Information Technology Reference
In-Depth Information
allows us to specify the impact of recent evidence and consultation for each agent.
This allows a hypothesis to remain available for consideration despite a run of
apparently falsifying observations (for example see Gooding and Addis
2004
).
This important virtue is in keeping with scientific practice (see, e.g., Kuhn
1977
;
Lakatos
1970
).
•
Second
, we reject the Bayesian assumption that the order in which events occur is
irrelevant, insofar as changes in belief will change the set of hypotheses available
for the interpretation of results (see Sect. 6.4).
Changes in belief can be induced by consultation as well as experimental evidence.
So different sequences of action performed by the agent will produce different pat-
terns of belief-revision in that agent (Gooding and Addis
2004
). Even in the case of
a single (non-consulting) agent, the order in which evidence appears is not predeter-
mined. Iterative, simulation-based methods allow us to explore the effects of what
sociologists of science call 'contingency' (Knorr-Cetina
1975
). Thus, they offer a
huge advantage over single-step discussions of the implications of a particular infer-
ence rule or confirmation strategy applied to a single sequence of events (Gooding
and Addis
2004
).
6.3
The Impact of Evidence: Hypotheses
It follows that we cannot use an inductive inference rule (such as Bayes rule) to
calculate the accumulation of evidence for a hypothesis. There is a long tradition
(dating from the work of Ramsey and de Finetti in the 1930s) of associating disbe-
lief with a prior probability of (or near) zero and certainty with a probability near
one. However, hypotheses do not achieve absolute certainty and they should always
remain hypothetical; scientists will re-classify them as a necessary postulate (as in
the case of the luminiferous ether), as a principle (as both Galilean-Newtonian and
Einsteinian relativity were), or as facts (e.g., elements transmute, species mutate).
Nor can a hypothesis be wholly, irretrievably disbelieved and still be hypothetical;
without belief, a hypothesis will be re-labelled as an artefact, a fiction or non-fact,
as a non-existent entity (phlogiston, the ether), or as a false (though once-believed)
principle (e.g., the immutability of chemical elements and of biological species).
To call something a hypothesis, H is to say that there is some empirical support for
H given evidence R
e
. This support is E
n
−
1
(H/R
e
), and is a value that lies between 0
and 1 for each hypothesis H. Given a new result R
e
from carrying out an experiment
'e', the probability which represents an agent's confidence concerning a particular
hypothesis can be modified to E
n
(H) by adapting the above equation to the following
one (see Addis
1985
, p. 260):
=
−
+
E
n
(H)
[(N
1)E
n
−
1
(H)
E
n
−
1
(H/R
e
)]
/
N
