We Have Met the Other and We’re All Nonlinear: as a Nonlinear Dynamic System

Ethnography as a Nonlinear Dynamic System

“ thnography” and “” have always had an awkward relationship. As far as research practice goes, ethnographers often feel that they have more in E common with investigative report- ers, historians, and intelligence analysts than they do with colleagues in For an ethnographer, what’s and economics. The usual interesting is the discovery of chant of “theory/hypothesis/measure- connections. MICHAEL AGAR ment/sampling design/significance level” does not describe much of what ethnographers do. For an ethnographer, what’s interesting is the discovery of connections. A holistic Michael Agar is with the Friends Social perspective is foundational for anthropology, sort of a proto-, which Research Center and Ethknoworks, helps explain why two anthropologists, Gregory Bateson and Margaret Mead, par- 1040 Park Avenue, Suite 103, ticipated in the post-World War II conference that founded . Baltimore, MD 21201. For additional It’s not that ordinary social science isn’t allowed to look for connections. It’s that writing on ethnography and ordinary social science begins with variables already given by some theory, and then , see forthcoming articles in tries to figure out how to locate, decontextualize, and measure those variables. A Human Organization and Journal of card-carrying holist notices a “variable” in a situation, maybe one that he/she had Artificial and Social never thought about before, but then he/she wonders what other things it might be Stimulation. connected with, in that situation and outside of it. The goal is to build patterns of many interacting things that include what was noticed, not to isolate what one was supposed to notice and measure it. It’s a problem that many students from traditional social science have discovered for themselves. I wish I had a nickel for every time this happened in my teaching days: A student from outside of anthropology walks in, looks around nervously, and says in a

Correspondence to: Michael Agar, 1040 Park Ave. #103, Baltimore, MD 21201, E-mail: [email protected]

