Any ideas for trust approaches that cannot be gamed by cliques in a distributed network? Anything I can think of would break the functionality of the network/remove all usefulness from it or open up other attack vectors.
You have to differentiate the actual network that holds the data (possibly as a DHT) and the network implicitely described by the statements.
The way you avoid clique is by staying within your clique. That is requiring that the entities you extend trust to be trusted by people close to you, who themselves trust each other.
But it's not just cliques. Say you have a system like eBay, based on buyer feedback. eBay can be fooled by playing a long game, and selling a lot of cheap items for a while before selling a big item.
If instead of eBay you have a distributed store of transaction report, you can constantly refine the trust algorithm. For instance, you might pay attention to the price of the items sold, or to the trustworthiness of the reviewers themselves, etc.
The more data goes in, the better your inference. Truth is coherent and entangled. The more data there is, the harder it is to convincingly lie (but the harder it is to remain anonymous as well).
If you stay within your clique you severely limit the number of possible merchants and if you trust them through personal relationship or foaf why go through the hassle of an escrow system like this?
>For instance, you might pay attention to the price of the items sold, or to the trustworthiness of the reviewers themselves, etc.
Ok so how about I just sell a bunch of high value "items" between my own accounts? Now my account has participated in a number of valid high value transactions. Since it is distributed there is no way to ensure the transaction is real.
>trustworthiness of the reviewers themselves
Who build their trustworthiness the same way. This is easy to game look at pagerank for an example of a similar rating system.
>The more data there is, the harder it is to convincingly lie
This isn't actually true though because you don't verify the data in any way(nor can you) so an attacker can craft the input to tell any story they want.
> If you stay within your clique you severely limit the number of possible merchants and if you trust them through personal relationship or foaf why go through the hassle of an escrow system like this?
Because as you increase the radius of your search, in a small-world netwok, you reach a very large number of real people while being minimally impacted by Sybil cliques. Check out the work on SybilGuard or SybilInfer for instance.
>Ok so how about I just sell a bunch of high value "items" between my own accounts? Now my account has participated in a number of valid high value transactions. Since it is distributed there is no way to ensure the transaction is real.
Ah, but you're looking for transactions reported by people who are themselves close to your circles of trust.
> Who build their trustworthiness the same way. This is easy to game look at pagerank for an example of a similar rating system.
A random surfer model is a naive model. With better probabilistic model, you become more resilient to gaming.
> This isn't actually true though because you don't verify the data in any way(nor can you) so an attacker can craft the input to tell any story they want.
This isn't arbitrary data, it's data connected to you in some way.
