RE: signal to noise proposal
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Aimee wrote: > > To wit, no two people can safely tell the same lie to the same person. Bah. I say it depends entirely on what the lie is, who's being lied to, and how confident and artistic the confidence artists are. Choate: > Actually they can, only one (or both, if we allow 3 or more agents, only > one is required to 'know' the lie) of the people must believe it is the > truth. If they were good enough (and their targets comfortable enough), all three could be lying their asses off about anything and nobody would ever be the wiser. Likewise, with three or more targets playing it the other direction. >Well, I doan' kno' nuttin' 'bout no agents. That fact has been established. Careful parsing is the spice of life... :P >But, you know, after pondering on that a bit...What if "the lie" was >supposedly "really secret stuff?" >You know, "ME LUCKY CHARMS!" >I know the little boys and girls are after me lucky charms. >If "3 or more agents" happen to run in the door with me lucky charms, Sounds about right. >that might smell really fishy to some people since leprechauns are hard to >catch. Somewhere over the rainbow. >Furthermore, if you ask them about these lucky charms in isolation, they >better know the lucky charms like the back of their hand, or further >investigation is likely to review not-so-lucky inconsistencies. The >"knowing" part can be rendered irrelevant by context, indeed it is >sometimes imperative that everybody KNOW so as to provide...uhm.secondary >alternative consistency. But what about when the unlucky charmers find they're actually the victims of a deceivers-deceiving-the-deceivers-deceiving-the-deceivers kind of thing. What shows that the snowers know they've slowly been snowed? Bet it keeps a lot of people awake at night, that one. Tricky, but fascinating. If anyone knows of any good links to counter-deception detection, drop me a line. Not sure how "on topic" it is, but something everyone here would do well to read about. Either that, or just default to not trusting anyone, ever. Works for me. >And, "lucky charm lies" can take many forms, including physical, which might >be subject to verification, additional investigation and other stuff I don't >want to happen to me lucky charms, because I might want the enemy to believe >they are TRULY "lucky," "charmed," and "mine." >I'm sure "it depends," but perhaps that wisdom came from just such a >situation. Oh really? *blink blink* like what? Didn't work, huh...damn, better go brush up on my social engineering. LOL ~~Faustine. *** The enemy resembles us. Therefore, he needs to be approached not as an assembly of 'targets' to be destroyed one by one; but as a living, intelligent entity capable of acting and reacting. - --Martin Van Creveld -BEGIN PGP SIGNATURE- Version: PGPsdk version 1.7.1 (C) 1997-1999 Network Associates, Inc. and its affiliated companies. (Diffie-Helman/DSS-only version) iQA/AwUBPJ/o8vg5Tuca7bfvEQIgLACeNITrMr+sYCicyHWyVw3anqpkwY4AoI8a KK/Xi3yMBzfoRiCzqjuiYtbK =+eZw -END PGP SIGNATURE-
RE: signal to noise proposal
> > To wit, no two people can safely tell the same lie to the same person. Choate: > Actually they can, only one (or both, if we allow 3 or more agents, only > one is required to 'know' the lie) of the people must believe it is the > truth. Well, I doan' kno' nuttin' 'bout no agents. That fact has been established. But, you know, after pondering on that a bit...What if "the lie" was supposedly "really secret stuff?" You know, "ME LUCKY CHARMS!" I know the little boys and girls are after me lucky charms. If "3 or more agents" happen to run in the door with me lucky charms, that might smell really fishy to some people since leprechauns are hard to catch. Furthermore, if you ask them about these lucky charms in isolation, they better know the lucky charms like the back of their hand, or further investigation is likely to review not-so-lucky inconsistencies. The "knowing" part can be rendered irrelevant by context, indeed it is sometimes imperative that everybody KNOW so as to provide...uhm.secondary and alternative consistency. And, "lucky charm lies" can take many forms, including physical, which might be subject to verification, additional investigation and other stuff I don't want to happen to me lucky charms, because I might want the enemy to believe they are TRULY "lucky," "charmed," and "mine." I'm sure "it depends," but perhaps that wisdom came from just such a situation.
