I would like to add in my 2 satoshis here.
I spent a long time thinking about this reputation issue; here's what I
came up with:
Trust systems, we all agree, are imperfect. Bitcoin replaced trust with
cryptographic proof, but it still needs consensus - the consensus works
well because it builds on verification rather than unreliable identity
reputation.
In an exchange system for fiat or physical goods, we cannot do this, so
I think it's worth "quarantining" trust/reputation. This can be done by
the use of voting pools - thus, arbitrators of disputes do need to have
some trust, but that trust can be "smeared" over a pool to reduce to a
minimum the central point of failure problem.
The great gain in limiting trust to arbitrators (and not users) is that,
combined with multisig, the arbitrators have no funds-stealing
capability in and of themselves. This can be further bolstered by
randomising the arbitrator-buyer-seller link. (These steps are what I
mean by "quarantining" trust).
Meanwhile, trust in users (buyers/sellers) can be removed entirely, as
long as the transfers of value are both (a)irreversible and (b) verifiable.
Consider the two obvious applications: 1 Silk Road 2 fiat-BTC trade.
1 Silk Road - transfer of goods is irreversible, but not verifiable.
2 fiat-BTC transfer of fiat is verifiable (ssl logging) but not fully
irreversible (although it can come VERY close).
Because of these limitations, users will always want some
identity/reputation tagging.
Thinking in this way, you could see that many users can and WILL create
or rely on existing reputation systems, no matter what we do. However
for special cases, they can and would use such systems anonymously, with
no reputation, and be very happy to do so.
These ideas are fleshed out at
https://github.com/AdamISZ/stegabank/blob/master/protocol.md
On 03/17/2014 12:44 PM, Manfred Karrer wrote:
> Hi Jaume,
>
> I did not want to say it has not value, often it matches well with the real expectation.
>
> But I think those reputation systems only have a limited validity from the data from where the reputation is derived. Like in bitcointalk its the number of posts and the activity (post over period of time, so regular long time posters has more value), and to fake that with sock puppets means a certain amount of work have to be spent, so sock puppets with "Hero Member" attribute are probably very rare.
> Of course it is not easy from a usability point of view to present the user a more complex reputation algorithm as pure data, so they tend to make easy to understand categories like "Hero Member", but at the end those are misleading. If the user only gets presented the naked facts, he need to derive in his brain a high level attribute from that (> 500 posts -> experienced user,….) and he would be more aware of the relation and limitations of that process.
>
> Back to the exchange project, I assume that reputation manipulation with self-trades cannot be prevented, so I am wondering if it is better to avoid reputation completely, as providing no data seems better than providing (possibly) false data (due sybil). And in the exchange project the reputation would only serve for punishing scammers who do bank chargebacksor use stolen bank accounts, as the rest is covered by arbitrators.
> But for those kind of scams a blacklisting by an arbitrator with a proof of the scam is justified, and then reputation is not needed as well.
>
> For other use cases a more soft approach might be justified also because often there are no hard facts available, so human intelligence/judgements is needed to derive meaningful conclusions. Like with "upvoting", "+1" or "like" its fine to have those tools which does not imply a precise valuation and have no critical consequences (other as in an exchange when you trust a sock puppet and can be a victim of a scam).
> But I guess all of us know how limited those approaches are as well. See how poor the reddit scoring works in realitiy. Gossiping outrival content too often….
>
> br,
> Manfred
>
>
> Am 17.03.2014 um 08:36 schrieb jaume <jaume@???>:
>
>>
>>> To provide measurable data is one thing, to derive from that data
>>> qualitative statements is problematic.
>>> As example see the bitcointalk "reputation" categories like "Hero Member",
>>> etc…
>>> It does not mean really much that one is categorized as "Hero Membe". It
>>> only means he has posted a lot. So better to display only that naked data
>>> and not derive any unprovable statements.
>>
>> Even though there is a history of "hero members" that were scammers, being
>> a hero member provides some kind of "proof of work". As the posts are
>> public, it is possible to evaluate the quality of such posts. A member
>> that has posted a large number of high quality posts over the years
>> deserves some "reputation" as it has contributed to the community. As this
>> reputation has a "value", the user has some incentive to preserve that
>> reputation by being legit. I have the feeling that the number of scammers
>> is higher in the ranks of "junior members" than it is in the ranks of
>> "hero members".
>>
>> My belief is that we are moving towards a "reputation economy" in which
>> what is important is what we contribute to the community and not what we
>> have.
>>
>> Peace :)
>> Jaum
>>
>> _______________________________________________
>> unSYSTEM mailing list: http://unsystem.net
>> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/unsystem
>
>
>
> _______________________________________________
> unSYSTEM mailing list: http://unsystem.net
> https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/unsystem
>