-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Michael Rogers wrote:
> filtering out automated spam. I've given this a hell of a lot of thought, and here's my best guess as to what we could do about it. "Posting a message on a newsgroup" is itself an action, albeit an unspecific one. But we can leave certain things unspecific, while still calling it a category of trust. I'm not sure how much lack of specificity can go into an action, or if actions themselves aren't aggregates of some other more fundamental unit of trust, but here's how I would handle spam. I'd have a category of trust something along the lines of "Posting a message on any newsgroup that is non-disruptive." There would be a second category for "Measuring the trust of people to be non-disruptive," that would allow for trusting people to decide the trust of others, to varying degrees. Individuals could decide which messages qualified under the first category, but they could also qualify other individuals under the second category. If there are 1000 automated spam messages and 100 posters, but 1 poster is diligent enough to only mark the valid posts, then that means only 1 person needs to look at 1000 spams. Without such a trust network, 100 people have to look at 1000 spams for 100,000 wasted views. Or you could divide the unread messages among 10 people, or any number of different strategies. And should one person attempt to falsely categorize a message, others who checked that message can catch them on it, and better define the topology of the trust network to exclude the now untrustworthy member. I think if we collectively and consistently approve signers of messages that are not disruptive to a newsgroup, mailing list, or other place of discussion (this would include off-topic ranting on groups with a topic), then that is exactly what we need to do to maintain a group's discussive cohesion, no more than that is necessary, and no less than that is sufficient. This is not to say that you can only trust other human beings. The 1 person reading through all 1000 spams would themselves trust a spam checker program to eliminate obvious spams while minimizing false positives. But that spam checker would only be one node in the trust network, so even if a message (falsely) registered as spam under it, another human being could vouch for the signer of that message, and have the message get past the spam filter to the moderator for verification that it was a false positive. That reduces load on anyone who has to check for spam messages, while also allowing for ways to remedy false positives in a bayesian filter, but it does not reduce any load or labor of the spammers. Thus even with a simple 1 person watching out for spam, this group could be spam free unless a spammer was 1000 times more powerful than that one person. You could even conceivably pay people to find non-disruptive messages and approve them, ala "certificate authorities" but without that nasty hierarchy. > Second, can we solve the problem of carrying over trust from one > category to another without an a priori hierarchy of categories? My best idea for that would be to have people who sign "relations" between categories, to allow trust from one to bleed into trust from another. Someone could mark "peanuts" and "peanut butter" as both related, so if you went looking for a trustworthy source of peanut butter, you might possibly find a source of peanuts as well. Trust would be lost over each relation of course... so unless you had a really crappy selection of peanut butter, you probably wouldn't get any people selling peanuts. But if it should be the case that all your peanut butter is filled with salmonella, you would want to know trustworthy peanut sellers so that you can make your own damn peanut butter. So "Relating two categories" would be itself a category of trust, or even "Relating two categories involving legume food products." Again I'm not sure how specific it needs to be. But someone you trusted in that category (or someone you trusted to trust people in that category) would be able to provide you with the information you need to connect one concept to another, and one source of delicious legumes to another. > Maybe > we could use something like collaborative filtering to detect similar > categories Again, individuals in this trust network could be automated machines, lexical parsers and the like. Their trustworthiness would be only as good as their programming though I can assure you. > But to > the extent that such an approach reduces the number of explicit trust > ratings I need to make, Yeah that's pretty important. Can't have everyone doing peanut butter inspections after all. It reminds me of one of those old "roleplaying" games I played once. You could choose a specific kind of sword to specialize in, a saber, a rapier, a broadsword etc, or you could choose a category of weapon to use, like one handed sword. Someone using a one-handed sword as a skill would take longer to get equally good at a rapier than someone using rapier as a skill, but the former person would be able to use a saber without killing themselves. Surprisingly though many people were in favor of the "generic" categories, as they had no way of predicting what would pop up in the treasure chests. So even though they had less skill than the specialists, they had a better chance of success because they were more likely to get a weapon they could wield. In the same way most people would just worry about general trust categories, and probably defer their trust over to a professional at that stuff, and the more specific trust categories would be used by the few who need to distinguish between those things. > Third, you mention the meta-category "recommending trustworthy people". > Is this actually a single category, or does the assumption that trust is > category-specific mean that I need to know who's good at recommending > trustworthy people in category A, category B, etc? Yeah, not entirely sure sorry. :/ There are two possibilities though. Either "action A" and "refer people who do action A" are two separate categories, or "action A" itself has some kind of a "maximum depth" property through which you can recurse. In most cases it does make sense that a plumber will be able to tell you who other good plumbers are, but on the other hand a plumber might not want you knowing about the competition. So just because you trust someone to be a plumber doesn't mean you trust them to recommend a plumber. So I'm not sure how to handle it, or even how to word it properly. You're right though, there can't be only one single "recommend trustworthy people" category. I could tell you an awesome local Mexican restaurant, but I couldn't tell you if there are any good Chinese places around here, so it would be silly to trust me to recommend all restaurants, and it'd be even sillier to trust me to recommend everything, including medical personel and reliable prostitutes. If there was a "trustworthy for anything" category, I can imagine that few people would have a very high level of trust in that for anyone. Some people though, you can pretty much trust for anything, or at least they'll let you know if they can't help! -----BEGIN PGP SIGNATURE----- Version: GnuPG v2.0.14 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iEYEARECAAYFAkuPaDgACgkQB/meY5RuPPQg3gCgvD8IQlug99hCvaMQmJ7gilKf GYQAoJ6yhokLrIFI7DPwEnR3mhs/F3Fr =PVlb -----END PGP SIGNATURE----- _______________________________________________ p2p-hackers mailing list p2p-hackers@lists.zooko.com http://lists.zooko.com/mailman/listinfo/p2p-hackers