y ago
Not exactly. It's the other way around. Just like herbstein said earlier theres no boolean flag that tells facebook that something is "inflammatory and tribal". Its not like facebook has some algo that says if a republican posts and a bunch of democrats comment then boost this to a wider audience (repeat ad nauseum for every category).
Their algorithm is just identifying content that lots of people interact with and then it boosts that to a larger audience. That content happens to be inflammatory and tribal content because people, unsurprisingly, like interacting with that sort of content.
If people didn't interact with it, it wouldn't get boosted. So ultimately its society's fascination with inflammatory and tribal content that causes the algorithm to boost it.
Again people pretend like facebook somehow looks at a picture and says "wow this person is attractive, wouldn't it be nice to affect someone's mental health by showing this to them so they can compare themselves?" Thats not how it works.
- hairofadog 5y agoIt sounds like you think that I think that they’re showing people content with the intent of sewing unrest. I don’t; I believe their intent in all cases is to make money and expand their empire. That’s their intent when they’re boosting inflammatory content, suppressing content that might challenge people’s biases, and that’s also their intent when they ban users whose content is so inflammatory that it actually invites real-world violence. It also sounds like the most important aspect to you is their intent? Like, did they mean to censor something, or were they just trying to increase engagement? So let’s say an engineer comes up one day and says, hey boss, I did some A/B testing with three different algorithms: “A” is just a chronological timeline, but B and C each use a different method of boosting and suppressing content. B increases engagement with some users but also drives away some users altogether. C shows content that generates fewer engagement “hot spots” but shows higher user retention. And the boss says, we’re in it for the long haul; let’s go with C! By the way, what’s it do? And the engineer says, well, an AI is trained to look for posts containing political opinions outside the mainstream and shows a cat picture in the timeline instead. Guess it turns out people really like cats. So would that be ok? After all, in this hypothetical situation their intent isn’t to harm anyone; it’s just to increase their user base and make more money.