You could blacklist everyone by default, then personally contact individual "smart" people and invite them to your darknet. The price of admission is to blacklist everyone but yourself and the people you personally approve of. When someone breaks this policy, you could just blacklist them again.
The first step (blacklisting everyone except me and people I approve of) is easy. Expanding network depends on other people joining the same system, or at least willing to send me a list of people they approve of. I think that most people use default settings, so this system would work best on a site where this would be the default setting.
It would be interesting to find a good algorithm, which would have the following data on input: each user can put other users on their whitelist or blacklist, and can upvote or downvote comments by other users. It could somehow calculate the similarity of opinions and then show everyone the content they want (extrapolated volition) to see. (The explicit blacklists exist only to override the recommendations of the algorithm. By default, an unknown and unconnected person is invisible, except for their comments upvoted by my friends.)
If you implemented this, I'd even be interested in trying it out, provided that I could see the two scores (smart-only as well as all-inclusive) side-by-side.
If the site is visible for anonymous readers, a global karma is necessary. Though it can be somehow calculated from the customized karmas.
Personally, though, I would prefer to err on the side of experiencing negative emotions now and then. I do not want to fall into a death spiral that leads me to forming a cabal of people where everyone agrees with each other, and we spend all day talking about how awesome we are -- which is what nearly always happens when people decide to shut out dissenting voices.
I also wouldn't like to be shielded from disagreeing opinions. I want to be shielded from stupidity and offensiveness, to protect my emotions. Also, because my time is limited, I want to be shielded from noise. No algorithm will be perfect in filtering out the noise and not filtering out the disagreement.
I think a reasonble approach is to calculate the probability of "reasonable disagreement" based on the previous comments. This is something that we approximately do in real life -- based on our previous experience we take some people's opinion more seriously, so when someone disagrees with us, we react differently based on who it is. If I agree with someone about many things, then I will consider their opinion more seriously when we disagree. However if someone disagrees about almost everything, I simply consider them crazy.
I think that most people use default settings, so this system would work best on a site where this would be the default setting.
I think this is a minor convenience at best; when you choose to form your darknet, you could simply inform the other candidates of your plan: via email, PM, or some other out-of-band channel.
It would be interesting to find a good algorithm, which would have the following data on input...
This sounds pretty similar to Google's PageRank, only for comments instead of pages. Should be doable.
...If the site is visible for anonymou
EDIT: Thanks to people not wanting certain words google-associated with LW: Phyg
Lesswrong has the best signal/noise ratio I know of. This is great. This is why I come here. It's nice to talk about interesting rationality-related topics without people going off the rails about politics/fail philosophy/fail ethics/definitions/etc. This seems to be possible because a good number of us have read the lesswrong material (sequences, etc) which innoculate us against that kind of noise.
Of course Lesswrong is not perfect; there is still noise. Interestingly, most of it is from people who have not read some sequence and thereby make the default mistakes or don't address the community's best understanding of the topic. We are pretty good about downvoting and/or correcting posts that fail at the core sequences, which is good. However, there are other sequences, too, many of them critically important to not failing at metaethics/thinking about AI/etc.
I'm sure you can think of some examples of what I mean. People saying things that you thought were utterly dissolved in some post or sequence, but they don't address that, and no one really calls them out. I could dig up a bunch of quotes but I don't want to single anyone out or make this about any particular point, so I'm leaving it up to your imagination/memory.
It's actually kindof frustrating seeing people make these mistakes. You could say that if I think someone needs to be told about the existence of some sequence they should have read before posting, I ought to tell them, but that's actually not what I want to do with my time here. I want to spend my time reading and participating in informed discussion. A lot of us do end up engaging mistaken posts, but that lowers the quality of discussion here because so much time and space has been spent battling ignorance instead of advancing knowledge and dicussing real problems.
It's worse than just "oh here's some more junk I have to ignore or downvote", because the path of least resistance ends up being "ignore any discussion that contains contradictions of the lesswrong scriptures", which is obviously bad. There are people who have read the sequences and know the state of the arguments and still have some intelligent critique, but it's quite hard to tell the difference between that and someone explaining for the millionth time the problem with "but won't the AI know what's right better than humans?". So I just ignore it all and miss a lot of good stuff.
Right now, the only stuff I can be resonably guaranteed is intelligent, informed, and interesting is the promoted posts. Everything else is a minefield. I'd like there to be something similar for discussion/comments. Some way of knowing "these people I'm talking to know what they are talking about" without having to dig around in their user history or whatever. I'm not proposing a particular solution here, just saying I'd like there to be more high quality discussion between more properly sequenced LWers.
There is a lot of worry on this site about whether we are too exclusive or too phygish or too harsh in our expectation that people be well-read, which I think is misplaced. It is important that modern rationality have a welcoming public face and somewhere that people can discuss without having read three years worth of daily blog posts, but at the same time I find myself looking at the moderation policy of the old sl4 mailing list and thinking "damn, I wish we were more like that". A hard-ass moderator righteously wielding the banhammer against cruft is a good thing and I enjoy it where I find it. Perhaps these things (the public face and the exclusive discussion) should be separated?
I've recently seen someone saying that no-one complains about the signal/noise ratio on LW, and therefore we should relax a bit. I've also seen a good deal of complaints about our phygish exclusivity, the politics ban, the "talk to me when you read the sequences" attitude, and so on. I'd just like to say that I like these things, and I am complaining about the signal/noise ratio on LW.
Lest anyone get the idea that no-one thinks LW should be more phygish or more exclusive, let me hereby register that I for one would like us to all enforce a little more strongly that people read the sequences and even agree with them in a horrifying manner. You don't have to agree with me, but I'd just like to put out there as a matter of fact that there are some of us that would like a more exclusive LW.