| |
Newell and Simon have developed many
artificial-intelligence systems, several of which
are described in their book titled Human
Problem Solving (1972).
Simon views scientific discovery as a special
case of human problem solving, and therefore
maintains that it can be examined with the
artificial-intelligence approach.
However, his artificial-intelligence systems
development work was not directed to scientific
discovery until later in the 1970's.
His principal publications pertaining to
scientific discovery are Models
of Discovery
(1977), which contains reprints of some
of his earlier articles relating information
processing concepts to scientific discovery, and
most notably his Scientific Discovery; Computational Explorations of the Creative Process
(1987), which describes several discovery systems
that rediscovered various historic scientific laws
and theories.
Just as examination of the evolution of the
contemporary Pragmatist philosophy of science
requires consideration of the issues in physics and
especially quantum theory, similarly examination of
the development of the artificial-intelligence
discovery systems requires consideration of issues
in the social sciences and especially economics.
Therefore, to appreciate Simon's views on
scientific discovery, it is necessary to consider
his views on human problem solving by
artificial-intelligence systems.
And to appreciate his views on human problem
solving, it is informative to consider what he calls
his most important contribution to economics, his
postulate of bounded rationality.
And to appreciate Simon's postulate of
bounded rationality, it is helpful firstly to review
both the various alternative rationality postulates
and Max Weber’s semantical thesis of “ideal
types.”
Neoclassical
Maximizing Rationality and Weber's Ideal Types
Simon proposes his thesis of bounded
rationality as an alternative to two other concepts
of rationality that have currency among economists.
The principal alternative to Simon's view is
the prevailing neoclassical postulate, which says
that consumers are rational because they maximize
their utility, and that producers are rational
because they maximize their profits.
The second alternative to Simon's is the
rational expectations postulate, which is a
distinctive extension of the neoclassical postulate
of utility and profit maximization.
The rational expectations view will be
considered below in the discussion of the BVAR
type of discovery system.
And since the rational expectations postulate
is an extended version of the neoclassical view,
Simon's critique of neoclassicism also applies to
the rational expectations thesis, which he
explicitly rejects.
Simon’s bounded rationality postulate is
similar to an earlier view originating in the U.S.
called “Institutionalist economics”, which will
also be examined below.
Before turning to Simon's bounded rationality
postulate, however, consider firstly the still
prevailing view in academic economics, the
neoclassical rationality postulate.
The neoclassical postulate of rationality has
its origins in Adam Smith's doctrine of self
interest set forth in his Wealth
of Nations (1776), the seminal document for
modern economics.
Smith was greatly impressed by Isaac Newton's
celestial mechanics.
In his Essay
on the History of Astronomy Smith described
Newton's celestial mechanics as the greatest
discovery ever made by man, and Smith aspired to
describe economic life as a harmonious mechanism, as
Newton had done for the heavens.
In Smith's system entrepreneurs' rational
behavior in pursuit of their economic self-interest
unintentionally produces a beneficial and harmonious
outcome for the national economy; this is his
doctrine of the "invisible hand.”
However Smith's perspective is not closed or
self-contained.
It was part of a larger moral universe of
natural laws, which Smith had earlier described in
his Theory of Moral Sentiments (1759).
In Smith's natural-law philosophy the pursuit
of economic self-interest is morally constrained by
men's natural sympathy for others and also by their
natural desire for the approval of others - a
distinctively sociological idea.
Later economists excluded from theoretical
economics Smith’s moral constraints on the pursuit
of self-interest.
In the twentieth century these constraints
came to be recognized as sociological or
institutional structures instead of natural moral
laws, and an attempt to re-introduce them into
economic analysis was made by the American
Institutionalists.
Almost one hundred years after the Wealth
of Nations a new development occurred in
economic theory, which is now called the "marginalist
revolution", and which might also be described
as the completion of Smith's agenda for a Newtonian
economics.
The term "marginal" means
incremental or differential, and the incremental
economic analysis lends itself to mathematical
expression with the differential calculus developed
by Newton.
