Information Technology Reference
In-Depth Information
In recognising the social dimension of inference, we accept that patterns of be-
lief revision depend not only on evidence, but also on the influence of others. This
introduces order effects. Similar sets of tables would be needed for subjects B and
C, detailing the pattern of construction and the probabilities of each possible obser-
vation, for every possible sequence of observation and consultation. For this very
simple case the first iteration would require two observation-construction tables for
each subject and could be calculated by hand. But for even a modest degree of com-
plexity simulation methods must be used. In science there are well-defined pathways
of communication and constraints on communication, but there are also many sources
of contingency (Knorr-Cetina
1981
). When we factor in real-science contingencies
such as access to experimental setups and the accessibility of other observers, the
situation becomes more complex still.
Confirmation theories and cognitive models of inference do not encounter such
difficulties because they treat agents in isolation, as do experimental designs based
on such models (e.g. Wason
1960
). These approaches fail to represent one of the most
important features of learning and belief-formation—its social character. It could be
argued that scientists use rules of inference and heuristic principles that constrain
variability due to social interaction and individual personality so that (say) actor
A
would always choose to make the most decisive experiments or to consult the most
competent actors.
This objection simply reasserts an article of faith of traditional philosophies of
science: for rationalists, that there are objective (i.e. universal or non-contextual)
principles of reasoning about observations, and for empiricists, that evidence even-
tually eliminates false hypotheses in favour of the true one. A further objection might
be that inference rules (other than the Bayes rule) apply only when all the evidence
is in. The entire set of hypotheses and associated observations could be included in
a single table that relates hypotheses to phenomena
via
sets of probabilities, rep-
resenting a final state of pooled knowledge. But this requires that all actors have
simultaneous access to reports of every type of observation. It would reveal nothing
about the dynamics of the process according to which each actor revises its beliefs
in response to each observation or consultation. Tables
7.2
-
7.4
illustrate a sequence
whereas information is acquired so hypotheses are introduced, amended or rejected.
7.3
A Historical Example
Simulated experiments will have those outcomes that are most probable in a world
in which the active, objective hypothesis is true. We argued in Chap. 5 that sci-
entists change the world by improving their experiments and engaging with other
researchers. In practice, even where an experiment is designed to test a particular hy-
pothesis, this development turns up additional hypotheses relating to other processes
that are implicated by the method of investigation. An example is the Michelson-
Morley-Miller experiments to detect ether-effects between 1880s and the 1930s.
These experiments were designed to detect a difference in the velocity of light due to
