Information Technology Reference
In-Depth Information
77,53
72,84
64,08
62,57
61,78
44.18
Random
Trust
Trust_L
Ability
Willingness
STAT
Figure 11.10
Experiment 1: comparison between many delegation strategies, 3 hits; (measuring the
success rates). (Reproduced with kind permission of Springer Science+Business Media
C
2005)
In order to allow the
statistical trustor
to learn from the experience, all the simulation
sets were divided in two phases (two halves). The first phase is meant for training only: the
statistical trustor delegates several times the tasks to all the agents and collects data from
successful or unsuccessful performance. It uses these data in order to choose to whom to
delegate in the second phase (in fact, it continues to learn even in the second phase). The
delegation mechanism is always the same: it chooses the agent who has the best ratio between
performed and delegated tasks; this number is updated after each result following a delegation.
In order to measure the performance of this strategy, we analyzed only experimental data from
the second phases.
The first experiment (
EXP1
) compares the
random trustor
(
RANDOM
), the best
ability
trustor
(
ABILITY
), the
best willingness trustor
(
WILLINGNESS
), the
statistical trustor
(
STAT
),
and two other cognitive strategies that differ only because of how much they weight the
environmental factor:
no impact
(
TRUST
) does not consider the environment, while
low
impact
(
TRUST L
) gives it a low impact (comparable to the other factors). In Figure 11.10 we
show 250 simulations for 100 tasks.
We can see that the cognitive strategies always beat the statistical one.
8
Moreover, it is
important to notice that recognizing and modeling the
external components
of trust (the
environment) leads to a very high performance: the cognitive trustor who does not consider the
environment (
TRUST
) beats the statistical one (
STAT
), but performs worse than the cognitive
trustor who gives a role to the environment (
TRUST L
).
8
The results are similar e.g. with five hits (250 simulations, 100 tasks):
RANDOM
: 26,24;
TRUST
: 57,08;
TRUST L
:
61,43;
ABILITY
: 40,58;
WILLINGNESS
: 48,0;
STAT
: 49,86.
Search WWH ::
Custom Search