Complexity and the Economy
Page 29
[ 156 ] Complexity and the Economy
REFERENCES
1. Bonner, J. T. The Evolution of Complexity. Princeton: Princeton University Press, 1988.
2. Constant, E. W. Origins of the Turbojet Revolution. Baltimore: Johns Hopkins University Press, 1980.
3. Darwin C. From Charles Darwin’s Notebooks, edited by P. H. Barrett et al., 422.
Ithaca: Cornell University Press, 1987.
4. Heron, S. D. History of the Aircraft Piston Engine. Detroit: Ethyl Corp., 1961.
5. Holland, J. Adaptation in Natural and Artificial Systems, 2nd ed. Cambridge: MIT
Press, 1992.
6. Holland, J. “Echoing Emergence: Objectives, Rough Definitions, and
Speculation for Echo-Class Models.” Mimeograph, University of Michigan,
1993.
7. Kauffman, S. “The Sciences of Complexity and Origins of Order.” Working
Paper 91-04-021, Santa Fe Institute, 1991.
8. Koza, J. Genetic Programming. Cambridge, MA: MIT Press, 1992.
9. Lieberman, P. The Biology and Evolution of Human Language. Cambridge, MA: Harvard University Press, 1984.
10. Lindgren, K. “Evolutionary Phenomena in Simple Dynamics.” In Artificial Life II, edited by C. Langton, C. Taylor, J. D. Farmer, and S. Rasmussen. Santa Fe Institute Studies in the Sciences of Complexity, Vol. X, 295–312. Reading,
MA: Addison-Wesley, 1991.
11. McShea, D. “Complexity and Evolution: What Everybody Knows.” Bio. & Phil.
6 (1991): 303–324.
12. Morowitz, H. Beginnings of Cellular Life. New Haven: Yale University Press, 1992.
13. Müller, G. B. “Developmental Mechanisms at the Origin of Morphological
Novelty: A Side-Effect Hypothesis.” In Evolutionary Innovations, edited by Matthew Nitecki, 99–130. Chicago: University of Chicago Press, 1990.
14. Ray, T. S. “An Approach to the Synthesis of Life.” In Artificial Life II, edited by C. Langton, C. Taylor, J. D. Farmer, and S. Rasmussen. Santa Fe
Institute Studies in the Sciences of Complexity, Vol. X, 371–408. Reading,
MA: Addison-Wesley, 1991.
15. Stanley, S. M. “An Ecological Theory for the Sudden Origin of Multicellular Life in the Late Precambrian.” Proc. Nat. Acad. Sci. 70 (1979): 1486–1489.
16. Waddington, C. H. “Paradigm for an Evolutionary Process.” In Towards
a Theoretical Biology, edited by C. H. Waddington, Vol. 2, 106–128.
New York: Aldine, 1969.
on t He evolu t ion of comPlexi t y [ 157 ]
CHAPTER 10
Cognition
The Black Box of Economics*
W. BRIAN ARTHUR
Most economists agree that when human agents face decision problems of complication or ones that contain fundamental uncertainty they don’t use deductive rationality. But what do they use in its place? Cognitive science tells us that in such situations we think associatively—we find similar situations from our repertoire of experiences and fit these to the problem in question, then derive implications from there. This essay explores the consequences for economics of this type of reasoning and it suggests ways such reasoning can be built into economic models. It argues that because a wide collection of memories and experiences of situations is necessary to our reasoning, economics students should be deeply versed in economic history, not just economic theory.
The essay appeared in The Complexity Vision and the Teaching of Economics, edited by David Colander, Edward Elgar Publishers, Cheltenham, UK, 2000.
In his autobiography Bertrand Russell tells us he dropped his interest in eco-
nomics after half a year’s study because he thought it was too simple. Max
Planck dropped his involvement with economics because he thought it was
too difficult. I went into economics because I had been trained in mathematics and I thought, as Russell did, that economics looked easy. It took me several
years to get from Russell’s position to Planck’s. Economics is inherently dif-
ficult. In this chapter I will explain one path by which I came to that view.
* Editor’s note: this is adapted from the conference keynote address upon which this volume is based. I asked Brian to keep the informal style as part of the chapter.
Whether one sees economics as inherently difficult or as simple depends on how one formulates economic problems. If one sets up a problem and
assumes rationality of decision making, a well-defined solution normally fol-
lows. Economics here is simple: from the problem follows the solution. But
how agents get from problem to solution is a black box; and whether indeed
agents can arrive at the solution cannot be guaranteed unless we look into this box. If we open this box economics suddenly becomes difficult.
Once in a while as economists, we do justify our assumed connection
between problem and solution. In a well-known paper, Rust (1987) tells the
story of Harold Zurcher, the superintendent of maintenance at the Madison
(Wisconsin) Metropolitan Bus Company. For 20 years Zurcher scheduled bus
engine replacement of a large fleet of buses—a complicated problem that
required him to balance two conflicting objectives: minimizing maintenance
costs versus minimizing unexpected engine failures. Rust figured out the
solution to this combinatorial optimization problem by stochastic dynamic
programming, and matched that optimization against Zurcher’s. He found
a reasonably close fit. The point of Rust’s article was that although this was an enormously complicated problem, Harold Zurcher found the solution and
therefore, at least in this case, the economists’ assumption that individuals
find optimal solutions to complex questions is not a bad assumption.
