Trust and Technology - Trust Theory: A Socio-Cognitive and Computational Model - page 344

The addition of the fourth source (certified reports) to the traditional first three allows trust
to measure in situations in which without this source there is no reference.
The FIRE 's authors show a set of experimental results in which the performance of their
system is really good and better than other developed systems. In particular they show that:

Agents using the trust measure provided by FIRE are able to select reliable partners for
interactions and, thus, obtain better utility gain compared to those using no trust measure.
Each component of FIRE plays an important role in its operation and significantly contributes
to its overall performance.
FIRE is able to cope well with the various types of changes in an open MAS and can maintain
its properties despite the dynamism possible in an environment.
Although decentralized, to suit the requirements of a trust model in open MAS, FIRE still
outperforms or maintains a comparable performance level with SPORAS , a centralized trust
model.

The main problem and limit with FIRE is about its assumption that agents are honest in
exchanging information with one another: The authors are aware of this limit and are working
to introduce reliability measures for witness ratings and certified ratings.

Different Kinds of Metrics

The computational approaches to trust can be classified also on the basis of the different
forms of metrics to rate the trust performances. An analysis about the metrics of the differ-
ent approaches is particularly relevant because it should give account about the underlying
assumptions of these approaches.
Discrete or continuous measures (in general in the numerical range (

− 1, + 1) where − 1

means no trust and 1 full trust) are the most common. But there are also discrete verbal

statements like those used in (Abdul-Rahman and Hailes, 2000): Very Trustworthy , Trustwor-
thy , Untrustworthy , Very Untrustworthy , that give a more direct representation of the human
evaluations: The problem is then to translate these statements in adequate misures for the
computational process.
We can cite (Schillo et al. , 2000) and (Banerrjee et al. , 2000) as examples of the use of
bi-stable trust values (good or bad) (Ramchurn et al. , 2004); while (Witkowsky et al. , 2001)
proposed the calculation of trust (a continuous measure) that deal with measurable quantities
of bandwidth allocation and bandwidth use (they presented a scenario for telecommunications
Intelligent Network in which bandwidth is traded by different agents).
Probabilistic approaches have the advantage of exploiting the consolidated apparatus of
the probabilistic methods: in particular they can profit from the different derivation methods
(from the probability calculus to the advanced statistical methods) (Josang, 2001), (Krukow
and Nielsen, 2006).
An interesting and useful approach is given by the Belief Theory that responds to the limits
of the probabilistic approaches with respect to uncertain information. In fact, the aim of belief
theory is to give a formal representation of the inaccurate and uncertain aspect of information.
In this case the sum of the probabilities over all possible outcomes do not necessarily sum up
to 1 (the remaining probability can be interpreted as uncertainty).

+