Information Technology Reference
In-Depth Information
Table 11.3
Source evaluation
Willingness Episodes
The source reveals itself not to be trustworthy (
I know John
intentionally lied to me
)
Competence Episodes
The source reveals itself not to be competent (
John did not lie but he
was not able to give an useful information
)
Self trust Episodes
The evaluator was responsible for a misunderstanding
(
I misunderstood
)
Accidental Problems
There were contingent problems and no revision is necessary (
there
was an unpredictable circumstance
)
that somebody intended exactly this (there was no misunderstanding) or if I do not remember
exactly what he said (or even if I am not sure that it was his opinion). In the latter cases, if I
am not certain of the source, it can be better to assign the error to my evaluation rather than
to ignorance of the source or even worse to his intention to deceive me. Such a change has no
impact on other interactions with the same sources (but it can lead to change my self trust value).
11.13.3 A Taxonomy of Possible Revisions
There are many possible ways to evaluate an episode of interaction in order to learn from it and
to decide to change one's beliefs. As we have shown, not only the sources' opinions, but also
the full set of interaction episodes have to be categorized; from this kind of categorization the
following belief revision process depends. For example, in order to comprehend the motivation
of the discrepancy between the source evaluation and my evaluation ('
John says that this doctor
is pretty good, but it results to me to be not so good
'). The first thing to consider is 'what
was the main factor' from which this discrepancy depends.
Obviously this decision process pertains to a cognitive apparatus and it is impossible at a
pure-belief level; so a cognitive agent needs some revision strategies that individuate the error
source and try to minimize it for the future. In Table 11.3 we propose a crude taxonomy of the
problems that can intervene in an episode of interaction.
Implicit revision performs better with regard to computational speed. However, explicit
revision has many advantages. First of all, taking into account single cognitive components
allows a better granularity; this can make the difference where fine-grained distinctions are
needed, for example in order to distinguish between trust somebody as an information source
and as a specialist, or to distinguish a deceiver from a not informed source. Also, a single
belief can be shared among many different FCMs, so this operation leads to the generalization
and reuse of the obtained results.
In general, explicit revision takes into account the single cognitive components of trust, and
this feature is one of our main desiderata. We derive trust from its cognitive components, i.e.
from single agent's beliefs. So it is better to store information learned by experience into the
same representation form (i.e. beliefs) rather than using compounded values (as an impact
is), in order to integrate them into the representational and reasoning system of the agent.
5
...
5
However, it has to be noticed that since we have a fine-grained distinction between different belief sources, even
the implicit mechanism results in being sufficiently accurate and specific for many purposes, even if it loses part of
Search WWH ::
Custom Search