The Zurcher example leaves us with a broad question: Can the assumption
that individuals find optimal solutions to economic problems be justified so
that we can avoid studying the details of the decision process? In simple cases the answer is yes. In most cases, however, it is no. Think of an ocean that contains all the well-defined problems that interest us in the economy, with ever more difficult problems at greater depths. Near the surface lie problems like
tic-tac-toe. Below that are problems at the level of checkers, and deeper still are problems like chess and Go. We might know theoretically that a solution
to chess exists, say in mixed Nash strategy form, but we can’t guarantee that
human agents would arrive at it. So the problems that are solvable the way
tic-tac-toe is solvable lie within two or three inches of the surface, but at levels deeper than this, problems cannot be guaranteed a solution. We can add to
these the many problems agents face, perhaps the majority they face, that are
not well specified. Zurcher’s problem lies on the boundary of what economic agents can accomplish by way of a “rational” solution. Deeper than this, economic “solutions” may not match “rationality” or may not exist.
What happens at these deeper levels? Human decision-makers do not back
off from a problem because it is difficult or unspecified. We might say that
when problems are too complicated to afford solutions, or when they are not
well-specified, agents face not a problem but a situation. They must deal with that situation; they must frame the problem, and that framing in many ways
is the most important part of the decision process. To consider that framing
you have to consider what lies between the problem and the action taken.
cogni t ion [ 159 ]
And between the problem and the action lies cognition. Between the problem and the solution there’s a lot going on, and if one considers what is going on, economics becomes difficult. To paraphrase my question then: How do people
problems? How do we really cognize?
In this chapter I want to consider cognition as a cognitive psycholo-
gist might look at it, and apply the findings to thinking about two different
issues: economic modeling and the education of graduate students.
NOTIONS OF THE MIND
In economics we have a simple and old notion of mind. Mind is a container
that holds data. The data are constantly updated by interaction with the world; and mind performs deductions based upon these data. All of this of course is
implicit; in economics we don’t talk about “mind.” But we do view mind—or
that which gives rise to ratiocination—as deduction upon collections of data
sets. In economic theory this is reflected in treating beliefs about the world as expectations of variables conditioned upon current data (or sigma fields)—
current information—and in formulating solutions based upon these. This is
a shorthand, the sort of reasonable abstraction that any science makes that
works well in many cases. But we need to get beyond it when we go deeper
than two or three inches into the ocean of problems.
Let me look at mind and the cognitive process then from a deeper view-
point—that of cognitive science. Imagine that at night you are reading a
novel, say Haldór Laxness’s Independent People and you’re enjoying it. What is actually going on? Actually, that’s complicated. The black and white marks on
the page are focused onto the light sensors or pixels at the back of your ret-
ina. These sensory perceptions are transmitted to the rear part of your brain, and map into certain visual structures there. Somehow letters and words are
parsed out, and somehow these fit together via an understanding of syntax.
(Where I say “somehow” I mean that cognitive scientists do not know the exact
mechanism of what is happening.) From syntax somehow “meaning” emerges.
But what is meaning? Meaning in this case is a set of associations. You might
read a sentence about rain: “Smoothly, smoothly it fell, over the whole shire, over the fallen marsh grass, over the troubled lake, the iron-gray gravel flats, the somber mountain above the croft, smudging out every prospect.” These
words trigger associations—associated memories really—and you form a pic-
ture, or a set of pictures. These associated memories and pictures in turn trigger what you might call “affect,” or feelings. The feelings are often subtle, the kind of feelings of what it might be like to be in Laxness’s world—the gloom
of the rain, the dreariness of the gravel flats, the oppressiveness of the mountain, the smell of the croft in the dampness. These are subtle feelings, and these
[ 160 ] Complexity and the Economy
feelings actually are our intelligence, are part of our cognition. They’re part of the meaning that we give to symbols. Reading and making sense of what is read
consist of associated memories and associated feelings. How all this happens is not well understood by cognitive scientists; it’s what French thinker Henri-Jean Martin calls a mysterious alchemy.
Here’s how the Princeton cognitive psychologist Julian Jaynes (1976, p. 1)
expresses this alchemy of mind:
O, what a world of unseen visions and heard silences, this insubstantial country of the mind! What ineffable essences, these touchless rememberings and unshowable reveries! And the privacy of it all! A secret theater of speechless monologue and prevenient counsel, an invisible mansion of all moods, musings, and mysteries, an infinite resort of disappointments and discoveries. A whole kingdom where each of us reigns reclusively alone, questioning what we will, commanding what we can.
A hidden hermitage where we may study out the troubled book of what we have
done and yet may do. An introcosm that is more myself than anything I can find in a mirror. This consciousness that is myself of selves, that is everything, yet nothing at all—what is it? And where did it come from? And why?
