I have to agree with most of ianG and James words. Note that your system is vulnerable to Byzantine attacks for example. However, I have the belief that trust could be made formal in some way, if done correctly. Maybe more work is needed. I'm not saying I know how to do it though :)
@Liam Edwards-Playne: Don't give up as you read our critiques. Maybe they can help you improve your idea. I recommend you to read about some other attempts of other people. Some of them that I can remember right now are: EigenTrust : http://en.wikipedia.org/wiki/EigenTrust P2PREP If you follow these articles you can find many more articles. If I remember correctly, some of these articles actually prove some ideas mathematically, which is really cool. Though they usually assume something more besides the generic symmetry between users. (For example: That some known set of peers are always trusted.) For me trust is connected with money or property in some way. It means to me that if I'm paying something, I know I get the goods that I wanted. (And if I don't, the other side looses about as much as I lose). Something like reputation in Ebay. In this context I want to ask another question. Given a known decentralized monitary solution (For example, bitcoin), do you have any idea about how to solve that trust problem? Naively speaking, If I want to create a file sharing program that allows to transfer files between users, and users actually pay in some way for the file transfer, How to make sure that clients in the network prefer to give good service, and not decide to cheat? Regards, real. On Sun, Jan 20, 2013 at 10:10 AM, ianG <i...@iang.org> wrote: > On 20/01/13 09:40 AM, James A. Donald wrote: > > On 2013-01-16 10:17 AM, Liam Edwards-Playne wrote: > >> Good day to you all, > >> > >> I've recently devised a solution for decentralized P2P trust > >> networking which you might all be interested in. It relies on an > >> artificial reasoning suite called Subjective Logic to provide methods > >> for deducing transitive trust. I've written a short 2 page paper to > >> detail the protocol - http://4abf.net/static/2013/jan/trust4.pdf > >> I'll be happy to answer any questions and discuss any critiques. > >> Regards, > >> Liam (liamzebedee) Edwards-Playne > >> > > > > It is not obvious to me how this performs should Bob create a million > > sybils to say that Bob is good. > > > > The system is designed to support generic trust. What are we > > trusting? Are we trusting Bob to be truthful, trusting Bob to pay his > > debts, trusting Bob to wisely curate content? Each of these is subject > > to subtly different attacks. > > > > Any trust scheme has to manage information about the probability of a > > probability, or rather the probability distribution of a probability > > distribution, abstractions that are hard to grasp, and I had a > > particularly hard time grasping them when they were not anchored to the > > probability of any particular event. > > > One thing that bedevils such systems is a weak definition (or a complete > confusion) as to what "trust" means: > > A naive approach would be to retrieve trust as it is needed, > receiving the latest trust information at the cost of inefficiency. > > Indeed, such makes no sense. It is not possible to "retrieve trust" any > more than it is possible to download love, transmit units of fear, or > sell hunger. > > This assumption that "trust" can be unitised is what underpins most > systems -- they create single data units which are moved around. This > is more because the designers are computer programmers, and to them, > everything is a datum. > > If one challenges that assumption, that "trust" cannot be unitised itno > a tidy data structure, then we're back to the drawing board. > > As intimated above, and perhaps in that paper, we instead want > statements. ”a node will provide the most correct responses” But an > analysis of the statements as they are applied to multiple nodes also > reveals that we need a much more sophisticated structure about those > statements. The statements can't be totally free. > > Here's a simple test: what happens if the statement proves wrong? The > answer to that will define the system far more than how the units are > moved around. > > iang > _______________________________________________ > p2p-hackers mailing list > p2p-hackers@lists.zooko.com > http://lists.zooko.com/mailman/listinfo/p2p-hackers >
_______________________________________________ p2p-hackers mailing list p2p-hackers@lists.zooko.com http://lists.zooko.com/mailman/listinfo/p2p-hackers