The result is an elegant mathematical
rendering of economic theory, in which the
rationality postulate became a matter of calculating
the global maximization of consumer utility and
producer profits by suitable values for the first
and second derivatives of the relevant
mathematically expressed demand functions.
The resulting theory of price determination
describes the allocation of goods and services in an
optimally efficient manner later called "Pareto
optimality" after the Italian economist,
Vilfredo Pareto.
A half century later there was another
revolution called the "Keynesian
revolution" named after the economist, John
Maynard Keynes.
Pre-Keynesian economic theory had assumed
that the Pareto optimum resulting from rational
maximizing behavior by each individual would also
maximize income and output for the whole economy, as
Adam Smith and the marginalists had believed.
In his General
Theory (1936), however, Keynes set forth a new
thesis saying that individual maximizing behavior
could result in less-than-full-employment
equilibrium, which he said had occurred during the
Great Depression of the 1930's.
This resulted in economists’ dividing
economics into the “microeconomic” theory of
price determination and the “macroeconomic”
theory of national income determination.
Keynes thus produced a revolution in economic
theory, but he did not explicitly attack the
classical economists’ rationality postulate of
individual human behavior.
His stagnation thesis of underemployment
equilibrium only attacked the classical
economists’ optimistic thesis of a maximizing
macroeconomic outcome.
Soon afterwards economists began applying
statistical inference techniques to estimate
equations with the macroeconomic data that were
being collected by Nobel Laureate economist Simon
Kuznets of the National Bureau of Economic Research,
in order to describe national economic conditions.
Both the availability of these data and the
development of the computer occasioned the evolution
of a specialty area in economics called
"econometrics", although earlier there
were Institutionalist economists whose statistical
analyses of economic data have also been called
econometrics.
Since Haavelmo, however, nearly all the
econometricians have been neoclassical economists,
who require that the selection of variables for the
equations constituting the econometric model be
“justified” by neoclassical theory.
Thus, until very recent years econometrics
was exclusively the application of statistical
inference and testing techniques to economic data
structured by neoclassical microeconomic and
macroeconomic theory.
Even today any econometric model that does
not result from such a
priori imposition of the neoclassical theory
upon the data is derisively referred to as "atheoretical.”
In this respect neoclassical economics still
bears a burdensome legacy from the Romantic era in
the history of European culture.
The above overview of the neoclassical
rationality postulate of human behavior reveals that
it is not viewed by economists as just one of many
alternatives; it has served as the foundation for
modern economics since its founder, Adam Smith.
Anyone attempting to overthrow the use of
maximizing rationality postulates is attempting a
new scientific revolution in economics that would be
much more radical than any of the revolutionary
developments within the history of neoclassical
theory.
Nevertheless, there have been dissenters such
as the American Institutionalists, and the reason
for their dissent has always been the empirical
inadequacy and heroic unrealism of the neoclassical
theory with its basis in rationality postulates.
Neoclassical theorists have not been
completely unaware of these problems caused by their
maximizing rationality postulates.
Before turning to Simon's alternative,
consider briefly Max Weber's thesis of the
"ideal type", a semantical contrivance
proposed to defend the neoclassical rationality
concept against its critics.
Simon does not refer to Weber, but Weber
explicitly proposes the same ideas that Simon
explicitly opposes.
Weber's discussion of his doctrine of the
ideal type or "idealtypus"
can be found in English translation from the German
in The Methodology of the Social Sciences (Tr. by Shils and Finch,
1949), and principally in the chapters titled
"'Objectivity' in Social Science and
Social Policy" and "The Meaning of
'Ethical Neutrality' in Sociology and
Economics", and in Max
Weber's Ideal Type Theory (1969) by Rolf E.
Rogers.
Weber’s philosophy of sociology contains
ambiguities that have been noted by recognized
Weberian scholars including “Weber’s dilemma”,
which is discussed below. The ideal type is
distinctive of the interpretative understanding to
which cultural sociology aims.
