The base problem isn’t really solveable, and its as much of a public discussion on what we want to do with speech first, before its a question of how we want Social Media firms to act.
In the end, there is no algorithm which can match the scale of bad content, no robust definition of bad content which can work without creating a flood of false positives.
Every false positive is now someone who had something valid to say who is silenced.
How are we going to decide which grey area speech is unwelcome (leaving out obvious things that are illegal).
————
The popular idea is increased human centric moderation, but thats still going to be 2k email escalations for one region per day, at a 10% escalation ratio from a base of 20k.
It only appears unsolvable because you've presumed that social media should exist in its current form. Yes, algorithms can't match the scale of the bad content before we hit AGI. But that problem only exists because we have for-profit companies hosting way more content than they can afford to police on the thin margins ads provide. (Twitter, for example, makes about $1 per user per month.)
Prior to the late 2000s, this problem didn't exist. In alternate universes, it surely doesn't; there are many ways this could have gone.
Not OP. My impression is that a lot of the focus is around safety and specifically privacy. Both play right into social media giants' hands.
Instead we need to target Ads. Almost all problems can be eventually traced back to ads. At the very least traced back to the money incentive that ads create, on any platform.
I would go one step further and suggest that ads are an issue for certain types of Economic games or Markets.
Any industry that depends on Ads tends to consolidate, and has an issue of incentives - the more people on the network, the more likely the network is able to survive.
On a tech forum people assume that challengers have better tech - but I would argue that challengers actually allow for more salacious/engaging content.
This is what creates the race to the bottom.
If the race to the bottom can be stopped - i.e. an incentive structure created that stops engagement being the primary metric, then the rest of the downstream problems are largely prevented.
Thats my root cause assessment of the situation. However once I get to this point, any solution seems to be a mess of intersecting fields ranging from morality, legal constraints, issues with press freedoms, free speech etc. etc.
So... I guess how do we set up incentives to not allow the most "engaging" content to dominate?
In the end, there is no algorithm which can match the scale of bad content, no robust definition of bad content which can work without creating a flood of false positives.
Every false positive is now someone who had something valid to say who is silenced.
How are we going to decide which grey area speech is unwelcome (leaving out obvious things that are illegal).
———— The popular idea is increased human centric moderation, but thats still going to be 2k email escalations for one region per day, at a 10% escalation ratio from a base of 20k.