Seeking Allies - Natural and Artificial Reasoning - page 178

min. Each person was asked to keep abbreviated notes on what they heard and to rate
the music on a single linear scale ranging from 1 to 10 in units of 1. Evaluation of
two pieces of music could be the same. They were then asked to form a committee
to discuss the music with the purpose of recommending that one of the pieces should
be included in a concert programme. The discussion was recorded. For completion
the committees were asked to arrive at a group opinion and rank all four pieces of
music on a scale 1-10.
The final part of the experiment, and as it turned out the most significant, was
to ask the individual participants to rank their fellow committee members in terms
of whose judgement they would take the most notice of when deciding to go to a
concert. This will be referred to as 'ally choice'.
The design of the experiment was primarily to explore the use of metaphor during
the discussion. The actual recordings taken during these discussions have yet to be
fully analysed. However, Dr. David Billinge and I were primarily interested in the
relationship between the music and ranking. We were also interested in how the
discussions might influence opinion, hence the individual-before and the group-after
rankings of ally choice. The test of the effectiveness of a person's internal models of
others was to be assessed from the individual ranking of a chosen advisor.
The delay in analysing the discussions was taken because we wished to view these
data in the light of a computer model of inferential semantics. There are arguments
to suggest that this is bad practice, since we are prejudicing our observations with
the computer model. Such prejudice will cause us to observe only that which will
support the model and thus our observations will be tainted and in doubt. We reject
such an argument because the history of science supports the need for an initial
theory (Kuhn 1985 ), provided such a theory can be tested and has the possibility of
being rejected (Popper 1959 ). Theories are particularly helpful in observing complex
situations, such as group discussions, because they do limit what should be observed.
The real test of a theory has been discussed above, but a theory also plays a further
important role by providing a basis for puzzlement and modification. If you don't
have a theory then you cannot be puzzled by what you observe. We therefore, required
a satisfactory model of discourse that could do the job of limiting what we observed
(at least initially) in the conversation.

14.3 The Computer Model

The process that models the inferential semantics is driven by the belief system
(Chap. 7, Addis and Gooding 1999 , 2008 ). The belief system, in this case, models
each user as a single undefined dimension of values for each piece of music involved
in a discussion. The values on the dimension are discrete and ordered. Each value is
considered an independent hypothesis that has a level of belief associated with it. The
actions open to the belief system, in order to update these associated beliefs, is limited
as to a fixed set of queries that may be addressed to another person. 'Person' in this
case is a part of the computer program that is normally referred to as 'Agent', but in
this chapter we will use the more appropriate and original term 'Actor' (Hewitt 1979 ).