Information Technology Reference
In-Depth Information
Chapter 7
Simulating Belief and Action
Reason's last step is the recognition that there are an infinite
number of things which are beyond it
Pascal, Pensées, 1670
.
7.1
Modelling Inferences About Observations
In Chap. 6 we noted that the concept of 'truth' was limited to deductive inference
only. For other kinds of inference we have introduced the new concept of 'belief'
to replace 'truth'. Our approach is now to represent experiments with a computer
model that allows for variability in four key components of the scientific process:
•
hypotheses
which are 'believed' or 'disbelieved' by agents,
•
where each hypothesis implicates a
phenomenology
•
that the agents produce by performing
experiments
•
made up of
procedures and apparatus
.
Thus, we have the interacting sets of hypotheses, results and experiments. The mem-
bership of each set can vary according to the hypotheses favoured by the agents. An
example illustrates the approach.
In Gruber's shadow box experiments, subjects are able to see an image that is
a 2-D shadow projection of an object hidden inside it (Fig.
7.1
). Different subjects
may see different shapes cast as different projections of the object (Gruber
1985
). A
cylindrical object can project a circle (endways), a rectangle (sideways), or a range of
capsule shapes (by oblique projection). Gruber was testing subjects' ability to modify
a construction based on their own perception (e.g., of a circular shape) by the different
perception of another subject (e.g. a triangular shape). This required individuals to set
aside the implicit presumption of the superiority of their own perceptions (Gooding
1990
). In some cases subjects could generate a correct solution only by exchanging
observations and conjectures with other subjects. Gruber found that adults are better
at collaboration (trusting the observations of others) than adolescents or children, and
that few subject groups generated more than one construct, even for simple (implied)
objects.
We use tables to relate such constructs (hypotheses) to the observations (data) via
the probability of observing (say) a triangle if the box contains (say) a cone. The
box works by projection so the simplest phenomenology will be that of a perfectly
