Information Technology Reference
In-Depth Information
Trust
0.0
Internal Factors
.5
.5
External Factors
0.0
0.15
0.0
0.15
0.15
Ability
Environment
Willingness
0.0
0.0
0.0
Figure 11.9
The FCM used by the
cognitive trustor
(Falcone
et al.
, 2003).
7
The values of three nodes (ability and willingness as internal factors
and environment as external factor) were set according to agent knowledge. The values of the
edges reflect the impact of these factors and are always the same in the simulations. It has
to be noticed that we never tried to optimize those factors: the results are always significant
with different values. An additional point: while in the experiments the environment modifies
the ability, in the 'mental representation' of FCMs this is not the case: this is not a piece of
information that an agent is meant to know; what it knows is that there is an external (positive
or negative) influence and it aggregates it by fulfilling the cognitive model. Figure 11.9 shows
an (un-initialized) FCM.
11.14.6 Experiments Description
The first aim of our experiments is to compare the
cognitive trustor
and the
statistical trustor
in different situations: their delegation strategy represents two models of trust: derived from
direct experience versus experience built upon a number of cognitive features. The random
strategy was added as a baseline for the difficulty of the setting. The
best ability
and
best
willingness
strategies are added in order to verify, in different settings, which are the single
most influential factors; as it emerges from the experiments that their importance may vary,
depending on some parameters.
In all our experiments we used exactly six agents (even if their delegation strategies may
vary); it is important to always use the same number of agents, otherwise the different sets of
experiments would not be comparable. In each experiment the task set by all agents is always
the same; their ability set and willingness, as well as the environment influence, are randomly
initialized.
The experiments are performed in a variable environment that influences (positively or
negatively) the performance of some agents in some tasks, as previously explained.
7
With respect to the general model, for the sake of simplicity we assume that unharmfulness and danger nodes are
always 0, since these concepts have no semantic in our simulations.
Search WWH ::
Custom Search