I like the line of thinking, but who actually provides the agent, and what are their incentives?
This is far from a perfect analogy, but compare it to the problem of email spam. People first tried to fight it with client-side Bayes keyword filters. It turns out it wasn't nearly as simple as that, and to solve a problem that complicated, you basically need people working on it full time to keep pace.
Ranking and filtering a Facebook feed would have different challenges, of course. It's not all about adversaries (though there are some); it's also about modeling what you find interesting or important. But that's pretty complicated too. Your one friend shared a woodworking project and your other friend shared travel photos. Which one(s) of those are you interested in? And when someone posts political stuff, is that something you find interesting, or is it something you prefer to keep separate from Facebook? There are a lot of different types of things people post, so the scope of figuring out what's important is pretty big.
This is far from a perfect analogy, but compare it to the problem of email spam. People first tried to fight it with client-side Bayes keyword filters. It turns out it wasn't nearly as simple as that, and to solve a problem that complicated, you basically need people working on it full time to keep pace.
Ranking and filtering a Facebook feed would have different challenges, of course. It's not all about adversaries (though there are some); it's also about modeling what you find interesting or important. But that's pretty complicated too. Your one friend shared a woodworking project and your other friend shared travel photos. Which one(s) of those are you interested in? And when someone posts political stuff, is that something you find interesting, or is it something you prefer to keep separate from Facebook? There are a lot of different types of things people post, so the scope of figuring out what's important is pretty big.