Hacker News new | past | comments | ask | show | jobs | submit login
Saving forums from themselves with shared hierarchical white lists (demon.co.uk)
39 points by alan-crowe on Nov 3, 2009 | hide | past | favorite | 26 comments

Maybe this is why Twitter works. You only see those you follow in your stream. You can't opt to see everything someone else is seeing, mostly because their platform can't really handle that sort of fan-out yet. But the basic idea of whitelisting is the same.

That's why Twitter's still fun for me regardless of how many social media consultants are around. I just don't follow them.

Exactly. This problem is why the friending / follower / newsfeed format has won. Everyone's accountable for curating their own feed. Want to expand the universe? Allow friend of friend, or friend of friend of friend views.

I'm imagining Outer Circle being implemented as a peer to peer application. In my example Albert->Carol->David. That is Albert is reading David's posts because Albert has white listed Carol (at level one) and Carol has white listed David.

How does the text of David's post get from David's machine to Albert's machine? We already have routing built into the semantics of the white listing hierarch. Since Carol is reading David's posts, Albert can get them from Carol.

I've read that one of the issues with amateur web comics is that good ones become popular and the artist suddenly finds himself with bandwith bills to pay. A peer to peer implementation of Outer Circle could cause this problem if an author becomes popular by millions of fans putting him on their direct lists, but I think that the common case is that a popular author reaches a large audience because he has been picked up by a hundred influention Circlers who are read at two removes by the Outer Circle. So the tree structure will usually give the system good fan out just because few users will wish to manage a direct list of more than 50 or 100 and will therefore access the system through several layers of hierarchy.

Have you considered PGP's web of trust?

PGP's web of trust shows that you can have a decentralized system without going full-blown peer to peer, and has many of the same issues with regard to trusting people and trusting people to trust people. It also shows that even this model - which is simpler than what you propose - has difficulty gaining acceptance.

I also don't see in what context you want to use this protocol, but there are a lot of fora that would be improved if only one had the ability to write Usenet-style killfiles (which can do a lot more than kill by author!). However, NNTP and SMTP clients largely already have these abilities to a sufficient level that you'll have a hard time displacing them, and web fora are very different and hard to script client-side.

I've heard the phrase "web of trust" and googled it, but found something very different to what I have in mind.

There is a tradition, going back to The Well, of having a serious adult forum invaded by children and their poo, pee, belly, bum posts. The adult forum resists and is destroyed.

My idea is that the different users are all legitimate and that success comes through coexistance. By coexistance I mean obliviousness. Chrome-Dome posts a cerebral contribution of great erudition. Chuckle-Head replies with vulgar mockery. In a traditional forum Chrome-Dome is driven away by this. The traditional response is to try to find a way to drive Chuckle-Head away instead.

The problem is that it is Chuckle-Head who is the popular figure, read and admired by hundreds of lame brains. Chrome-Dome is read and admired by dozens. So systems of voting and karma end up reflecting the fact that the mob greatly out numbers the elite.

The vision driving Outer Circle is that Chrome-Dome is happy with a white list of authors of some intellectual pedigree. If Outer Circle is a success, with millions involved, Chrome-Dome's, err, corner of the circle has thousands of authors, more than he can actually read, so he never gets round to expanding his white list to the point that he sees Chuckle-Head's post. Chrome-Dome is oblivious to the vulgar mockery around him and is not driven away by it.

Simultaneously, no administrator is banning Chuckle-Head, and Chuckle-Head is not retaliating by organising a DDoS of the site. Why would he? He has white listed his mates and is larking about, mocking the occasion boring adult he stumbles across, and enjoying acting his age.

The goal of Outer-Circle is the coexistance of incompatibles through invisibility.

Suppose that Middle-Brow, whom Chrome-Dome reads, comes across Chuckle-Head's vulgar mockery and replies to it, complaining that it is rude. We don't want this making it visible to Chrome-Dome. That would be a failure of invisibility and might drive Chrome-Dome away.

