First, it is always important to remember that representative democracy has known fundamental scaling problems from a game-theoretic perspective, and we can enjoy the consequences globally across the developed world today. Establishments and incumbents getting back to old good populism and enforcement, no matter left, right, or center. Representation works as a bootstrap mechanism, but it should be treated as something to be transformed into efficient direct democracy, otherwise we are doomed. And here we come to the practical "efficiency" of governing local communities. It seems especially important to discuss possible solutions here on HN, because much of the solution space clearly lies in tech.
So all in all, small communities do not fail because online participation is impossible. They fail because obstruction is cheap and accountability is absent: one person can repeatedly delay a decision, while everyone who supported a controversial decision simply moves on.
I think we are in that point in history where participation could be dramatically improved by LLM in pipeline (and there are real works behind it
https://github.com/AIObjectives/tttc-light-js
https://www.science.org/doi/10.1126/science.adq2852)
- Chat and voice messages becomes a source-linked with signed summaries (llm definitely can help most ppl structure their thoughts into words)
- disputes are classified as factual, distributive, value-based, procedural, or rights-related
- it becomes more real to agree on explicit success (or dealbreaking) criteria for every decision
- votes and objections also accounted as "prediction market" bets of their voting power.
- evidence is more reviewable with LLM
- and what's most important the outcome updates domain-specific reputation that weakens future vote weights of those who sabotaged, or joined to crowd for populist decisions proved to be failure.
If a blinded LLM assessment and the community votes agree, the decision proceeds. If they disagree, at least it is much rare occasion and interesting to consider the causes.
The key innovation is an auditable path from messy discussion to decisions, evidence, outcomes, and consequences. Reputation becomes the non-monetary cost of being confidently wrong or strategically obstructive.
I hope to see more of such platforms and survive experiments. Current situation feels like desperation
So all in all, small communities do not fail because online participation is impossible. They fail because obstruction is cheap and accountability is absent: one person can repeatedly delay a decision, while everyone who supported a controversial decision simply moves on. I think we are in that point in history where participation could be dramatically improved by LLM in pipeline (and there are real works behind it https://github.com/AIObjectives/tttc-light-js https://www.science.org/doi/10.1126/science.adq2852)
- Chat and voice messages becomes a source-linked with signed summaries (llm definitely can help most ppl structure their thoughts into words)
- disputes are classified as factual, distributive, value-based, procedural, or rights-related
- it becomes more real to agree on explicit success (or dealbreaking) criteria for every decision
- votes and objections also accounted as "prediction market" bets of their voting power.
- evidence is more reviewable with LLM
- and what's most important the outcome updates domain-specific reputation that weakens future vote weights of those who sabotaged, or joined to crowd for populist decisions proved to be failure.
If a blinded LLM assessment and the community votes agree, the decision proceeds. If they disagree, at least it is much rare occasion and interesting to consider the causes. The key innovation is an auditable path from messy discussion to decisions, evidence, outcomes, and consequences. Reputation becomes the non-monetary cost of being confidently wrong or strategically obstructive. I hope to see more of such platforms and survive experiments. Current situation feels like desperation