Weber defined the ideal type as a mental
construct that has two basic features: Firstly it
involves one or several points of view. According to
Weber's theory of knowledge this perspectivism is
characteristic of all concepts including both
natural science and social science concepts, because
no concept can capture reality in all its
potentially infinite variety of aspects. Weber
explicitly rejects the copy theory of knowledge,
which he finds in the Historicist philosophy of
social science, and he refers to the Historicists'
claim of pure objectivity in science as the
"naturalistic prejudice.”
In the present context what is noteworthy is
that the rational aspect of human behavior is a
central aspect and perspective of reality that Weber
includes in the ideal-type concepts in pure economic
theory.
The second of the two features of the ideal
type is that it involves a one-sided accentuation or
intensification of the perspective or point of view
in the ideal type concept.
Nonrational considerations are not denied,
but the maximizing postulate is knowingly made
unrealistically extreme as a limiting case.
Weber explicitly rejects the charge that the
ideal type is a complete fiction, but he calls it
"utopian", since historical concrete
individuals do not conform in their behavior to the
accentuated, maximizing rationality described by the
ideal type.
Thus individual instances not conforming to
pure economic theory do not falsify the theory
containing the ideal-type concepts; as Weber states,
the ideal type is not a hypothesis, and it is not
tested by its application to reality.
Weber says that the ideal type is used to
compare theory to reality, in order to reveal by
contrast the irrational aspects of human behavior.
What neoclassical economists call “pure
theory” utilizes ideal-type concepts exclusively,
and it makes certain assumptions, notably the
maximizing assumptions, which almost never
correspond completely with reality but rather
approximate reality in varying degrees.
The ideal type is a semantical contrivance
like Heisenberg’s concept of a closed-off theory,
because it is what Popper calls a
“content-decreasing stratagem” to evade
falsification.
It is unfalsifiable, because it is protected
from falsifying evidence by the stratagem of
restricting its applicability in the face of
contrary evidence and thus of denying its
falsification.
Pure economic theory with its ideal-type
concepts is true where it is applicable, and it is
not nonapplicable wherever it would be falsified.
In other words all observed human behavior is
“rational” and suitable for economic analysis
wherever neoclassical economic theory applies to it,
and it is “irrational” and unsuitable for
economic analysis wherever the theory does not
apply.
If there is anything that distinguishes the
ideal type thesis, it is that the evading denial of
the falsifying consequence of contrary evidence is
very explicit.
It may also be noted that when the Weberian
neoclassical economist compares his ideal type with
observed behavior in order to detect irrational
behavior, he is not using it as a counterinductive
"detecting device" as Feyerabend
advocates.
When Galileo was confronted with the
Aristotelian tower argument opposing the Copernican
heliocentric theory, Galileo’s response was to
revise the language describing observation.
And when Heisenberg was confronted with the
apparently Newtonian track of the free electron in
the Wilson cloud chamber, his response too was to
revise the Newtonian language describing the
observation.
These are examples of counterinduction.
But when the Weberian neoclassical economist
is confronted with observed anomalous
"irrational" behavior, no attempt is made
to reconcile the reporting language of observation
with the ideal-type language of neoclassical theory,
much less to revise the theory.
Instead the reported anomalous observations
are simply excluded from economics.
The Weberian regards the observed
"irrational" behavior as a phenomenon to
be excluded from neoclassical theory rather than as
one to be investigated either for a more empirically
adequate post-neoclassical economic theory or for a
new test-design language.
Many contemporary economic theorists are only
less explicit in their dogmatic adherence to
neoclassicism with its definitive maximizing
postulates.
They are reluctant to dispense with the
elegantly uniquely determinate mathematical
solutions enabled by merely setting the first
derivative to zero and checking the second
derivative for a maximum inflection point.
They are scandalized by the observed absence
of optimizing behavior and the rejection of their
maximizing postulates, because it implies that
paradigmatic problems thought to have been solved
elegantly after two hundred years of theoretical
development in the neoclassical tradition have not
actually been solved at all.