: CDR: Re: signal to noise proposal
>>All these mental efforts are not cost effective. Tens of people spending hours on filtering/rating schemes is expensive. Choate's whereabouts are well known, deploying persuasion contractors will cost a fraction of the proposed engineering efforts. Think of it as of simulation run of AP. Just my 5000 cents.<< Count me in for 5 as well and all,APster negative reputation capital,RUST never sleeps.
RE: signal to noise proposal
Tim wrote: > A lot of the current/recent "reputation schemes" make a fundamental > mistake: they attempt to assign a scalar value to "the" [emphasis] > reputation of an actor. Even the schemes which attempt to assign a > vector rating, e.g, "Declan' s rating of Detweiler is..., Tim's rating > of Detweiler is...," make a fatal mistake. Sure seems so to me. > There are no reputations attachable to actors in this way. Different signals are salient to different receivers, and more so when a group departs from targeted goal-seeking, that being determinative of saliency. Salience is not based on reality, but upon congruence with a perceptual framework and mind-set. Based on my observations, anything that does not fit within that framework is treated as a hostile signal. Groups end up engaging in nothing more than perceptual filtering. The result is often high "sprignal" instead of high "signal." (Recognizing the blur between your concept of signal and mine.) You are a maximum entropy channel, probably in part due to the increasing discord among goals, capabilities and perceptions, but also because of the interactions that you allow, high receptivity, etc. Most reputation schemes fail to separate the concept of source reliability and accuracy, much less take in considerations of coherence, credibility or intrinsic confirmation. It strikes me as destructive. (A high channel error rate is suggestive of a manipulated/corrupted channel, rather than a "good moderation system.") The biases involved with "past behavior" ratings are many, but some strike me as particularly relevant: (1) people tend to attribute the behavior of another person based on their "nature" or "character" -- rather than the situation. In contrast, we judge ourselves by situational factors. (2) probability estimates are affected by recall and like events. (3) anchoring bias. This is not suggest that "noise" reduction strategies are inherently bad, just to maybe stimulate some thinking in brighter minds than mine. And earlier > The familiar saw about two people being able to keep a secret...if one of them is dead. To wit, no two people can safely tell the same lie to the same person. ~~Aimee
Re: signal to noise proposal
On Sun, 24 Mar 2002 [EMAIL PROTECTED] wrote: > ahem, yes I am aware any simple system is easily > circumvented & defeated, but that doesnt imply > that it will be. > > Ive noticed many objections to any new proposal often > take the form, "but that would be different > than what we have now!!!" wow, amazing, no kidding!!! > > I can come up with all kinds of objections to > my own proposal, basically identical to what > everyone else wrote. > > ok, fine, status quo stays the same :p > lets just gripe,bitch,moan to the list > for another few years. wheee > I thought things might be different after > a half decade of cyberspace lightning, but > so nice that some things just dont change. > > yes, its cypherpunk stalemate as usual. > I fully agree with TCM. who writes about it every > few weeks for the last ten years. hahahaha > cypherpunk == grandiose ideas, no execution. > 99% inspiration, 1% perspiration hahahaha Don't talk about the solution then. Just do it. Cypherpunks write code. Patience, persistence, truth, Dr. mike
Re: signal to noise proposal
All these mental efforts are not cost effective. Tens of people spending hours on filtering/rating schemes is expensive. Choate's whereabouts are well known, deploying persuasion contractors will cost a fraction of the proposed engineering efforts. Think of it as of simulation run of AP. Just my 5000 cents.