The point I want to make here is the meaning that’s abstracted from the
book is not in the book; it is in the mind. It’s a point that starts to get recognized in philosophy in the 1700s by Kant, but isn’t fully articulated until the twentieth century. We construct meaning by the associations we make. If this
seems strange, imagine a page in Russian of Dostoyevsky shown to a Russian
reader and a non-Russian reader. Each gets exactly the same data, but the
Russian has the associations to parse the Cyrillic script and make the written sense data come alive. The non-Russian sees exactly the same data; but his
associations if he does not speak the language are nil and there is no meaning.
Meaning therefore is imposed. It emerges by our imposing associations. It’s
not Dostoyevsky or the book Independent People that brings meaning to me—
that’s an illusion. It’s me that brings meaning to Independent People. I’ m making sense, I’ m imposing associations, I impose meaning on what I’m seeing.
Not just any old meaning, but the meaning that emerges from the associa-
tions the book makes with my neural memory.
Let me give you another example because I want to hammer on this point
and derive a few things from it. There’s a Yeats poem that goes something like this: “Down by the salley gardens my love and I did meet; she passed the salley gardens on little snow-white feet. She bid me to take life easy, like the grass grows on the weirs, but I was young and foolish, and now am full of tears.”1
1. Editor’s note: The lines are lines 1, 2, 7, and 8 of an eight-line poem, ‘Down by the Salley Gardens’, by W. B. Yeats. From W. B. Yeats, the Poems, ed. Richard J. Finneran, New York; Macmillan, 1983 (p. 20). Down by the salley gardens my love and I did meet; cogni t ion [ 161 ]
These words will have different effects on different people—different meanings. Ask yourself what meaning you get out of weirs. For me this has enor-
mous meaning because I and my friends played near weirs as children. (Weirs
are little dams in a stream, usually covered with algae and some form of green trailing grass.) I also know what salley gardens are. But those who are not
Irish will probably be affected differently. They may wonder: what are salley
gardens anyway? Maybe Salley had a garden. Maybe there’s such a thing as the
Salley Gardens—maybe they exist on some estate near Dublin. In the absence
of knowing what salley gardens are, you probably have an image of a garden
well kept, surrounded by flowers and tended by keepers. But it’s not that. The word in Gaelic is s-a-i-l-e-a-c-h, and it means “willow.” So Yeats is near willows, and therefore likely near water. If there’s a weir, the water is a stream or river.
Once one has these associations, immediately the initial picture shifts. My
point is that different meanings can be imposed on the same data. Different
meanings that come from different associations.
Data—literary or economic—have no inherent meaning. They acquire
meaning by our bringing meaning to them. And different people, with differ-
ent experiences, will construct different meanings.
THE MIND AS A FAST PATTERN COMPLETER
What conclusions does modern cognitive psychology draw from such exam-
ples? The first conclusion is that our brains are “associative engines” to use a phrase of Andy Clark, a philosopher and cognitive scientist from Washington
University in St. Louis (Clark, 1993). We’re wonderful at association, and in
fact, in cognition, association is just about all we do. In association we impose intelligible patterns. To use another of Clark’s labels,
we are fast pattern completers. If I see a tail going around a corner, and it’s a black swishy tail, I say,
“There’s a cat!” But it could be a small boy with a tail on the end of a stick who’s trying to fool me. But I don’t do that. My mind is not built to do that.
If I were strongly skeptical, I could do that, or if I saw some small boy playing pranks I could say, “Well, it’s either a cat or a small boy.” But in the absence of a small boy, all I’m really saying is, “Hey! I see a cat.” But I didn’t see a cat. I saw a black tail. A famous Bertrand Russell story makes the same point.
A schoolboy, a parson and a mathematician are crossing from England into
Scotland in a train. The schoolboy looks out and sees a black sheep and says,
/She passed the salley gardens on little snow-white feet. / She bid me take love easy, as the leaves grow on the tree, / But I, being young and foolish, with her would not agree.
/ In a field by the river my love and I did stand, / As on my leaning shoulder she laid her snow-white hand. / She bid me take life easy, as the grass grows on the weirs; / But I was young and foolish, and now am full of tears.’
[ 162 ] Complexity and the Economy
“Oh! Look! Sheep in Scotland are black!” The parson, who is learned, says, “No.
Strictly speaking, all we can say is there is one sheep in Scotland that is black.”
The mathematician says, “No, still not correct. All we can really say is that we know that in Scotland there exists at least one sheep, at least one side of which is black.”
Cognitive science repeatedly tells us that we don’t think deductively as
the mathematician did, we think associatively as the schoolboy did. And for a
very good reason: evolution has made it so. Our ability as humans a hundred
thousand years ago to sniff the air and associate a fleeting humidity with the presence of water a few miles away had real survival value. Completing patterns fast, surmising the presence of water from the faintest of clues, helped us survive. Deductive logic did not; and in all but the most trivial of cases
we do not use it at all. In fact, cognitive psychologists tell us that deductions themselves are primarily associative. I may say I can solve such-and-such a