Academic economists who have dutifully
labored and groveled for years to earn their
doctorate credentials and publish their papers in
the prestigious refereed journals, do not welcome
being advised that their purportedly empirical
theory depends on a content-decreasing stratagem, a
self-deceiving linguistic contrivance, which makes
their received theory only slightly less
semantically vacuous than the formal differential
calculus used to express it, and hardly more
ontologically realistic than the Ayn Rand
Romanticist utopian novels used to propagandize it
for the nonprofessional general public.
Yet in truth not all economists are
philosophically atavistic neoclassicals.
In recent years the ascendancy of
econometrics has made such evasion of empiricism
more difficult, because the “rational” and the
“irrational” are inseparably commingled in the
measurement data.
The econometrician constructing models from
time-series historical data would prefer to make
statistically acceptable models, than to dismiss
large error residuals in his statistical equations
as merely “irrational” behavior that can be
ignored.
While the ostensible practice in academia
today is still the Haavelmo agenda (discussed
below), in which equations are specified on the
basis of neoclassical theory, a growing number of
economists are evolving into closet Pragmatists.
They have turned increasingly to empirical
data analysis for the determination of their
equation specifications, and they include in their
equations even noneconomic variables such as
demographic, sociological or political variables,
which are never found in textbooks preaching
neoclassical microeconomic or macroeconomic theory.
And some economists, such as Simon, are so
heretical as to reconsider even the sacrosanct
maximizing rationality postulates axiomatic to the
neoclassical orthodoxy.
Simon's
Postulate of Bounded Rationality and "Satisficing"
In his autobiography Simon relates that in
what he calls his first piece of scientific work, a
study in 1935 of public recreation in the city of
Milwaukee, he saw a pattern that was the seminal
insight for what was to become his thesis of bounded
rationality.
For this study he was examining the budgeting
process for the division of funds between playground
maintenance, which was administered by one
organization, and playground activity leadership,
which was administered by another organization in
the Milwaukee municipal government.
He found that the actual budget allocation
decision was not made as economic theory would
suggest.
What actually occurred was that both of the
two organizations wanted more funds for their proper
functions, and he generalized from his experience
with this budgeting decision, that people bring
decisions within reasonable bounds by identifying
with partial goals for which their own
organizational units are responsible.
This insight was taken up in Simon's Ph.D.
dissertation (1942), which he later published as Administrative
Behavior (1947), the book referred to by the
Royal Swedish Academy of Sciences as an
"epoch-making book", when they awarded him
the Nobel Memorial Prize for Economics in 1978.
In his autobiography Simon says that his
entire scientific output may be described as a gloss
on two basic ideas contained in his Administrative Behavior:
(1) human beings are able to achieve only a very
limited or “bounded” rationality, and (2) as a
consequence of this limitation, they are prone to
identify with subgoals.
The first of these ideas is fundamental to
Simon's critique of neoclassical rationality, and
the second is fundamental to his theory of human
problem solving.
In his autobiography Simon says that his
"A Behavioral Model of Rational Choice"
(1955) reprinted as chapter fourteen in his Models
of Man (1987), was his first major step toward
his psychological theory of bounded rationality.
In that early paper he states that the
neoclassical concept of rationality is in need of
fairly drastic revision, because actual human
behavior in making choices does not satisfy three
basic assumptions underlying neoclassical maximizing
rationality.
Those three assumptions are: (1) a decision
maker has a knowledge of the relevant aspects of his
environment, which if not absolutely complete, is at
least impressively clear and voluminous; (2) a
decision maker has a well organized, consistent, and
stable system of preferences; and (3) a decision
maker has a skill in mental computing, that enables
him to calculate for the alternative courses of
action available to him the one course that will
enable him to reach the highest achievable point in
his preference scale. Then in his "Rational
Choice and the Structure of the Environment"
(1956) reprinted as chapter fifteen of Models of Man, Simon proposes replacing the neoclassical postulate
of maximizing behavior with his more modest
postulate that he calls "satisficing"
behavior.
