Information Technology Reference
In-Depth Information
7.6.1
Non-consultation Run (Solo)
In the following run there is just one actor ('Solo') with access to a simple (single-
image) shadowbox. In this case the act of 'consultation' will be equivalent to 'no
experiment'. On each cycle the object may be observed as either a circle, triangle or
circle point. The plot (Fig.
7.3
) indicates that an initial bias in favour of the pyramid
hypothesis is quickly overcome by increased confidence in the sphere hypothesis. By
cycle 8 both have been displaced by the cone hypothesis. Cone and sphere are again
considered almost equiprobable later in the sequence (cycles 35-40). This indicates
that an emerging bias may be overcome by new evidence. The pattern of events
(self-consultation, experimentation or inaction) that shapes these patterns of belief
is shown in the narrative sequences in Tables
7.9
and
7.10
. At cycle 38, following
a sequence of 'circle' observations, the actor is for the moment undecided between
them.
These narratives (in Tables
7.9
and
7.10
) also illustrate how revising confidence
in hypotheses based on experimental evidence produces behaviour that
appears
to
switch between confirmatory behaviour (as between cycles 32 and 38) and falsifying
behaviour (as between cycles 7-10 and 38-42). Although this fits with strategies used
by scientists (Tweney
1985
), it is misleading to construe the simulation behaviour in
these terms.
The inference process updates confidence in light of recent events (in this run the
event-window
=
3.3 events). Reducing the event-window further to just a single event
(flexibility
1.0) produces more abrupt changes in belief (Fig.
7.4
). However, even
with a single-event window an agent's behavior does not reflect simple falsification.
The narrative (Table
7.10
) shows that the changes plotted in Fig.
7.4
are due to
two experimental results separated by a series of self-consultations. It may seem that
the sphere hypothesis is 'confirmed' at cycle 2 prior to being 'falsified' at cycle 9,
given that a sphere could produce the circle seen at cycle 2 but cannot produce the
pointed oblique projection seen at cycle 9.
However, when an actor can also consult other actors we cannot interpret these
behaviours simply in terms of such rules. This is because actors' belief profiles will
reflect other opinions as well as experimental evidence. Whereas logical models of
confirmation (or falsification) exclude influences other than observational data, our
simulation calculates the consequences of allowing actors to consult. This can be
shown by comparing the 'solo' sequences Table
7.9
with a similar scenario in which
there are two consulting actors in Tables
7.11
and
7.12
.
=
7.6.2
Consultation Run for Two Agents (Actor1 & 2)
We argued in the section '
Why Inference can't be modelled
' that consultation
complicates belief-revision models based on inference rules because consultation
introduces the influence of opinions (rather than evidence) and introduces order
