Andrew_Rowe proposes a “respect” system where real humans earn respect votes and AI-generated content is down-voted, with higher respect granting users more powerful voting rights. This aims to encourage human interaction and reduce administrative workload, though he questions whether such a system could be gamed by spammers using sock puppets.
NateDhaliwal suggests leveraging the Post Voting plugin but raises concerns about AI accounts reaching high trust levels by gaming interactions. He notes that shifting promotion criteria to point collection could fundamentally alter the forum’s gamification dynamics.
nickdb argues that the line between human and AI is blurring, as many users rely on AI to draft posts. He believes a respect system would fail because it requires a shared value system that large online communities lack, and users may unknowingly act as “willing proxies” for AI content.
Ed_S views the issue as an economic “arms race” where both humans and AI agents will inevitably form voting rings. He recommends studying Slashdot’s metamoderation scheme as a historical example of tackling these challenges.
merefield explores using community network visualization tools to identify and mitigate coordinated voting rings or “cabals.”
Yee_Li (translating from Chinese) supports treating respect as a reference signal rather than a direct permission determinant to avoid manipulation. He suggests prioritizing sustained, cross-topic interactions over single-post votes to better identify genuine engagement. Yee_Li also asks how to prevent new members with low respect scores from being buried by the “chicken-and-egg” problem, ensuring their quality contributions remain visible.
The other topic which asked community administrators if they allow AI responses got me to thinking. With the new workflows feature and a few other goodies like a customized gamification plugin, it wouldn’t be hard to knock something together where real humans get respect votes from other users and AI gets down-voted. The more respect a user has, the more powerful the vote they cast, up or down.
The trust system built into Discourse is basically the same thing
I’ve seen first hand how the spammers try to game the system. They post an AI topic, then up-vote their own topic using another account. The AI spams itself
Would my ‘respect’ system just get gamed or do you think it could self-police the forum and encourage human interaction while relieving administrative workload?
Seems like the Post Voting plugin would work really well with this.
I wonder, have there been AI-powered accounts so much so that they like posts, intersct and reach TL3? Because then that could be a problem.
Perhaps then the solution is to change promotion criteria to amount of points collected. But that will be a fundamental shift in the whole system. It would promote a more aggressive gamification (literally) of the forum to earn points. But you know, aggressively posting good, non-AI content could also be a good thing…
The challenge is more humans are using an AI to form their thoughts. We see this in the growing numbers of people who cut and paste a topic into the composer - I actually have a bot that can query this metric so it adds another dimension to check to the “AI-ness” of a post.
When your humans become willing proxies and believe the digitally processed content they provide is their own, then a respect system will fall apart.
You need a common foundation, or value system, across all members in a forum for this to work, imo that just isn’t possible in any large group of people, let alone an online community.
I think sock puppets and voting rings are already a thing that humans do - and will inevitably become a thing the agents do, if they find it helps their goals.
I think it’s an arms race, and therefore not winnable, and it’s a matter of economics as to which countermeasures are worthwhile.
I might previously have mentioned Slashdot’s metamoderation scheme - I was always pretty impressed by that. It’s probably not too unlike a voting plugin, but possibly with extra mechanisms. Slashdot was an early popular site which needed to tackle these issues - it’s a bit of history we can learn from, if we so choose.
I find this direction really interesting, but I also agree with the concerns mentioned above: as long as weight can be exchanged for visibility, someone will inevitably find ways to game it.
My own approach would be more conservative: instead of directly tying weight to permissions, treat it as a reference signal. For example, pilot it on a small scale to see which interactions are sustained and cross-topic, rather than focusing solely on one-off votes on individual posts. Single votes can easily be amplified by cliques, whereas interactions over time are much harder to fake.
Another question I’d like to ask: if a new member starts with little weight, will their thoughtful replies actually be less likely to be seen? How would you avoid this “chicken-and-egg” problem?
