Invite only. If I'm running a federated service for me and my family, the moderation is going to be astronomically better than anything big providers are doing because a couple minutes of moderation by me every week is astronomically more on a per user basis than what someone like Google or Facebook is doing.
Drew's point is that people who can have or acquire that expertise are force multipliers — it's easy _enough_ that one person can run a node that carries less-technical users into the ecosystem. But I don't think we're really there yet, and worse the value isn't there in federated social media to offset the costs of time, money, and change.
We've seen distributed networks capture enough imagination and be usable enough to get mass adoption, but they did things like distribute tremendous proprietary media libraries on the cheap and without permission.
Services like Mastodon aren't as easy to spin up for a group as, say, running Napster was, or even seeding a torrent is now. But people, even non-techincal ones, either put up with those services' occasional unreliability or recruited more technical users to work them on their behalf because the relative value was so great.
So to me the problem with federation is making the value exceed the cost. It's a squeeze from both sides. We won't get there until an ActivityPub service is as easy to stand up as running an app on your device, AND the value of running one is large enough to make it worth the trouble.
It feels like messaging should be where the value end should have already taken off, but federated group chat is still a rat's nest of incompatible or feature-starved services. Federated social is still just on the wrong side of complexity and the value it provides honestly isn't that great to a mass audience.
Sandstorm is/was most of the way there; it makes installing server apps really easy (assuming a hosting provider installs Sandstorm for you).
Not for 99% of the users, for now. But there is an increasing faction that is taking issue with what any company does with their data.
At first, caring about one's own data was mostly a political thing. Now, abuse of data is getting so ridiculous that how much time you spend in the supermarket queue may depend on the aggregated data supermarkets have on you (and yes, supermarkets often aggregate their customer data via third-party reputation aggregator companies). Not to mention credit interest rates, insurance premiums, visas, and airport security checks. Soon people will too realize that taking their data off from ad networks is a good way to save money. At that point, one either goes silent or searches for a limited, well-protected social network.
I research in that direction: http://mro.name/ShaarliGo#install--update
Perhaps extend that one or two levels into the graph, with exponential decay. So if you invite someone, and that someone turns out to be a spammer, they get banned, you get banned, and I lose posting privileges for a week. And the person who invited me loses some privileges for a day. Etc. Add some level of manual review and algorithmic detection of clusters of accounts that exist only to invite other accounts that invite spammers, and...
... well, did anyone ever try this out? What are the obvious holes in this system that I'm missing?
It's worth not just "a shot", but lots of earnest attempts.
Another feature that I wish other places adopted is a public moderation log (https://lobste.rs/moderations) where users can see all the actions by the mod team. This helps keep the mods accountable.
Yes, protection against a million fake accounts is the key scientific problem to solve (Sybil attack).
Our university lab has worked on this problem for 15+ years, its hard. Early indications are that ActivityPub and federation are not going to scale. Bot attacks are too ferocious for amateur admins. Big instances attract more people, however a single instance fails to scale.
Nobody knows how to build trustworthy self-organising systems with "self-governance". Somehow you need to deduce the reputation of others from their signed interaction history. Their privacy needs to be protected, but you can still calculate a reputation score. After 20 years of scientific progress around "distributed reputation systems" with 100+ scientific articles they have one thing in commons: they don't protect from the Sybil attack.
Server federation is not going to help if there is no trust fabric to build upon. The effortless onboarding plus mobile-first requirement of today means you even need to have smartphones talking to eachother directly, P2P-style. <plug> thankfully students in our lab now have carrier-grade NAT traversal operational between Android devices. Google: "Technology Stack for Decentralized Mobile Services" (quotes required).
If you are, while i think email has proven that moderation is much much easier at scale if the attackers are spammers (astroturfing is a bit of a different story though)