A Fuzzy Implementation for the Socio-Cognitive Approach to Trust - Trust Theory: A Socio-Cognitive and Computational Model - page 327

shows four cases:

1) learning in normal environment, task in normal environment (normal/normal);
2) learning without environment, task in normal environment (no/normal);
3) learning without environment, task in negative environment (no/neg);
4) learning without environment, task in positive environment (no/pos).

11.14.7 Using Partial Knowledge: the Strength of a Cognitive Analysis
The results achieved in the above experiments are quite interesting, but rather predictable.
More interesting and with high degree of difficulty of prediction is the experiment in which
we try to individuate the level of approximation in the knowledge of a cognitive trustor about
both the properties of other agents and of the environment. In other words, we would like give
an answer to the questions: when is it better to perform as a cognitive trustor with respect to
a statistical trustor? What level of approximation in the a priori knowledge is necessary in
order that this kind of trustor will have the best performance?
To answer this interesting question we have made some other experiments (as EXP5 ) about
errors in evaluation. As already stated, all the values we assume about cognitive features are
true values: each agent knows all the real features of the others.
This is an ideal situation that is rarely implemented in the real world/system. In particular,
in a multi-agent system environment there can be an evaluation process that is prone to errors.
In order to evaluate how much error the cognitive trustor can deal with without suffering from
big performance losses, we have compared many cognitive trustors introducing some different
levels of 'noise' in their data.
Figure 11.12 shows the data for the random trustor (RANDOM); the best willingness
trustor (WILLINGNESS); the best ability trustor (ABILITY) ; the normal cognitive trustor
(NOERR) , as well as some other cognitive trustors (ERR 40, ERR 20 ,

) with 40%, 30%,
20%, 10%, 5%, 2.5% error; the statistical trustor (STAT). While all the experiments used
a set of six agents, we present aggregated data of different experiments. We have ordered
the strategies depending on their performance; it is easy to see that even the worst cog-
nitive trustor (40% error) beats the statistical trustor. Under this threshold we have worse
performances.

...

Real Time Experiments

We have performed some real time experiments, too. EXP6 (see Figure 11.13) involves three
cognitive strategies in a normal environment (250 simulations, 500 tasks).
The differences between the cognitive trustor without environment and the two with envi-
ronment are statistically meaningful; the difference between the two cognitive trustors with
environment are not. The results are very close to those that use turns; the differences depend
on the limited amount of time we set for performing all the tasks: by augmenting this parameter
more quickly, strategies become more performing.
Another experiment ( EXP7 ) aims at testing the performance of the first trustworthy trustor
( FIRST ). Here there are two parameters for performance: Credits and Time . Time represents