You're just shifting around your trust problem. You need to handle 4chan level manipulation (million of users coordinating to manipulate polls), or Scientology depth (getting thousands of people in to USA government jobs in order to get recognised as a religion). If it's "we'll catch it in moderation" then whoever wants to manipulate it just gets a moderator ...
"Super-moderation": will a dictatorship work here? I don't see how.
"Meta-moderation": you're back to bad actors manipulating things with pure numbers.
But think of how we solve this problem in our personal interactions with other people, and this should be a clue for how to solve it with computational help. We have a pretty good idea of which people are trustworthy (or capable, or dependable, or any other characteristic) in our daily lives, and based on our interactions with them we update these internal measures of trustworthiness. If we need to get information from someone we don't know, we form a judgement of their trustworthiness based off of input from people we trust--e.g. giving a reference. This is really just Bayesian inference at its core.
We should be able to come up with a computational model for how this personal measure of trustworthiness works. It would act as a filter over content that we obtain. Throw a search engine on top of this, sure, but in the end you'd still need to get trustworthiness weights onto information if you want it to be manipulation-resistant. This labeling is what I mean by manual curation. You can't leave that up to the search engine or the aggregator because those can be gamed, like the examples you gave for aggregators and SEO for search engines have shown.
We really don't. People get surprised all the time that someone had an affair, or cheated, or ripped someone off, or whatever. "But I trusted you" ...
It's actually relatively easy to fool people in to trusting you, as many red team members will probably confirm.
Look at someone like Boris Johnson, people are trusting him to lead the country knowing that he's well known to betray people's trust and that he even had a court case lodged against him based on his very blatant lying to the entire country. You can even watch the video of him being interviewed where the interviewers says (paraphrasing) "but we all know that's a half truth" and BoJo just pushes it and pushes it and refuses to accept that it's anything other than absolute truth.
>If we need to get information from someone we don't know, we form a judgement of their trustworthiness based off of input from people we trust--e.g. giving a reference. //
This is domain authority again - trust some domains manually, let it flow from there. If that domain trusts another domain then they link to it, trust flows to the other domain, and so on. Maintaining such trust for a long time adds to a particular domains trust factor, linking to domains not trusted by others detracts from it.
>This is domain authority again - trust some domains manually, let it flow from there. If that domain trusts another domain then they link to it, trust flows to the other domain, and so on. Maintaining such trust for a long time adds to a particular domains trust factor, linking to domains not trusted by others detracts from it.
This can be gamed if you're able to update the trustworthiness of a domain for other people, and that's why a trust metric needs to be mostly personal, and should update dynamically based on your changing trust valuations.
Seriously, I'm not so sure -- I try to trust first and then update that status as more information becomes available; but that's more of a religious position.
I don't think it's necessarily instructive to look at my personal modes here. I guess my main point is that if you're going to say "well humans have cracked trust, we'll just model it on that" then I think you're shooting wide of the mark.
