the spam/bot problem is real but i think the more subtle challenge is keeping quality high even with all real humans. most online discussions degrade not because of bots but because the incentive structure rewards reactive emotional responses over thoughtful ones.

what's interesting about polis's approach is that it surfaces agreement clusters instead of amplifying disagreement. most comment systems optimize for engagement, which in practice means conflict. if you optimize for "where do people actually agree despite appearing to disagree" you get a completely different dynamic.

the invite-tree idea someone mentioned below is interesting for the same reason: it's not just that it keeps bots out, it's that it creates social accountability. you're more thoughtful when your reputation is linked to the people you invited. same principle as why small communities tend to self-moderate better than large ones.