On the other hand, Chrome-Dome isn't reading all the replies to his posts. Ordinarily when Middle-Brow replies to a reply to one of Chrome-Dome's posts, the Outer-Circle software should promote the middle post to Chrome-Dome's attention, least there be a break in the threading. So the fine detail, of letting Middle-Brow control whether his replies do or do not implicitly endorse the comments to which they are attached, is actually essential to the overall plan.

I'm not convinced you understood what I was saying in the way I intended it to be understood, so let me clarify.

PGP's web of trust is designed to answer the question "is john@example.org who he says he is". A user is "valid" (in the above sense) if you have marked it as such, or if a sufficient number of users of sound judgement vouch for it. (There are multiple levels of "vouching", and the meanings of "of sound judgement" and "sufficient number" are user-configurable; and "users" are really "keys".)

By default, PGP believes that users who are vouched for as in the above model are valid but not of sound judgement (i.e. we don't let this users vouch for others), but you could easily change that.

The point I was trying to make is that, if you replace "PGP" by "Outer Circle" and define "valid" to be "worth reading", the above sounds a lot like what you want to achieve. It has also been deployed in the real world for quite a while, so you can see to what extent it has worked (it's rather succesful in a very small part of the population).

Additionally, PGP is effectively decentralized, even though day-to-day operation uses keyservers. This model is a lot easier than full peer-to-peer (which, in the presence of NAT, is probably not possible without centralized servers anyway), and a combination of clever crypto and fallback options means that PGP does not rely on keyservers. Look at the documentation for more details.

Notwithstanding the above, which is honestly intended to be good advice, I don't see how your system is ever going to become sufficiently widespread to be worth learning for the average user, and I don't really see the value in keeping a medium that's 99% nonsense usable for the last few erudite users - just let them migrate.

Thank you for expanding on that. In trying see the relevance of PGP's web of trust to Outer Circle, two things throw me off the scent.

First, identity in Outer Circle is simply a public key used to decode each signed post. When an identity is highly regarded, others will wish to impersonate it, in order to get their writing or advertisements read, but they are unable to do so because they do not know the private key.

Prevention of passing off is the whole of identity in Outer Circle. We do not care if users have multiple accounts nor do we care whether group accounts are acknowledged or pretend to be an individual. Circle jerks, be they conspiracies or sock puppets, do not undermine the Outer Circle system because an identity, even in good standing, has no right of recommendation. If some-one on your direct list starts reading crap you "shallow" him or drop him, and that is part of the normal operation of the system as tastes change (his or yours) or time pressures squeeze.

So it is confusing to me to read about web of trust because Outer Circle has no notion of who somebody really is. Ossian is welcome.

Second is Outer Circles core concept of the coexistence of incompatible through invisibility. Whether somebody is who he says he is is a valid concept that is either true or false. I see the idea that a post is worth reading or not worth reading is a trap. There is no truth of the matter and that is the heart of the reason that traditional forums with moderation and meta-moderation collapse. They are trying to forge a consensus when they should be facilitating disagreement.

You doubt the value in keeping a medium that is 99% nonsense usable for the last few erudite users. That foregrounds the issue of scaling. I picture Outer Circle growing from a thousand elite users to million users, 99% hoi polloi 1% elite. So the proportion of erudite users falls, but the absolute number rises tenfold. I see the endless migration of the nomadic elite as an important problem, preventing them reaching critical mass.

Do you still think that PGP's web of trust is relevant? How do I relate "is john@example.org who he says he is", which has type boolean, to "john is worth readng", which has type function from users to boolean?

Hmm, you had not previously expanded on the prevention of impersonation, although that is indeed desirable. In fact, piggybacking this whole thing on PGP (with special "is interesting" messages) may work. Somewhat.

I don't see why the web of trust-approach would fail. Certainly, I admit, "is worth reading" is not (intended to be) a global value, while "is really that person" is.

But ultimately the web of trust answers the question "do I trust that person to be who (s)he says (s)he is". For instance, someone who trusts a lot of people is likely able to verify that key D5327CB9 belongs to Wietse Venema, while someone who has just created a key but doesn't yet trust anyone cannot. Therefore, while this key belongs to Wietse or does not belong to Wietse irrespective of who asks this question, the answer may differ.

There are obvious parallels to "do I trust this person to produce interesting content/tag appropriately".

