Applicability of reputation systems in information systems

    Paul proposed a conjecture regarding the validity of using reputation systems in the context of identity systems. This (and some discussion on the IDGang list) inspired me to dig again through some of my notes regarding the ontology of physical reality (and thus–by extension–quantum theory).

    My personal position in the discussion on the most sensible approach to physial ontology was always firmly rooted in the realist corner: I completely reject positivism and–mostly– empiricism on fundamental principle. There is no doubt in my mind that there is an objective physical reality, independent of human (or any other) observer^[1].

Reputation in information systems

    Now, a reputation scheme can easily be interpreted as mechanism to determine the value of an entity’s attribute by averaging over the subjective values of that particular attribute, as seen by an ensemble of parties interacting with the entity in question. So, for example, to determine the “trustworthiness in business transactions” of user A of an auctioning site, one can average over the subjective opinion of business partners of user A on his trustworthiness.

    This approach is valid, and as many social (or even business) sites indicate very useful. It can be applied reasonably well to attributes of an entity that are either non-counterfactual definite (i.e. completely subjective), or not measurable by an objective and reproducible measurement approach.

    “Trustworthiness” is a good example for a subjective attribute, and credit-worthiness of a company or individual might be an attribute of the later type: while the fundamentals of a company determine its ability to shoulder a certain about of debt without collapsing, there is (to my knowledge) no definite algorithm to compute a simple “creditworthiness” attribute. However, the averaging over the credit ratings from different rating agencies (i.e. a kind of “credit reputation”) is normally a good approximation of this attribute^[2].

    However, there are some attributes that cannot be averaged over: those attributes are counterfactual definite, i.e. objective and can be measured by a repoducible mechanism. A good example for such an attribute is my physical height,  my employment status with a given company, or my gender. All of these might change in time, but at a given point in time, they can be easily determined and have an objective value–even if nobody measures it. Applying a mean-value approach to these does not make any sense.

    One might interject, that for such a counterfactualy definite attribute there might be a different perception of its value with other entities. For example, while my actual height is 187cm (~ 6′ 1″), some people might think that I am taller or shorter.  Now, my actual height does not change because a number of people are thinking so. It is my perceived height that changes and this attribute is entirely different from the former.

    So, in the end it is very important to evaluate carefully if a given attribute of an entity in an information system lends itself to be used in the context of reputation systems. In some specific cases this does make sense, but in others it is entirely pointless.

[1] Yet, while realism is vital to my world view, I am much more inclined to abandon local reality than counterfactual definiteness.

[2] The current financial quagmire is an example of how such a reputation system can fail.