A Fuzzy Implementation for the Socio-Cognitive Approach to Trust - Trust Theory: A Socio-Cognitive and Computational Model - page 316

importance of trust components/determinants (for different personalities or different situ-
ations). Our experiments show that the relative importance assigned to each feature may
drastically change the results. Most of the differences in FCM's behavior are due to the strong
causal power assigned to ability (routine visit scenario) and accessibility (emergency visit
scenario), even if the basic beliefs values are the same.
11.12.5 Evaluating the Behavior of the FCMs
We have conducted several experiments modifying some minor and major beliefs' sources in
the FCM of routine visit scenario for the doctor. This allows us to evaluate their impact on the
overall results.
We can see that the FCMs are quite stable: changing minor factors does not lead to catas-
trophic results. However, modifying the values of some major factors can lead to significant
modifications; it is very important to have a set of coherent parameters and to select the most
important factors very accurately.
However, our first aim is not to obtain an exact value for trustfulness for each FCM; on the
contrary, even if we consider the whole system as a qualitative approach, it has to be useful
in order to make comparisons among competitors (i.e. the doctor and the machine in our
scenarios). So, an important question about our system is: how much can I change the values
(make errors in evaluations) and conserve the advantage of a competitor over the other?
In the routine visit scenario the two trustfulness values are far removed from one another
(0.57 for the doctor vs. 0.23 for the machine). Even if we change several factors in the
machine's FCM its trustfulness does not overcome its competitor's one.
11.12.6 Personality Factors
Given the way in which the network is designed, it is clear that the weights of the edges and
some parameters of the functions for evaluating the values of the nodes are directly expressing
some of the personality factors. It is true that some of these weights should be learned on the
basis of the experience. On the other hand, some other weights or structural behaviours of
the network (given by the integrating functions) should be directly connected with personality
factors. For example, somebody who particularly cares about their safety can overestimate the
impact of danger and unharmfulness, or even impose a threshold on the final decision. Each
personality factor can lead to different trust values even with the same set of initial values for
the beliefs sources. Many personalities are possible, each with its consequences for the FCM;
for example: Prudent : high danger and unharmfulness impact; Too Prudent : high danger and
unharmfulness impact, additional threshold on danger and unharmfulness for decision; Auto :
high direct experience impact, low impact for the other beliefs sources; Focused on Reputation :
high reputation impact, low impact for the other beliefs sources.
Some personality factors imply emotional components, too. They can lead to important
modifications of the dynamics of the FCM, for example modifying the choice of the heuristic
for combining homogenous and heterogeneous fonts.
To summarise, we can say that our experiments aim to describe the dynamics of trust and
to capture its variations due to belief sources variation, and the different importance given
to the causal links and personality factors. The scenarios presented here fail to capture many
factors; in addition, we have assigned values and weights more as a matter of taste than through