I can understand your desire to bring the elite together, but what is the big added value over, say, Twitter or blogs? These are fairly stable, fairly reliable identifiers of people, with connections between them. Filtering out the nonsense is so easy that it's hard to notice that we're even doing it, and "clusters" of smart people form more or less automatically. (Which is not to say that either Twitter or blogs are the best possible communication medium, especially for small messages like this, but they seem to get this mostly right.)

What is the big added value? I was prodded into making another attempt to bring coherence to the swirling mass of ideas I call Outer Circle by a recent thread here on Hacker News: http://news.ycombinator.com/item?id=916448 Other people consider the forum problem important.

When Kuro5hin died, I started reading The Edge, http://www.edge.org/. However The Edge is what I call an Inner Circle site. Outsiders get to spectate, but there is no way for outsiders to ask questions or raise objections. The Edge seems introverted to the point of sterility. My dream for Outer Circle is that it should impose a wholesome discipline on outsiders. They are not confined to spectating but can ask questions and raise objection. On the other hand outsiders have to work for access, playing nicely, and earning a place on insider's white lists.

The big win happens when you have people you don't know making unexpected positive contributions. The hard question is how to permit that without suffering the usual forum problem.

I understand that people believe this to be important, and I see the value in interesting communities. It's just that I don't consider the forum model something to emulate, but rather a regression from earlier models - I much prefer mailing lists or Usenet groups, both of which allow a user much more control via their client, and both of which seem to, on average, suffer less from "the idiot problem" than their web-related counterparts.

Part of the reason may have been that many web fora try to appeal to the broadest possible audience, while many mailing lists and Usenet groups have a formal or informal charter.

This is interesting but whitelists/blacklists worry me because newcomers will either see everything or nothing, both of which are pretty unfriendly.

If it spreads by word of mouth, you can use your friend's white list. If you see everything you can pick the best of a dozen posts and then use the authors white list. Once you have a human written white list to start from you can manage the volume by controlling the depth that you descend the search tree, and start adding authors you like to your direct list. Every time you add an author to your direct list you chose a number to set a search depth with him as the tree root and his direct list as the first level, so I think you quickly take control and adjust the list to your taste.

I'm more worried about how a new author gets read. Nobody sees his comments because he is on nobodies white list. I guess there has to be a slush pile. When people get bored they can turn to the slush pile and maybe find a gem. Alternatively one may wish to see all the replies to ones own posts, not just those from the authors on ones white list. If that is common practice it gives a new author an audience of one simply by replying.

I think this is a very serious possible problem with this sort of system, but there are ways around it. The system could take everyone's whitelist and derive a "public whitelist" that it can use to filter articles and comments for people that are new to the site and have unknown preferences. This "public whitelist" could be derived from an average of everyone's whitelist, with other heuristics thrown in. As the system learns what type of posts a new user prefers their whitelist is updated.

One can also imagine how a user could use a whitelist on their own posts to maintain a separation between the things they would say in public vs. their private thoughts (which others may find offensive or unpopular)

I don't know if I'm just lazy, but there is no way I would use something like this.

Yes, but it would save everyone else from having to read your drivel.

Can you honestly see yourself ranking people based on how many levels of the people they ranked you want included in your white list? Do most people even use the filtering capabilities available today? I don't think I've ever blacklisted anyone on a forum, even if they drove me absolutely insane.

Can you honestly see yourself joining a knitting community?


That's different. People join knitting communities because they enjoy knitting. Very, very few people enjoy rating things just for the sake of rating them. People join knitting communities, but not rating communities. People rate as a means to an end. The end has to be worth the hassle of the means. This rating system sounds substantially more complicated than anything in popular use today, and for me, even what's in use today isn't worth it. I don't think I'm in the minority in not using blacklists and similar capabilities. If people won't even do that, how will you get them to go through some of the complications of this system?

Is there a way that hierarchical whitelisting could be derived from up/down votes like on HN and reddit?

For a user of which you have voted on many posts, we can define "interesting" directly. If we use I(you, u) to denote our proxy of the chance that you find user u interesting, we define I(you, u) = (number of upvotes to this user) / (number of total votes on all users).

