I’m more excited about looking at small online forums through the lens of institution design.
I think of virtual moderation as a central example of an institution that is relevant to online communities. What kind of institution do you have in mind?
The prediction itself could be done using machine learning, prediction markets, contractors who are paid based on the quality of their predictions, voting with informal norms about what is being voted on, etc.
Smart people are already reading LW—it shouldn’t be hard to gather their opinions as they read, and any attempt to approximate their opinions will likely be inferior
Those smart people’s judgments are the main data that is being used, we’re not trying to e.g. use textual features of comments to predict what smart people will think, we’re trying to use what smart people think to predict what some trusted moderation process would output.
I was responding specifically to the machine learning version of your proposal. My contention is that for a community that’s the size of LW, using machine learning to solve this seems a bit like killing a rat with a nuclear bomb. I also suspect that using an institution where all of the important decisions are made by humans is more likely to lead to knowledge that can reapplied outside of the context of an online forum.
I think of virtual moderation as a central example of an institution that is relevant to online communities. What kind of institution do you have in mind?
The prediction itself could be done using machine learning, prediction markets, contractors who are paid based on the quality of their predictions, voting with informal norms about what is being voted on, etc.
Those smart people’s judgments are the main data that is being used, we’re not trying to e.g. use textual features of comments to predict what smart people will think, we’re trying to use what smart people think to predict what some trusted moderation process would output.
I was responding specifically to the machine learning version of your proposal. My contention is that for a community that’s the size of LW, using machine learning to solve this seems a bit like killing a rat with a nuclear bomb. I also suspect that using an institution where all of the important decisions are made by humans is more likely to lead to knowledge that can reapplied outside of the context of an online forum.