Re: signal to noise proposal
ahem, yes I am aware any simple system is easily circumvented & defeated, but that doesnt imply that it will be. Ive noticed many objections to any new proposal often take the form, "but that would be different than what we have now!!!" wow, amazing, no kidding!!! I can come up with all kinds of objections to my own proposal, basically identical to what everyone else wrote. ok, fine, status quo stays the same :p lets just gripe,bitch,moan to the list for another few years. wheee I thought things might be different after a half decade of cyberspace lightning, but so nice that some things just dont change. yes, its cypherpunk stalemate as usual. I fully agree with TCM. who writes about it every few weeks for the last ten years. hahahaha cypherpunk == grandiose ideas, no execution. 99% inspiration, 1% perspiration hahahaha q. whats the difference between a group of cpunks & a group of arbitrary people chosen at random??? a. the arbitrary people OCCASIONALLY AGREE WITH EACH OTHER!! they also OCCASIONALLY WORK TOGETHER!!! hahahahha just for my own amusement I may write a quick perl script to do some of the basic statistics I suggested over a few weeks & post them. if anyone knows of .tar.gz cpunk archives somewhere, I could do it that much faster. some may object & think this will only add to the noise, but hey, as I always say, if you cant beat em, join em :p
Re: signal to noise proposal
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Declan wrote: >There may be the germ of an idea here, but I'm hardly convinced an >automated mechanism such as you describe will work. Even if it did, getting people focused on improving their popularity ratings rather than contributing ideas is hardly going to improve content. The only thing it would accomplish is promoting conformity of thought: "disagree with the group and be punished". It's far too easy to manipulate, anyway: have you considered the possibility of some vindictive loser with nothing better to do or a group of feds orchestrating reputational attacks against key posters? (spoofing, vote-rigging, etc.) As long as nyms unconnected to real names have votes, the system will always be wide open to this kind of thing and the numbers will be meaningless from the beginning. It'll turn into just another way for the offended to disrupt the group: think of all the people who used to post but left, angry and humiliated. They'll be back. Anyone who reads this list on a regular basis has a perfectly good picture in his or her mind of basically what they can expect from any given poster. How is a number going to express anything you haven't already figured out for yourself? Who I like to read most around here is entirely independent of my personal opinion of them, whether I agree with their posts or how nasty I get when I argue with them. I like to think I'd be able to get past the third grade playground mentality and give them a 10 or whatever when they deserve it: sadly I know as sure as I'm sitting here these very same people would do their damndest to obliterate me from the board forever. What a terrible waste of time and talent. This rating system is only going to make people more petty and vicious than they already are. As tempting as getting mickey-mouse revenge on your "enemies" may be, shouldn't we do what we can to just cut the bullshit squabbling and have an honest exchange of ideas with each other? I don't think the subjects of the list deserve anything less. >Perhaps an easier way to do it is to have everyone post their kill.rc >files publicly for everyone else's delectation. :) Seriously, a great idea. Quick, dirty, and to the point, everybody vents and moves on. Something else which might be worthwile is for each poster to go to the inet - -one or MARC archives and do a little statistical analysis of his or her own posts. What are you really accomplishing here? Are you an asset or a liability, a help or a hindrance? Are you bickering or contributing? Mee-tooing or saying something original? Are you fixated on anybody? boring the shit out of people? What can you honestly say you bring to the forum? A little more self-examination wouldn't hurt any of us. ~~Faustine. *** He that would make his own liberty secure must guard even his enemy from oppression; for if he violates this duty he establishes a precedent that will reach to himself. - --Thomas Paine -BEGIN PGP SIGNATURE- Version: PGPsdk version 1.7.1 (C) 1997-1999 Network Associates, Inc. and its affiliated companies. (Diffie-Helman/DSS-only version) iQA/AwUBPJ40r/g5Tuca7bfvEQIcGACfTCpO+OR8/RXTmMrJ1/eTYDZLrGIAoJuk SzYifCjwdfA709i730GuYVDD =WNvE -END PGP SIGNATURE-
Re: signal to noise proposal
