Ming's Dynasty
I've found myself stopping by Flemming Funch's Ming's
Metalogue daily, and reliably find a useful post. Today he's discussing
reputation
systems:
Alex
Halavais talks about an experiment with a karma/reputation system in a
class he was teaching. The idea being that one had a certain number of points,
and one could give them to others for doing good deeds, according to a simple
system. But people cheated and the system fell apart.
I've noticed myself that it is rather difficult to make a functional reputation
system. There is one in NCN, where people
mark others as being 'acquaintances', 'friends' or 'comrades', meaning that
they're somewhere on a scale between 'I know them' and 'I would trust them
with my life'. Some of the problems I've noticed are:
- People have different norms. Some people feel they trust everybody
unconditionally.
- Many people feel obliged to be reciprocal, even if they don't quite
mean it.
- Some people try to have several virtual personalities, so they can
give each other points.
- If there is a list of people's reputation ratings as numeric values,
ordered in descending numeric order, people change their behavior and get
competitive about getting better numbers.
- If I made the system, and I'm first on the list, people get suspicious.
- People who are very active get high ratings.
- Some people end up hating reputation systems.
Aside from that, it works fairly well. I just think I need to get rid
of the comparative listing.
Of course, Xpertweb is nothing more than a reputation system:
- Sellers deliver value for a proposed fee, say $100.
- The Buyer rates the seller 1-99% and adds a written comment.
- A grade above 85% receives $100
- A grade between 50-85% receives $50-85
- A grade below 50% is failing - no payment
- Grades are compiled into a reputation for the seller and
the buyer.
- Mentors who train others who get good grades are well rewarded.
- Various mechanisms counter the foreseeable manipulation possibilities.
We've decided to just put the system in place rather than model it in a controlled
environment. The reason: Grades must be dollar-denominated to mean anything.
The only way to understand how something really works is just to do it.
Our work on Xpertweb has convinced me
there are 4 problems with real-world reputation systems:
- People no longer work for a living. They hold jobs for a living, which pays
better.
- People are incidentally loyal to a company, but deeply loyal to an accounting
system.
- Accounting systems are designed to buy work cheap and sell it dear.
- Managers of accounting systems are hostile to exposing success or failure,
whether within or outside of the organization.
Where programmers and bloggers take delight in finding bugs and celebrating
quality, all organizations, like most people, are insecure. One study (can't
find it now) discovered that most high achievers are afraid they'll be found
out. Transparency exposes incompetence which is why any organization
that manages an accounting system as its primary activity (is there any other
kind?) cannot and will not support reputation-building. Only humans would consider
such exposure and then only those highly qualified or motivated to do so.
We need an economy of people who solicit suggestions for improvement from customers
and mentors. I guess we'll just have to start small.
Funch-y Musings
Flemming Funch has a series of entries on the subject of organization and they're
worth a look:
2002-12-19: Reputation
Systems
2002-12-10: You
can't shut up a network
2002-12-04: Dynamic
Facilitation
2002-12-02: Online
Business Networks
2002-11-30: Power-law
distributions on the web
2002-11-30: People
Tour
2002-11-29: Compliance
or Creation
2002-11-26: The
State of Grace Document
2002-11-25: Self-Organization
2002-11-22: Fertile
soil for group-forming
5:20:41 PM 
|
