Reputation system

From OVN wiki
(Redirected from Reputation)
Jump to navigation Jump to search

Reputation applies to agents.

Reputation is public. It can be construed from metadata about activity / behavior. It is information about an agent's past activities that helps other agents make decisions about possible future relations, action or outcomes.

On the one hand, reputation is about predictability, as in about trusting an agent (in context) to perform a specific (set of) activity. Thus, reputation is essential for building predictable environments and socioeconomic processes.

In a game theoretical approach, an affiliate will act predictably (which can also mean rationally, but not always), to increase his/her reward, and the worth of the entire network / community, which is in part transferred back to the same agent / player as reward. In other words, the reputation of an OVN affiliate is related to the probability that this affiliate respects the formal rules and norms that govern co-creation and transactions within the network, which, if the network is properly designed, should maximize reward for the affiliate, which in turn aligns with what makes the network sustainable. If the economic model is well-designed, there should be very little or no conflict between individual interests and collective interests (see also tokenomics in the context of DAOs).

Reputation has also a qualitative and subjective dimension, related to how actions are performed, quality of deliverables, attention, social accommodation and inclusiveness, etc.

Reputation is also a form of wealth that individuals nurture in social contexts and organizations. It provides access to socioeconomic processes as a dimension of credentials. In that sense, it is a non-transferable type of wealth, it belongs to one particular agent. When reputation is made explicit in a socioeconomic context as a condition to access (processes, resources) people become conscious about it and model their behavior and personality to nurture their reputation, as expected within that environment. Therefore explicit reputation parameters is also used to model behavior.

Reputation can be aggregated over a network to provide a measure of network (organizational) quality, which is a network type of wealth.


Reputation system

The OVN requires a comprehensive and fair reputation system to process and manage this data for collective intelligence and decision making (see governance).

The reputation system incentivizes good behavior within the OVN and helps to focus attention. Moreover, it plays an important role in innovation and production processes by filtering participants for adequate tasks (credentials). See also resource allocation). These systems are also designed for network-to-network interface.

The reputation system can also be seen as a self-exclusion mechanism. In other words, access to processes (or priority for access) can be granted (modulated) programmatically based on reputation. Affiliates that acquire bad reputation can thus be gradually excluded from processes to the point of eliminating motivations to stick around.

Metrics

Metrics can be divided into two general parts: for individual agents and for a network.

Reputation metrics for the whole network can provide an aggregate of reputation as quality of a network, which is part of the wealth structure of the network.

Design

  • multidimensional and composable, best represented by a graph, not just a scalar number.
  • computable, at least in large part
  • transportable (see individual profile)


Objective and subjective reputation

Some people argue against subjective reputation in open networks. In fact, the prevalent web3 culture is based on trustlesness.

Commitment can be seen as an objective dimension of reputation: A promises to deliver X for a given date (time), and X satisfies a set of predefined requirements (quality).

Being sociable and collaborative can be seen as a subjective dimension of reputation, although some people can argue otherwise based on patterns of engagement extracted from digital traces of activity.

Negative reputation

Where data about reputation is not collaterally shared by the agent in question. Ex. credit history, data about unpaid loans, negative reviews and complaints from business partners.


Contribution Accounting

The contribution accounting system records every affiliate's contributions, and provides opportunities for feedback from peers (based on quality of execution of tasks). Revenues are distributed in terms of affiliate's contributions, factored by the benefit redistribution algorithm which incorporates some reputation data. This system outputs a map of sources and flows of valuables.

The role system incites voluntary subordination and plays an important role in self-organization. contributions, reputation and roles interact with each other. All these systems are necessary to induce tight self-organization within the OVN, and to render the network creative and productive.

Reputation factors are linked to the contribution accounting system and can play a part in the benefit redistribution algorithm. In other words, reputation can modulate someone's ability to benefit from the network. Thus, the reputation system provides a mechanism of continuous self exclusion. If someone gradually loses reputation to a point where the rewards become lower than what the individual can get somewhere else, this individual loses his/her incentives to stay with the network and might decide to quit.


Reputation with respect to documentation of work

Documentation is a mechanism of accumulation, which makes an OVN more valuable over time. In other words, the sum of digital assets produced and shared across the network has generative effects and is considered as a form of network wealth. Thus, it is very important that affiliates share their work. It is not enough to contribute, it is also important to deposit something within the commons of the OVN. Proper documentation is important, for different reasons. First, others can build on someone else’s work, this is for the cumulative effect. Second, others can intervene early to correct a situation if someone is taking a wrong path, this for quality and the two combined make the OVN more efficient. Third, this documentation will be integrated to the feedback system, used to provide other affiliates with a big picture, in real time.

