so many of these problems have the same root cause:
we don't have an effective data driven reputation system. we use gameable heuristics to track social capial.
when metrics for evaluation are flawed, people behave in ways that exploit the flaws even if they increase the likelihood of failure.
"we are not rewarded for necessary grunt work as much as shiny advances", for example. That's a failure of the reputation system to account for the value of that work.
My solution to this problem is a mathematical reputation system based on the same concept as page rank. The system is available here:
In the past few years I have been building up good reputation at various stores both online and offline. It bothers me I cannot use that reputation. For example a major supermarket chain here in The Netherlands rolled out self scanning from 2006[1]. They do random checks at the checkout, and for many years now they know I never forget to scan something. This resulted in the amount of random checks going down for me. That should mean something, and I should be able to use this trust/karma elsewhere. All these companies building data on me, and I can't use it myself.
The challenge is getting the companies to (from their perspective) "share" the trust information with the world, which includes their competitors.
I suspect we might need a "taxonomy of trust" so to speak, that allows the trust data to be anonymized and aggregated into commonly-accepted meanings of trust contexts, trust roles, trust relationships, etc. That might let these companies to release the trust data into such a format through a blockchain perhaps, and be able to participate in consuming the aggregated data. I'd need someone well-versed in game theory to figure out if an advantage is conferred to "leeches"; a company in such a scenario who only consumes the aggregated data but never send into the blockchain what they accumulate on their own customers. I think that's a real danger with such a scheme, but am not sure how to strongly dissuade that behavior.
Well, watch out. That sounds great if you trust everyone you interact with to give you good reputation. But pretty quickly that can lead to the recent stories about the Chinese social credit (assuming the stories are true). It can become a very subtle and powerful way to control people. Not to mention a legal minefield. Just look at credit ratings.
I like this very much, thank you for bringing it up. Take a look at Fabriq [1; PDF], trying to accomplish similar in a blockchain context. I especially like the work you put into explicating the implied respect issue.
What are your thoughts modeling respect not just as a unitary quality (still highly useful for quick, ad hoc, high-level evaluation), but also along crowd-created and crowd-defined axes? Then people can refine their description of respect and for example, say they agree with one crowd-group's definition of "good manager" for a specific person, but at the same time that person is not respected as another crowd-group's definition of a "good leader".
Gaming reputation systems over extended time periods and via aliased entities is a perennial problem. What are your thoughts on random latencies before respect scoring is evaluated on new respect data for an entity, securely tying a hash based upon fully-sequenced DNA to real-person accounts, interaction of entities in a specific context (someone might be respected as a great athlete, considered toxic in one of the companies they own, but respected in a different company), and tracking corporate aliasing (through mergers, acquisitions, spin-offs, name changes, etc.)?
we don't have an effective data driven reputation system. we use gameable heuristics to track social capial.
when metrics for evaluation are flawed, people behave in ways that exploit the flaws even if they increase the likelihood of failure.
"we are not rewarded for necessary grunt work as much as shiny advances", for example. That's a failure of the reputation system to account for the value of that work.
My solution to this problem is a mathematical reputation system based on the same concept as page rank. The system is available here:
github.com/neyer/respect
I'd love your feedback.