Modelling Experiments - Natural and Artificial Reasoning - page 60

These considerations explain the failure of attempts to model theory-change solely
in terms of inference rules operating on observational data. More generally, they
indicate the difficulty of reconciling a universal, principled account of science with
the variability and contextual nature of practice in the many domains of science. To
understand how any rule of inference would work in practice, it would have to be
implemented in a way that reflects the contingent and socially situated character of
scientific thinking. Simulation methods allow us to do this. We can then evaluate
the assumptions of our models of science by playing out the consequences of the
properties of a situation, of inference rules and of the attributes of actors. For example,
an actor may be biased, or insensitive to the opinions of others, or have access
only to certain other actors or to certain experiments. We will advance a stronger
argument that introducing communication into a model of science produces a degree
of complexity that cannot be handled by models defined only as static, semantic
structures.
The context in which beliefs are formed and confirmed or rejected includes mul-
tiple networks of epistemic practices. Science studies recognize the existence of
a range of influences, constraints and sources but reject the notion of a single set
of procedures, rules, norms or institutions sufficient to explain in general terms
how the sciences work. The sciences do produce results—some of which turn out
to be robust—without being constrained by centralized authority, or by standard-
ized protocols, or exclusively by consensual factors, or by an objective world that
determines outcomes. Each actor, site or node of a scientific community has a view-
point, a partial view consisting of beliefs, local practices, local constraints, norms
and resources. None of these are fully shared across all sites. For example, experi-
mental results often cannot be replicated unless expertise (tacit knowledge) is also
transferred (Collins 1985 ). Nevertheless, there is sufficient transfer of descriptors,
concepts, methods and expertise to allow for communication between domains and
between differing theoretical positions (Star and Griesemer 1989 ; Galison 1997 ;
Gorman 2005 ), and for negotiation leading to the aggregation of elements from dif-
ferent viewpoints (Star 1989 , p. 45). This explanation emphasises the diversity and
context of practices, styles, and discourses (Galison and Stump 1996 ). It does not
follow that there are no common methods and strategies, but if many communities
of practitioners conduct science then we cannot expect a finite set of unambiguous
rules to govern belief-revision and theory change.

5.4 Dynamic, Socially Mediated Inference

If complexity and variability defeat formal and semantic models, must we then
conclude that scientific processes cannot be modelled? Here we adopt an alternative
approach. This involves modifying some assumptions of traditional computational
models and of philosophical theories about how confidence in hypotheses relates
to experimental evidence. Drawing on findings of science studies about the social
aspect of belief, we propose a way of modelling the dynamical rationality of science.