Proper documentation is made timely, with enough detail, contains failed results and explanations about why they might have failed. Furthermore, documentation must be done in accordance with the logic of stigmergy and it plays an important role in the good functioning of the open network.

Ideas

Different levels of organization and reputation

Yasir proposes to split reputation according to structural dimensions: project, enterprise, network, and network of networks.

p2p infrastructure for reputation - ideas

The reputation of an affiliate describes HOW this affiliate contributes, or in what manner. The Reputation system interfaces with the individual profile and has write access. In reality, reputation of an affiliate lives in the heads of all other affiliates who have shared experiences together. There is no central platform for reputation. To verify the reputation of an affiliate, one needs to query, or asks other affiliates. Reputation on a p2p platform should mimic that. Reputation parameters about every affiliate can live as data that belongs to every other affiliate. A search engine can be used to query all affiliates that have reputation data about another affiliate, and present a reputation aggregate or summary. The affiliate should not be allowed to edit his own reputation data.

Think of 'reputation' not only in terms of multiple measures, but multiple measures measured in different types of relationships. Just because my friend thinks that I'm lazy doesn't mean my boss thinks I'm lazy too. It's a different type of relationship, laziness features differently in them. My future boss doesn't care that my mom thinks I'm lazy. Her rating would contaminate that measure and should be kept separate. VDML supports this thinking.

Praxis

There are no examples of advanced reputation systems for for an OVN. SENSORICA's prototype is far from being functional.

Reputation features in the current NRP

People can voluntarily commit to perform some listed tasks. NRP-CAS registers those commitments, which will usually be connected to one or more deliverables, which in turn will be required by other processes. So if someone does not fulfill a commitment, and it affects the next process, the agents at the next process will be notified. One thing they can do is to ding the reputation of whoever committed to to the work that is now holding them up.

The commitment has 3 properties that could be evaluated and affect the rep of the committer:

  1. was it fulfilled at all? (that affects anybody who was expecting the deliverable)
  2. was it fulfilled on time? (that only affects a process that was expecting the deliverable at that time)
  3. was the quality of the deliverable any good (that affects anybody who used the deliverable)

1 and 2 only affect people who had prior plans that expected that commitment, which would also be known in advance to the committer. It's not anybody in the world, it's a small set. If we are talking about large numbers of people who later decide they want the thing, the timing was none of their business, nor was the doing at all.

Likewise if the quality of a deliverable does not meet the requirements of the next process, the people at the next process will down-rate the quality, and that will be reflected on the reputations of the people who created the deliverable.

If it is something like a design or a document and they actually do use it, they could have a chance to rate the quality, but in that case the thing would probably have a lot of ratings that would be subject to some kind of averaging algorithm.

And in some cases, the quality could be set fairly objectively, too. E.g. software in a project that has tests: does the new commit pass the tests? then shut up. Lots of other products could also be tested. E.g. Sensorica tests some of their products.

If the person who committed to do the work did it on time and the quality rating was also good, their reputation will be increased.

The commitments exist now. The notifications have started to exist, but nobody is ready to use them yet. The quality ratings have not been deployed yet, but could be in short order. The reputation system does not exist yet.

When someone commits to a work he/she doesn't necessarily know who is waiting for the results. We have not decided yet exactly how the reputation factors will be handled if and when the late work starts to delay the next process: could be automatic, or could be by choice of the people working on the delayed processes.

There is also a notification system in place to inform the contributor and those who are waiting for the work about the progression of the work.

Another reputation factor that exists in the server-based NRP-CAS now, that could be developed a lot more, is scores for skills. The NRP-CAS keeps track of all of the work people do by type of work (for example, electronics, optical, 3D design, etc. etc.), and accumulates scores for each person for each type of work. The scores now are just the total number of hours worked, but they could also include the quality of deliverables. That datum is in the system, but has not been used yet.

So people could see who has good scores for electronics design, for example, and know who to approach for training or questions or to ask for help.

Problems with reputation systems

It is about how good one is. The problem is that this is always in a context. People have lot's of qualities that are often not acknowledged or picket up in a reputation score. The organization misses opportunities when applying metrics to construct a reputation score.


Current reputation systems are:

  • vulnerable to user privacy leaks;
  • vulnerable to false ratings;
  • subject to manipulation by the storing authority;
  • involve a high number of transactions for voting or reputation queries.

References