“Satisficing” means that instead of
optimizing, the decision maker's limited information
and limited computational ability require that he
adapt "well enough" to achieve his goals
instead of optimizing.
The first chapter of his Sciences
of the Artificial (1969) reveals that Simon
identifies exactly the same things about
neoclassical rationality that Weber identified as
the two basic features of the ideal-type concept.
Firstly like Weber's thesis of viewpoint in
the ideal type, Simon calls neoclassical rationality
an "abstract idealization", because it
selectively directs attention to the circumstances
of the decision maker's outer environment for his
adaptive behavior.
Similarly in the chapter "Task
Environments" in his Human
Problem Solving (1972) he says that it is the
task that defines the "point of view"
about the environment, an idea that is comparable to
Weber's thesis that the ideal type contains a point
of view determined by one’s interests.
Secondly, just as Weber said that the
accentuated rationality in the ideal type is
"utopian", Simon calls neoclassical
rationality "heroic" to describe its
unrealistic character, and later in 1983 in his Reason in Human Affairs again without referencing Weber, he
describes optimization as "utopian.”
But unlike Weber, Simon does not relegate to
the status of the "irrational" all the
decision making that does not conform to the
neoclassical ideal type of rational maximizing
behavior.
Instead Simon considers the empirical
inadequacy of neoclassical rationality to be good
reason for replacing it with his behaviorally more
realistic concept of bounded rationality.
In the second chapter of his Sciences
of the Artificial and then in his "From
Substantive to Procedural Rationality" in
Models of Bounded Rationality Simon uses the
phrase "substantive rationality" for the
neoclassical maximizing rationality, which considers
only the decision maker's goals and outer
environment.
And he uses the phrase "procedural
rationality" for the satisficing psychological
cognitive procedures including the decision maker's
limited information and limited computational
abilities consisting of what Simon calls the
decision maker's inner environment.
The study of cognitive processes or
procedural rationality is interesting only when the
substantively rational response is not trivial or
instantly obvious.
It is usually studied in situations in which
the decision maker must gather information of
various kinds, and must process it in different ways
to arrive at a reasonable course of action for
achieving his goals.
Simon refers to the Pareto optimality
described in the economists' theory of general
equilibrium, which combines the individual
maximizing choices of a host of substantively
rational economic actors into a global optimum for
the whole economic system, as the "ideal"
market mechanism.
Then he says that there is also a
"pragmatic" market mechanism described by
the Nobel laureate economist Frederich von Hayek
that is more modest and believable, because it
strives for a measure of procedural rationality by
realistically tailoring decision-making tasks to the
limited computational capabilities and localized
information available to the economic decision
maker, with no promise of optimization.
He quotes at length a passage from Hayek's
"The Uses of Knowledge in Society" in American
Economic Review (1945), in which Hayek asks:
what is the problem we wish to solve when we try to
construct a rational economic order?
And Hayek answers that the economic calculus
does not describe the optimization problem, since it
is a problem of the utilization of knowledge that is
not given to anyone in its totality.
The price system is a mechanism for
communicating information, and the most significant
fact about it is the economy of knowledge with which
it operates, that is, how little the individual
participants need to know in order to be able to
take the right course of action.
Simon maintains that it is Hayek's
"pragmatic" version which describes the
markets of the real world, and that the substantive
rationality of neoclassical theory is worthless,
since it is not backed up by executable maximizing
algorithms.
He says that consumers and business firms are
not maximizers, but rather are satisficers.
They accept what is "good enough"
because they have no choice.
The rationality that they actually use is a
satisficing procedural rationality.
Examination of the limits of rationality
leads to consideration of the price system mainly as
an institution that reduces the amount of nonlocal
information which economic actors must have to make
"reasonable", i.e. satisficing, decisions.
|
|