16 COMPLEXITY © 2004 Wiley Periodicals, Inc., Vol. 10, No. 2 DOI 10.1002/cplx.20054 quiet voice. “I’m working on my disserta- massive overdetermination of pattern [3], tion and I’ve got a problem. I’m supposed concepts that I’ll draw from in mercifully “,” to take another example, to do a linear regression to explain this brief fashion later in this article. For now, is a core concept, the observation that dependent variable. But…” I’d like to see if I can carry the banner a a lot of interacting elements can self- And with that the student pulls out a little further by looking at a system of organize into surprising system results. piece of paper with stains made of equal ideas that ethnographers are now begin- parts of tears and pizza grease. A diagram ning to consider. shows a large number of tiny boxes. The During the 1980s and 1990s a new featured NSF and NIH presentations. boxes are filled with all manner of things research framework grew at an increas- No, this was claimed to be the real that the student knows are relevant to ing rate under the popular—and often thing, a shift in the epistemological tec- what he or she is trying to understand. misinterpreted—monikers of “chaos” tonic plates by which the powers that Arrows fill the spaces between the boxes, [4] and “complexity” [5]. The growth be define what “science” in fact is. What’s especially interesting about all It looks like spiders of different sizes was driven by a problem shared by re- this, for an ethnographer, is how familiar pressed flat on the paper. “Can you help searchers of all kinds—how are we to this new epistemology sounds. Connec- me with this?” asks the student. understand systems with so many inter- tions, the missing piece from traditional “No,” I say, or learned to say with acting elements that it’s hard to say science, in fact turn into the point of it all. experience. I tell them to go back to what they will look like in the future? “Emergence,” to take another example, is their department and never talk to any- This growth was enabled by advances in a core concept, the observation that a lot one like me again until their disserta- computer technology, because simula- of interacting elements can self-organize tion is signed. They will just make a tion is the only way to model and ex- into surprising system results. As I enjoy mess of their exams and annoy their plore such systems. We rapidly ap- saying to my colleagues in this new field, faculty, and someone like me as one proach the time when you can pick up “anthropology said emergence before outsider among five examiners won’t be enough computer power to run such emergence was cool.” The notion that able to prevent the slaughter. Writing a models at the convenience store. most things one is trying to figure out dissertation is as much, if not more, a This comparatively new framework— aren’t just a matter of a few variables political act as it is a scholarly act. Ask new for science, not for Buddhists—in- causing another to act in a certain way— Foucault. But their goal—the compli- trigues people like me and, presumably, it’s not a surprise if you do ethnography. cated picture of pattern, the many links readers of this journal. Gaddis [6] recently Let me begin referring to this “new among many different things—is ex- explored the framework for history, not- kind of science” with the standard actly what an ethnographer aspires to. ing that social science has always trou- phrase “complex adaptive systems,” ab- Understanding how the social world bled him because he can’t find any of the breviated “CAS” for ease of writing. I works is poorly served by traditional so- “independent” variables they all seem to like this phrase rather than the many cial research approaches. Truth-in- insist on. If the venerable Annual Reviews other descriptive terms for the field be- packaging compels me to say that coun- are any indicator, then the framework has cause it summarizes the basics, with terexamples to this argument exist and arrived with recent review articles for one caveat. In social research, “adap- should be consulted, like Russ Ber- both sociology [7] and anthropology [8]. tive” has problematic connotations. A nard’s overview of anthropological re- And from within the framework itself, blind commitment to adaptive may well search with just this mainstream social studies of “artificial societies” [9] and re- miss the point of what is going on, and science framework in the foreground ports on them in the Journal of Artificial many of the social systems I deal with [1]. And toward the end of my own Societies and Social already are better described as “maladaptive.” methodology book I show the relevance have a pedigree of several years’ duration. Such are the dangers of cross-disciplin- of traditional methods when built on In a landmark event that placed this ary metaphors. In more precise terms, ethnographic findings in a “grounded framework squarely in the middle of adaptive just means that both sides of theory” sort of way [2]. public , Wolfram published a the structure/ opposition must But still, theory-driven linear causal book called A New Kind of Science [10]. be taken into account in analysis and models that equate prediction and expla- This event raised the stakes, because explanation. On this both ethnogra- nation just aren’t the flags most of us fly now we (the public) were talking about phers and CAS can agree. under. That leaves the problem of design- a paradigm shift in the classical Kuh- Many are the ways to think about the ing some new flags. Along with many nian sense—not just a little conceptual relationship between ethnography and other colleagues, I’ve tried to do this over tinkering. I say “public” as reflected in CAS. For awhile now I’ve been arguing the years. My flags went by names such such places as mainstream book re- that ethnography, in this day and age, as rich points and abductive logic and views, congressional testimony, and actually labels a kind of logic rather