On Sunday, March 24, 2002, at 12:30 PM, Declan McCullagh wrote: > On Sat, Mar 23, 2002 at 06:45:20PM -0700, [EMAIL PROTECTED] wrote: >> - posts to the list are like currency. lurkers > > Not a useful analogy. For some people, the more they post, the > lower their reputation falls. A lot of the current/recent "reputation schemes" make a fundamental mistake: they attempt to assign a scalar value to "the" [emphasis] reputation of an actor. Even the schemes which attempt to assign a vector rating, e.g, "Declan' s rating of Detweiler is..., Tim's rating of Detweiler is...," make a fatal mistake. There are no reputations attachable to actors in this way. What there are are _beliefs_ about certain actors held by others. "I have seen 8 years' worth of posts from Detweiler, with some mult-year gaps, and I have seen some fraction of comments made by other people, some of whom I respect (believe to some extent) and some of whom I don't believe (ignore, criticize, believe the opposite of usually). Based on all of these inputs, but very heavily weighted by my own past (Bayesian) experience, I tend not to take Detweiler's posts very seriously, even when he attempts to behave and attempts to put forth content rather than gibberish." (Readers will recognize that this goes beyond even a tensor, the generalization of a vector, and involves "stories" (possible worlds semantics, a la Kripke). Belief is partly Beyesian (or Dempster-Shafer-centric), partly a semantic net of many factors. Evolution has given us very good tools for assessing danger, deciding who's worth listening to and who's not, and how to plan for certain futures. Most of the mechanistic models for reputation are overly simplistic.) To paraphrase, "I made not be able to define bullshit, but I know a bullshitter when I see one." I encourage Detweiler to reify his ideas into code. Shouldn't take more than a short while programming in Python or Squeak to generate a filtering method he can apply to _his_ instance of the list. Though from what I just read a few minutes ago, with him excoriating the list for not rushing to begin implementing his latest ideas, it looks like DejaNews all over again. I give him 3 weeks of pounding headaches before he begins referring to An Metet as a tentacle of me, Koder bin Hackin', as a tentacle of Declan, etc. --Tim May
Re: signal to noise proposal
On Sat, Mar 23, 2002 at 06:45:20PM -0700, [EMAIL PROTECTED] wrote: > - posts to the list are like currency. lurkers Not a useful analogy. For some people, the more they post, the lower their reputation falls. > - mailing list records # of times someone wrote > a post that was replied to. posts that get replies > are generally an indicator of interest, useful Or people flaming them for being idiots, or trolls, or posting off-topic messages, or forwarding links to Slashdot items... > - everyone is subject to the same rules. there are Except the person running the list. There may be the germ of an idea here, but I'm hardly convinced an automated mechanism such as you describe will work. Perhaps an easier way to do it is to have everyone post their kill.rc files publicly for everyone else's delectation. :) -Declan
Re: signal to noise proposal
[EMAIL PROTECTED] wrote: [snip] > - ... the mailing list > simply records # of posts written by each poster. call > this "P" > > - mailing list records # of times someone wrote > a post that was replied to. ... call this "R" > > - pseudoreputation is a measure of the above two > parameters. one can experiment with different > metrics/weightings as a combination. e.g. x*P + y*R etc Aha, a simple formula for personal signal-to-noise. This could be extended to include other factors, such as quoting style (where Q is pecentage of mail quoted without reply). Many Fidonet groups used to contain frequent statistics on best and worst posters/quoters/etc. An automated reputation system based upon postings and replies is all to open to abuse though. Especially in a group such as this, where people post under several different addresses. Reputation-bumping through conversations with self (enlightenment turned practical?) would be too easy without further measures that hindered the simplicity of mailing lists. > - mailing list outputs current reputation alongside > peoples posts. I.e. if my reputation value is currently > [x], there's a "reputation-value: x" field output > in outgoing msgs. this is available for filtering > by end users. Fine for those of us whose mail clients don't support customized headers, or can't telnet to port 25. > - mailing list might also support a filter > such that people can toss out msgs from sources with > too low a reputation by their specification. Or with too high - must be some kind of niche market for people that want to read crap, look at the kind of forwards I get every day ;) > - I propose that those with low reputations are not > bounced from the list, only given quotas. say the > lower the reputation goes, the fewer msgs per day > they are allowed to post. Ick - as soon as you start to limit people's postings then you're probably in for trouble. While it may be of benefit to the list as a whole to prevent a person from repeatedly posting nonsensicals, the reputational scalability infers that people who would otherwise be able to comment as many times as they liked on any thread that they liked would suddenly find themselves having to choose which messages to reply to, and lo, the whole idea in this case of both the list and moderation of that list, that of encouraging conversation, is endangered. perhaps some kind of cut off point *might* work, with some experimenting as to its boundary, but I am much more in favour of client-specific filtering of message reading