Im no software engineer so i’d like to know why do social media platforms rely mainly on user feedback/reports to get rid of fake accounts instead of software. I feel like we’re at a point where AI could be use to identify fake accounts with parameters like repetitiveness of words, number of posts per hour/day, words used, age of account, content of the account, username, etc. Why is is so hard?
In: 0
> I feel like we’re at a point where AI could be use to identify fake accounts with parameters like repetitiveness of words, number of posts per hour/day, words used, age of account, content of the account, username, etc.
You might feel that way, but it isn’t really true. (Or rather, it’s partly true, but it’s hard to do that reliably without catching some legitimate users, and good bots can get past it. Sites do in fact ban plenty of bots automatically, but you’ll notice the ones they miss.)
If nothing else, a lot of models these days train *adversarially*. What that means is that instead of building a model to talk like a person, you build *two* models: one that talks like a person, and one that detects machines trying to talk like people. You train the one that talks like a person to evade the one that tries to detect it until the one trying to detect it can’t do so reliably anymore.
Latest Answers