I'll be interested to see how they leverage the user tree. I'm no expert in social media bots, but it seems like the user invitation tree could be important to identifying not just networks of bots that are generating low-value content and trying to amplify certain viewpoints, but the people inviting those bots into the network in the first place.
I think the point here is that if someone starts spouting rubbish, the person who invited them loses "face" - it reflects badly on them. So invitations aren't hard to come by, provided you have a reasonable reputation for saying things that are sensible.
This is a throwback to an early age when you would never talk to someone in social circles without first being introduced. If someone turned out to be a "bad sort" then the mutual acquaintance who had introduced you had proven themselves not to be reliable, and so you don't trust any future introduction from them. The scene in "Pride and Prejudice" where Mr Collins goes up to talk to Mr Darcy without having been introduced is a classic example of breaking these social norms, and the loss of reputation that ensues.
There is a lot of current research on "reputation" and "trust", and the "invitation tree" is one very simple way to provide some of the benefits.
That's the flip side, and part of the reason research on the question is active, ongoing, and occasionally contentious. It's a choice to make, and a mechanism to employ. One alternative is the HN model wherein everyone can contribute and you have no idea of their experience, expertise, or trustworthiness.
It's a choice, and it has consequences.
One thing I'd like to see added to HN is the ability for me to assign trustworthiness scores to other users, and then filter based on that. There are people here who I've come to know will provide value, even if their views are unpalatable, and their karma is low. There are others who have reasonably high karma, and yet who I know add little of value (for me!) to conversations.
Again, that would risk a filter-bubble, but it would reduce the number of comments I need to wade through before getting to those of value -- all from my personal view, of course, and your judgements would be different.
He's really good at that. I can't recall if he disclosed his methods. Most don't to avoid giving the people doing it a chance to adapt. Let them stay at low-effort attempts. :)
Users and mods have used it to detect sockpuppets and voting rings. We've had almost zero trouble with bots because of the invite system.
It's not obvious from browsing the site (though of course it's visible in the source: https://github.com/lobsters/lobsters/blob/791b21f60a2e952cab...) but users can lose the ability to invite if they reach a net score of -5 (rare, they have be a bad actor right out of the gate) or from a mod removing it (also rare, they typically have to invite several problem users or run a voting ring).