For any user u on which you have not cast a lot of votes, calculate I(you, u) as the sum of I(you, u') I(u', u) over all users u' (other than you or u), plus the above formula.

There are some issues with this: it'll take some work to make it perform, and it strongly favours established accounts. But if you can solve the first issue, you can probably live with the second - karma has the same problem.

If so, then that would be really cool. I'm just saying I don't think the average user will do it on their own.

You are not asking when and why.


Forums have a life cycle. They start small, full of interesting people. Hoi polloi find them and join, to spectate and snark and lark. The signal to noise ratio falls and the interesting people leave.

Forums self limit by drowning in mediocrity. If some mechanism could prevent this a forum could grow larger and have more interesting people on it. There may well be a critical mass of interesting people that no forum has reached yet. Failing that it would still be a big win if a forum could persist. Having to move on and find somewhere else is a pain and people would use a complicated system if it actually delivered on the promise of keeping the signal to noise ratio up. What is at issue is the individual signal to noise ration, what you see given your white list. It doesn't matter if the forum as a whole goes down the tubes as long as the interesting people never find out. It also doesn't matter if people disagree about what is signal and what is noise, each to his own white list.


Outer Circle starts small. So participants can simply use the ALL option, and not bother with rating. However, the Outer Circle client should offer each user vote up and vote down puttons. These are private. As the forum grows and ages, two things happen: there is plenty to read, the signal to noise ration falls. Each participant has a breaking point at which he asks the system to use the his personal voting tallies to auto-build him a white list.

Perhaps it should include those with positive totals and leave out those with negative totals. That could be done with putting all the positives on level zero, but if the forum is thriving the list will be thousands long. No. It is important that the auto-build takes some time to discover some structure so that it can build a hierarchical list.

There is a social science hypothesis lurking here. Outer Circle assumes that the authors you rate very highly will read authors that you rate highly and that there is a clustering of preferences sufficiently tight that hierarchical sharing of white lists becomes an efficient way of constructing and managing large white lists.

There after rating is pain-driven. When the signal to noise ration that you experience drops too low you think to yourself "Outer Circle is dying, where do I go to instead?" Then you remember your white list. You have continued to upvote and downvote, it is a habit, even though it doesn't seem to do anything. Now you can ask your Outer Circle client why the forum has become crap and it can tell you the paths by which you see the articles and suggest deepenings, shallowings, and directings to improve your experience.

I think that there is a threshold effect. If the rating system is elaborate enough to keep the forum from dying it is worth the hassle, but a slightly simply system, which merely slows the decline and thus prolongs the agony is worthless.

You assume that online communities die because there are too many (sub-)mediocre people. I admit that this is a reasonable hypothesis that explains part of what can be observed, but it is a rather big leap to get to "a community will not fail as long as its populated by the elite". Even the elite has a finite attention span, after all, and any sufficiently large community is likely to splinter into sub-communities anyway. Either that, or people have to stop posting all but their best posts - and that does not seem likely.

I think the take-away point is that this system would provide a way to have (one or more) communities within the community. The various subcommunities would overlap and cross-pollinate, so to speak, but if there were a small community of serious thinkers, they'd generate interesting talk on their own. This doesn't exclude the larger community - in fact, it permits the existence of a larger community to continue, and occasionally to cross over into the inner community.

I'm not expressing this well, on reading back over it. But I think you're miscontruing the point. It's not that online communities die because there are too many sub-par people; it's that that many sub-par people dilute the original community and everybody throws up their hands and finds another place. This allows the original community to persist amid the din.

'S perfect. I like it.

I was specifically responding to what I construed as "this will allow a critical mass of smart people to emerge/collaborate": due to people's finite attention span and inability to keep track of too many different people, there is a natural limit to the size of a community anyway, and it is possible that "elite" communities run into this bound before they achieve critical mass (which is rather nebulous anyway, but I interpret this as "genius happens".)

I'm really not trying to misconstrue your point, BTW; if it didn't interest me, I'd have stopped responding long ago, and I have better things to do than trolling.

This is an excellent explanation. I'm not 100% convinced, but I'm definitely more convinced than I was before.

Applications are open for YC Winter 2022

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact