| |
The empirical criterion operates in the test
of a scientific theory.
At the time of the test of a theory all the
statements in the state description may be viewed a
segregated dichotomously into two classes: those
that are proposed for testing and those that are
presumed for testing. The former are the theory statements. And there may be more than one theory. The latter, the statements presumed true for testing the
theory, are the test-design statements.
Theory statements are included in the state
description, because at least one member of the
profession, presumably the proponent of the theory,
believes that his theory is true.
But the test-design statements are accepted
as true by all the members of the profession, since
these statements supply the semantics that
characterize the problematic phenomena independently
of any theory, identify the cognizant profession,
and define the object language that is relevant and
thus included in the state description.
Execution of the empirical test in accordance
with the previously agreed test-design changes the
state description for the cognizant profession, when
it eliminates one or several theories by a
falsifying test outcome.
By prior agreement the test-design statements
are those that will be regarded as true in the event
of falsification. Regardless of the test outcome, these statements contribute
parts of the meanings of the descriptive terms
common to both the test design and theory
statements. But
the parts of the meanings contributed by the theory
statements change depending on whether or not the
theory was believed true by the particular scientist
before the test, and depending on whether or not the
theory was falsified by the test.
The advocates of the theory believed in it
before the test, and therefore believed that its
statements supplied a true characterization of the
problematic phenomenon in addition to the definitive
characterization supplied independently by the
test-design statements.
Both test design and theory statements
contribute parts to the meaning of each univocal
term common to them, until falsification makes at
least one term equivocal. The falsifying test outcome motivates the proponents and
advocates to reconsider, such that the semantics of
their theory is no longer thought to supply a
characterization of the problematic phenomenon.
However, in the event of falsification some
of the theory's advocates may choose to reconsider
their prior agreement about the defining role of the
test-design statements.
Other members of the profession may dismiss
this behavior as prejudicial or foolishly stubborn. But even if the response to a falsifying test outcome is
merely a stratagem to evade falsification, so long
as the execution of the test is not questioned,
reconsideration of the test design creates a role
reversal between theory and test design.
It redefines the problem into a new one and
in effect creates a new state description, in which
the falsified theory in the old state description
assumes the role of test-design statements in the
new one. Observation language is merely a change of
quantification of some of the universal test-design
statements, such that the reconsideration of the
test design in response to falsification, which
redefines the semantics and reverses the relation
between theory and test design, creates a new
observation language.
This reversal is enabled by the artifactual
character of the semantics of language, which was
noted by Duhem in his thesis of physical theory,
when he stated that a falsifying test does not
locate the error that caused the falsifying outcome.
Furthermore reconsideration is not an
irresponsible evasion of the empirical constraint
discipline, when the recalcitrant advocates propose
a new theory for the new problem, a theory that
purports to explain why the old test design should
be rejected. In
fact this is the outcome of Feyerabend’s
counterinduction thesis, which he illustrated with
Galileo’s creation of a new observation language
for the Copernican theory to extend the Copernican
theory to redescribe the observations used as
objections. The Copernican theory and its extensions became a new
observation language, and Feyerabend is correct in
saying that Galileo had created his own observation
language.
The thesis of artifactuality is contained in
Quine’s "Two Dogmas of Empiricism",
where he stated that it is possible to preserve the
truth of any statement by redistributing
truth-values. Unlike Duhem, Quine does not limit the
artifactual character of language to physical
theory, and he therefore admits to no restriction on
redistribution of truth-values.
But while anything can be reconsidered, not
everything actually is reconsidered, and there
continues to exist semantical continuity due to more
remote and unchanged beliefs. Complete semantic
incommensurability could never occur, even when new
semantic values are introduced. The web of beliefs
is not a logically complete axiomatic system in
which every provable theorem has been derived.
It is a cultural artifact – connected but
perpetually fluctuating and frayed. The propagation
of semantical change is damped by vagueness, logical
inconsistencies, undetected implications and
continuing alterations.
Furthermore the test-design statements may be
modified for reasons other than a falsifying test
outcome. They
may be refined by the addition of statements
describing new test procedures that offer more
accurate measurements or more refined observation
techniques, so that the testing may be more
critical. Feyerabend
notes that this outcome may result from developments
in “auxiliary sciences.”
These new test-design statements have the
effect of reducing the vagueness in the semantics of
the descriptive terms in the test-design statements.
All descriptive language is always vague, and
vagueness can never be completely eliminated, but it
can in principle always be reduced.
Vagueness occurs to the extent that
descriptive terms have not been related to one
another in universal affirmations or negations
believed to be true.
Refining the test design has the effect of
resolving some of the vagueness in the descriptive
terms in the test-design statements, and the outcome
of the consequently more critical test may be the
falsification of previously tested and nonfalsified
theories.
Finally, turn to the topic of scientific
discovery or theory development.
The critical elimination of theories from the
state description by empirical testing requires
consideration of the state description at one point
in time. But
for the constructional introduction of new theories
into the state description, it is necessary to
consider the historically accumulated object
language from both falsified and nonfalsified
theories in many past state descriptions for a given
scientific problem.
This is because falsified theories have scrap
value; their constituent descriptive vocabulary can
be salvaged for new theory construction.
In some circumstances the construction of new
theories can be predicted and therefore effected by
use of the salvaged object language in a cumulative
state description.
Hickey distinguishes three types of theory
construction with the objective of identifying those
circumstances: (1) theory extension, (2) theory
elaboration, and (3) theory revision.
Given a new state description with its
statements of test design that identify its
scientific problem, the first type of theory
construction that the cognizant profession will
attempt for a new problem is theory extension.
This initial conservative response to
falsification suggests Quine’s principle of
“minimum mutilation.”
The existing beliefs give phenomena what
Hanson called their "intelligibility", and
scientists are reluctant to sacrifice
intelligibility by disturbing their current beliefs.
Furthermore language habits are strong, and they
motivate minimizing semantic mutilation.
Theory extension creates minimal disturbance
to current beliefs, and it consists of using the
statements of an explanation already accepted as a
solution for another problem, and then extending
that explanation to address this current problem,
perhaps because the current problem is viewed as a
special case of the solved problem.
This extension is something more than just a
logical transformation. It may consist of relating the explanation to the terms or
variables in the test-design statements for the
current problem by the addition of new statements,
and these new relating statements constitute the new
theory, which is tested and may be falsified.
Falsification of these new relating
statements would not affect the validity of the
employed explanation as an explanation of the
problem that it had already solved.
If successive attempts at theory extension
fail to solve the current scientific problem, then
some of the members of the cognizant profession will
become more willing to depart from the existing
stock of accepted explanations.
But theory extension may also employ analogy
with some currently accepted but unrelated
explanation. The resulting reorganization in the
science in which the new analogy is applied may
produce a new theory that seems quite revolutionary
to the affected profession.
Theory elaboration is the next most
conservative approach. It offers minimal deviance
from accepted explanation, and it involves a
modification to some previously proposed but since
falsified theory for the problem.
The falsification is typically recent and is
motivated in an attempt to save the falsified
theory. The
modification consists of the introduction of some
new descriptive term or variable as a “correcting
factor” or “hidden variable”, that will change
the previously proposed and since falsified theory
thereby transforming it into a new theory.
It may also occasion introduction of new
semantic values, and thus create semantic
incommensurability.
This effort does not “save” the falsified
theory, but instead produces a new one, since the
modification changes the theory’s claim and its
test outcome. Different
members may propose different correcting factors as
strategic in their theories, but their theories will
typically display a recognizable similarity to the
extent that they are basically modifications of
shared older beliefs.
Empirical testing may result in persistent
falsification of theories produced in this
conservative manner.
Some members of the profession will therefore
become more willing to deviate more radically, and
their theory construction will make new theories
that bear increasingly less similarity to past
theories produced by theory extension or theory
elaboration. As
the permutations permitted to theory construction
become greater, the only remaining control on the
exponentially increasing number of constructional
possibilities is the size of the descriptive
vocabulary in the state description. But this size approaches a limit, as the persistent failure
of theory elaboration provides reason to expect that
the solution to the current problem does not consist
in the further search for more still hidden
correcting factors, but instead consists in
restructuring statements containing a selection from
the descriptive vocabulary already in the cumulative
state description, the last vestige of continuity
with the past supplied by the test-design language
and the only remaining available language.
Hickey calls this third type "theory
revision", and he maintains that as increasing
numbers of researchers abandon theory elaboration in
favor of theory revision, the prospects increase for
producing an empirically satisfactory explanatory
solution by the mechanized theory revision of the
object language available in the cumulative state
description. The
key idea in this strategy for mechanizing theory
development is that the descriptive vocabulary that
serves as input has been identified, is small, and
is available. Hickey
notes that the conditions occasioning increased use
of the strategy of theory revision might resemble
something similar to what Kuhn called a
"crisis", and also that theory revision
produces a much more radically new and different
theory, that would readily be called
"revolutionary.” Hickey maintains that the principle of minimal mutilation
dictates that the introduction of new semantic
values does not typically occur during theory
revision, and that the introduction of new semantic
values occurs prior to theory revision.
Therefore since no new semantic values are
involved, there is typically no semantic
incommensurability in revolutionary transitions. Ironically revisionary theory development is most often
viewed as the most mysteriously muse-inspired type,
while according to Hickey's metatheory the
availability of object-language input from a
cumulative state description makes it the type that
is most easily mechanized. Mechanization takes the mystery out of musing.
Hickey does not accept Kuhn's early thesis
that every scientific revolution is a wholistic
gestalt switch to a new "paradigm"
producing an institutional change.
Nor does he accept Feyerabend's radical
historicist thesis that there are semantically
incommensurable revolutionary developments involving
Whorfian covert categories, even when the new theory
uses a new patterning mathematics, or his thesis
that science should be in a state of perpetual
revolutionary change.
In Hickey's metatheory of semantical
description the semantical continuity through theory
revision is exhibited in the unchanged semantical
contribution to the descriptive vocabulary made by
the test-design statements, if as Popper says, one
"sticks to the problem.” And the semantical discontinuity is exhibited by the
semantical contribution to the descriptive
vocabulary by the radically new statements
constituting the new theory.
Due to the semantical continuity, even the
most radical scientific revolution does not create a
completely new world view that is semantically
incommensurable with the past and that ipso
facto constitutes an institutional change.
Thus there is no semantical basis for
maintaining that radical change in theory
necessitates institutional change in the science,
although historically it has on a few occasions
produced such change.
The extent of semantical restructuring in the
new theory produced by theory revision produces a
correspondingly high degree of cognition constraint
for the inventor working with no discovery system,
and a comparably high degree of communication
constraint for the profession with or without a
discovery system.
Furthermore, the contemporary Pragmatist
philosophy of science with its theses of semantic
relativism and scientific realism liberates theory
from any particular semantics and ontology.
This is the institutional change belatedly
recognized by philosophers of science when
confronted with the development of the quantum
theory, although due recognition must be given to
Popper, who earlier concluded that science is “subjectless”,
when he was confronted with the development of the
relativity theory.
When the Romanticist and Positivist
philosophies of science prevailed, on the other
hand, they attempted to make all future scientific
theory metaphysically bound to the prevailing
theory’s distinctive semantics and to the ontology
its semantics described, thereby giving that theory
institutional status.
Any revision of theory therefore actually
required an institutional change in the views and
values in the affected science.
The philosophies of science advanced by Kuhn
and Feyerabend describe institutional views and
values that characterize earlier periods in the
history of physics, when science's institutional
views, as Hanson noted, defined such concepts of
explanation and causality.
Pragmatism avoids this outcome by making
ontological commitment depend exclusively upon
empirical adequacy, rather than including any
ontology in the criteria for scientific criticism.
This practice of scientific realism simply
means that even the more obdurate physicists and
philosophers of science have learned something.
Of course institutional change will continue
to occur in sciences in which Pragmatism prevails,
because it is impossible to predict what the
post-Pragmatist philosophy of science will look
like. But
in those sciences that have not yet matured
institutionally the adoption of the contemporary
Pragmatist philosophy of science will produce an
institutional change: in due course psychology will
drop Positivist Behaviorism and sociology and
neoclassical economics will outgrow their retarding
Romanticism. Then
they will have achieved the maturity they envy in
other sciences.
Hickey's
METAMODEL Discovery System
Hickey is the first philosopher of science to
design and create an artificial-intelligence
discovery system for philosophy of science, although
he is reluctant to call his system
“artificial-intelligence”, since no one knows
what “natural intelligence” means, and since he
furthermore makes no psychological claims about his
system design.
His METAMODEL discovery system constructed while at San Jose College,
San Jose, CA, antedates Simon's applications of his
problem-solving theory of heuristic search to the
problem of scientific discovery by about ten years,
and Hickey’s system has an original design that is
not the same as the heuristic-search discovery
system design used by Simon and his colleagues at
Carnegie-Mellon in the 1980’s or by their later
followers including Thagard.