rather than writing. A supplementary web-based interface would be almost essential under a filtered system, I think. The ability to refer to messages you would otherwise never have read is important. > the tweaking would have a lot to do with the > weighting of the reputation, etcetera. it could also be tweaked differently according to the type of group, too - discussion and announce lists could be within the same barn, just viewed under different parameters. > none of this requires moderation or a lot of extra > activity, which I think is absolutely crucial in > any workable system. nobody wants it to be any work > at all. And how many people, in _general_ (Yahoogroups, MSN communities et al, as opposed to on CP), would probably not bother with moderation at all, but be content to simply hit "delete" for any irrelevant messages? In fact, how many would actually *reply* to the trolls? ;) .g
Re: signal to noise proposal
On Saturday, March 23, 2002, at 05:31 PM, Graham Lally wrote: > Adam Back wrote: >> Apart from my recent comments about NoCeM's and on onspool NoCeM >> reader, another perhaps simpler idea would be to do it all with simple >> CGI stuff and a web archive. I'm sure this has been discussed before >> in the past, but I don't recall anyone actually trying it out: > > TBH, I'm surprised nobody's mentioned Slashdot yet... I did, just a couple of days ago. Slashdotting is a major, if not _the_ major, contributor to short attention span noise. Choate relies heavily on Slashdot...'nuff said. --Tim May "To those who scare peace-loving people with phantoms of lost liberty, my message is this: Your tactics only aid terrorists." --John Ashcroft, U.S. Attorney General
Re: signal to noise proposal
hi guys, thanks for the feedback.
I have in mind a system that would work from
a few factors, with minimal intervention.
I specifically think that a
system of appointed moderators is **not**
an ideal solution for a bazillion reasons.
that is precisely what I have "not" in mind.
the whole moderator thing is a step backward,
imho. cyberspace needs some kind of self-sustaining
all-user-supported system.
yes, I did forget to mention slashdot, they
do have a decent/workable reputation system.
I agree, I dont want the system to run over
the web as a required feature (but as a
supported one, OK).
the basic ideas I am toying with are the following:
- posts to the list are like currency. lurkers
are not really useful to mailing lists as
far as advancing the dialogue. so the mailing list
simply records # of posts written by each poster. call
this "P"
- mailing list records # of times someone wrote
a post that was replied to. posts that get replies
are generally an indicator of interest, useful
posts, etc.. call this "R"
- pseudoreputation is a measure of the above two
parameters. one can experiment with different
metrics/weightings as a combination. e.g. x*P + y*R etc
- mailing list allows anyone to flip a bozo bit on
another poster ("plonk") or "unplonk".
but the plonk is weighted
based on the plonker's "reputation" by above. this
is so anonymous plonks dont have any value, but someone
who has built up some reputation can have an effect
on plonking other users. users can negotiate with
each other behind the scenes or on the list to ask
for "unplonking". the plonk note to a server can include
a "reason for plonking msg" & gets sent to the plonked user.
- mailing list outputs current reputation alongside
peoples posts. I.e. if my reputation value is currently
[x], there's a "reputation-value: x" field output
in outgoing msgs. this is available for filtering
by end users.
- mailing list might also support a filter
such that people can toss out msgs from sources with
too low a reputation by their specification.
- I propose that those with low reputations are not
bounced from the list, only given quotas. say the
lower the reputation goes, the fewer msgs per day
they are allowed to post.
- everyone is subject to the same rules. there are
no moderators or appointed keepers of the sacred
signal-to-noise. however those with greater reputation
have greater effect. as you can see reputation is
based on (a) posts, (b) replies, and (c) avoidance
of plonks. but note the system is workable even
without plonking.
the tweaking would have a lot to do with the
weighting of the reputation, etcetera.
none of this requires moderation or a lot of extra
activity, which I think is absolutely crucial in
any workable system. nobody wants it to be any work
at all.
Re: signal to noise proposal
On Sun, 24 Mar 2002, Adam Back wrote: > Are there people who already read cpunks regularly via the web? > > (Reading email and mailing-lists via the web always seemed clunky to > me, even on broadband, but there are apparently vast numbers of people > who use only web-email by preference, and to this group presumably a > web archive is preferable to subscribing to a list and reading it's > contents via their web-email account page.) Any time a URL of mine is posted to the cpunks list, I get a large number of hits with the inet-one archive listed as the referrer. More interestingly, most of these are either .mil or .gov addresses, or unresolved IPs that disappear within the ncsc mesh.
