Simulating Belief and Action - Natural and Artificial Reasoning - page 93

developments that can be discerned according to historical evidence. We would also
programme changing sets of probabilities to govern actors' access to existing and
new apparatus and techniques, and to other actors as these enter or leave the scene
(e.g. Lorentz, Fitzgerald, Poincare, Einstein, Miller).

7.4 Simulating Experimental Science

We argue elsewhere (see Chaps. 4 and 5, also Addis and Gooding 2004 ) that there is
no a priori reason to prioritise either the hypotheses (treating phenomena as deductive
consequences) or the phenomena (from which hypotheses are induced or abduced).
In science, hypotheses and phenomena develop jointly by a dynamic that combines
inductive, abductive and deductive forms of inference. To consider the dynamics we
shall examine examples based on simulation of the Gruber shadowbox. In Gruber's
experiments, subjects have the option of consulting another subject. At every iteration
stage of our simulation, each agent decides whether to experiment, consult or do
nothing, and must select an experiment to perform or an agent to consult. The decision
procedure is complex. An agent is motivated to reduce the uncertainty in its view of
the world (see Chap. 6, Addis and Gooding 1999 ; Gooding and Addis 1999 ). This
view consists of the agent's changing belief profile (a set of confidence values) and
dynamic variables derived from this (e.g. scientist's belief entropy or indifference,
or probability of consulting another scientist is reduced). Uncertainty consists of not
knowing which hypothesis best describes the world. A scientist with a flat belief
profile (having no bias towards a particular hypothesis) is less confident of his view
of the world than is a scientist who is confident about only one or two hypotheses.
The belief-revision simulation tracks and reports agent (a simulated scientist)
behaviour in two ways: as macro-behaviour, a trajectory of changing confidence
in each hypothesis (a changing belief-profile , as plotted in Fig. 7.3 ) and as micro-
behaviour, a narrative or history of actions . In Figs. 7.4 and 7.5 , Table 7.10 the agent's
display which of two experiments are selected (and the outcome) or which of two
agents is consulted. During the process of experimenting or consulting an agent's
indifference levels changes and is shown as a dynamic threshold I n (A). This dynamic
threshold defines which of the hypotheses are believed (see Chap. 6).

7.5 How Experiments Mediate Between Hypotheses and

Phenomenology

An experiment consists of a physical setup that is acted on by an agent procedure.
Each apparatus-procedure pairing can produce a range of results. Which results can
occur also depends on which hypothesis is being considered. So in our computer
simulation, the range of possible results is expressed as a probability distribution for