In his autobiography Simon distinguishes
three types of discovery systems: expert systems,
generate-and-test systems, and heuristic-search
systems. Unlike
Simon's heuristic-search type, Hickey's generative
grammar most closely resembles the generate-and-test
type of system.
The generate-and-test procedure in the METAMODEL
discovery system does not proceed through a lengthy
sequence of dependent decision points.
Instead the design is a combinatorial
procedure that generates and tests independently a
very large number of structured nonredundant
combinations of language elements.
The METAMODEL
is an exhaustive cognitive exploration of
revisionary theory-constructional possibilities that
are latent in the input state description.
The principal disadvantage of the
generate-and-test design is its extensive
utilization of computer resources in comparison to
the heuristic-search design.
On the other hand the principal advantage is
that unlike heuristic search, it does not risk
overlooking or preemptively excluding theories that
are worthy of consideration.
In other words it is not a satisficing
system, but rather is an optimizing system that
outputs a small number of constructionally generated
and empirically tested theories.
As the computer hardware technology continues
to improve, the trade-off between efficiency and
thoroughness will continue to move in the direction
of thoroughness.
Hickey’s METAMODEL system is designed exclusively for creating longitudinal
models.
Hickey's Introduction to Metascience is divided into two parts.
The first part is an exposition of his
metatheory, as described above in its essentials. The second part sets forth the design of his METAMODEL
discovery system together with a description of an
application of the system to the trade cycle
specialty in economics in 1936, the year in which
John M. Keynes published his General
Theory of Employment, Interest and Money.
The METAMODEL
performs revisionary theory construction to
reconstruct the development of Keynes theory, an
episode now known as the "Keynesian
Revolution" in economics.
The applicability of the METAMODEL's
revisionary theory construction for the rational
reconstruction is already known in retrospect by the
fact that, as Lawrence Klein says in his Keynesian Revolution (1966, [1947]), all the important parts of
Keynes theory can be found in the works of one or
another of his predecessors. The METAMODEL
discovery system has an input and an output
state description, and Hickey firstly describes the
cumulative input state description containing the
object language given to the system.
The test-design statements are not explicitly
displayed in the input state description, since they
do not change through the execution of the discovery
system. They
consist of statements describing the phenomena
symbolized by the descriptive variables occurring in
the trade cycle theories that had been proposed by
economists up to 1936, together with the statements
describing the measurement procedures for collecting
the associated data.
The measurement data are those representing
the U.S. national economy, which were originally
published at the time in annual issues of the U.S.
Department of Commerce Statistical Abstract, and since reprinted in their Historical
Statistics of the United States (1958).
Hickey searched both the books and the
periodical literature of the economics profession
for the interwar years prior to 1937, which
pertained to the trade cycle problem.
The American Economic Association's Index
of Economic Journals was a useful bibliographic
source, which also revealed that the number of
journal articles fluctuated in close correlation
with the national average unemployment rate with a
lag of two years.
This examination of the relevant professional
literature yielded ten economic theories of the
national trade cycle, which he translated into
mathematical form.
The ten theories were those of J.A. Hobson,
Irving Fisher, Foster and Catchings, J.M. Clark, F.A.
von Hayek, R.G. Hawtrey, Gusatv Cassel, Gunnar
Myrdal, Johan Akerman, and A.C. Pigou.
The descriptive vocabulary occurring in
these theories was a highly redundant, and yielded a
set consisting of eighteen variables.
The data for these variables are annual time
series for the period 1921 through 1934, which were
available to any economist in 1936.
These time series data were converted to
index numbers of period-to-period change rates, and
together with variable names including one time lag
are the input to the METAMODEL
discovery system for the historical simulation.
The output state description was expected to
contain an econometric model of Keynes theory
constructed by the discovery system.
Therefore Keynes' theory like the other
theories was translated into mathematical form.
The theory is actually a static theory, but
it was made dynamic by including considerations
contained in an appendix to the General
Theory titled "Notes on the trade
cycle", in which Keynes explicitly applies his
theory of income determination to the phenomenon of
the trade cycle.
Keynes theory contains ten variables and
seven equations with three exogenous variables.
All ten variables occur in more than one of
the preceding trade cycle theories, and most in
several of them.
There is no question that all the variables
needed for a recognizably Keynesian theory are
available in the existing literature in 1936.
|
|
