Information Technology Reference
In-Depth Information
The range of queries given provides a basis for the process of question and inference
required to update the model.
We (David Billinge and myself) use the initial decisions made by the participants
during their evaluation of the music heard in experimental sessions to initiate the
model. We then compare the predicted order of ally choice made by each agent in the
model as assessed from the agent's perception of others, as drawn from a simulated
conversation, with the actual outcomes of the sessions. We plan to modify the model
to show from recorded conversations of the participants in our experimental sessions
how the patterns of questioning compare with that generated by the model. The model
has
n
actors and
m
aesthetic objects. The
m
objects in our experiments are pieces
of classical orchestral music to which our actor/participants have a response. Actors
have a response-scale for each piece of music representing their own subjective
impression. Further, an actor has a separate scale of belief for the response to each
piece of music for each other actor as derived from the conversation. In the model the
actors can ask questions in turn and can only ask one question per turn. Only the actor
asking the question can update its scales of belief from another actor's response.
We made three assumptions:
first,
that each actor accepts that other actors initially
have the same perceptions of the heard music and thus the same ratings;
second,
that
each modelling actor tends to ask the other actor about the music of which it, the
questioner, has the most uncertain belief scale (e.g. highest entropy—see Chap. 6);
and
finally
that the modeller can have no doubt of its own experience. In this way an
actor tracks the subjective experience of others; an experience that may change over
time. We also assumed that, as supported by our observation of the experimental
work with people, an actor will choose as an ally the fellow actor who is closest in
m
scale distance. The significance of this result is that it should show that a subjective
1
experience can be inferred through conversations and we also suggested that this
might actually be the major purpose of peoples' discussions.
Finally we describe how we expect to adapt the model to fit in with observations
to take into account other factors that decide group decisions. Our model should
thus become more able to predict conversational behaviour and final group decisions
from knowing individuals' perceptions within this scenario.
14.4
Analysis of the Experimental Results
Seven experiments were carried out with four people and one with only three peo-
ple. Where ally-ranking information is important only the seven fully populated
experiments are considered here.
The basic ordinal and numerical data from these experiments includes:
•
Each subject's personal scale for each piece of music.
•
The order in which the music was played.
1
An evaluation, of a potentially shareable event, that is accessible only by a single actor and related
to that actor's observation and assessment of that event.