© 2004 Wiley Periodicals, Inc. COMPLEXITY 17 than a unit of study. Waldrop [5], in his plex” a system is. The one that makes overview of complexity, offers a tanta- most sense here is “algorithmic com- Traditional social research is lower on the lizing description of CAS as a kind of plexity.” An algorithm is just a set of algorithmic complexity scale compared “qualitative holistic math.” There are procedures for doing something. So, with ethnography. interesting epistemological parallels to one measures algorithmic complexity explore here. In fact, if you take CAS by answering a question: “Is there an seriously and want to do social re- algorithm to produce the expression of how different it is when compared to search, ethnographic logic is where you interest that is simpler than the expres- traditional social research. An ethnog- have to go. One sketch of that logic will sion itself? How much simpler is it?” raphy will always be higher in algorith- be included later in this article. One informal example is the contrast mic complexity. When you study meth- A second way to explore the relation- between paint-by-numbers and a Jack- odology in traditional social science, ships between ethnography and CAS son Pollock painting. Paint-by-numbers what do you learn? You learn some al- lies in the question of just what it is we is simple by comparison. Pollock in an gorithms. The idea is, you can imple- produce at the end of a study. The old interview was asked what he did about ment those algorithms to do research in “domain plus detail” model of tradi- “mistakes” and he just laughed. You any number of areas. tional ethnography was shredded long can’t have mistakes if you’re not follow- So you learn how to design and val- ago. No more “Chapter III: Kinship: Part ing an algorithm. idate instruments, for instance. I: Mother’s Brother.” Modern ethnogra- At the ridiculously simple end of the You take courses that nowa- phy and CAS agree that what they’re ethnographic scale, suppose you devel- days are concepts plus computer pack- after are ways to describe systems that oped an algorithm that went like this: ages. You take courses that teach you mix order and disorder, systems that Say “hello how are you” to everyone the difference between a true random move and change, sometimes in small within a one-mile radius of where you and a stratified sampling design. You ways that react to circumstance, some- stand when the algorithm begins. learn these algorithms so that you can times in major ways that change the Record the answers. When no one is left execute them in sequence to do a full nature of what it means to be a partic- you haven’t said hello to, stop. Publish a study. It’s not very nonlinear or dy- ipant. There are interesting representa- list of the answers in the American An- namic, but then it’s not supposed to be. tional issues here, and a few will be thropologist. This looks like a Monty Py- It’s supposed to be linear and predict- described at the end of this article. thon version of Newton and Locke. Al- able and, above all, controlled. Take a A third way to explore the relation- gorithmic complexity is extremely low. research problem, plug in the algorith- ship—the way I will emphasize here— But what the algorithm produces is cer- mic modules in the right sequence, and involves how the research process itself tainly not any kind of ethnography. bingo, it’s a study. Doesn’t matter who mirrors the epistemology and the rep- Now consider something from the you are or when or where you do it, as resentation. In other words, ethno- ridiculously complex end of the scale. long as you follow the algorithms. graphic research is, in and of itself, a Well, I can’t write it, because the com- Traditional social research is lower . The process plex end means that the algorithm on the algorithmic complexity scale involves an ethnographer, at least one, would be as long as the study. In other compared with ethnography. An eth- and different people that he/she spends words, every moment of the study nographer of course also learns some time with, and in this day and age lots of would call for procedures unlike any- algorithms in graduate school. And information from other sources as well. thing that had been done before. It some, like the algorithms learned by The process begins in comparative dis- would be a social constructionist fan- more traditionally trained social scien- order, shifts and changes through time, tasy, every moment built up in a unique tists, will have to do with things like and typically winds up with conclusions way. No pattern, no repetition, no algo- asking questions, and figuring out that were not expected at the beginning. rithm. The description of a study would whom to talk with next, and comparing Is this an interesting way to think be as long as the study itself. Algorith- different kinds of data by way of analy- about ethnography? I’d like to explore the mic complexity would be extremely sis. The ethnographic versions will be question here by looking at a few key high. This is not any kind of ethnogra- much more varied and loose, though. concepts from CAS to see if they clarify phy, either. Maybe Hunter Thompson And our ethnographer-to-be will issues for ethnography and suggest new aims in this direction, at least in the old also learn a meta-lesson. (“Anything ways of thinking about those issues. Let’s days as he swallowed a handful of pills you can do I can do meta,” as a col- begin with the question of whether or not before describing a scene, but not eth- league is fond of saying). The meta-les- ethnography is an CAS at all. nography. son says, learn as many algorithms as The algorithmic complexity of an you can, but understand that you won’t ETHNOGRAPHIC COMPLEXITY ethnographic study will be between know which ones will apply, at what On the face of it ethnography is obvi- these two extremes. But the point here point in the study, in what kind of com- ously a CAS. Several measures have isn’t to measure ethnographic complex- bination, until you’re actually doing the been proposed to diagnose how “com- ity in a precise way. The point is to show study itself.

18 COMPLEXITY © 2004 Wiley Periodicals, Inc. As if that weren’t bad enough, the that doesn’t have much evolutionary cesses, the problem being you never student will absorb a second meta-les- potential. One of the strengths of com- know at the beginning which way it will son. That lesson teaches that issues plex adaptive systems is that new cir- turn out in the long run. considered “methodological” will come cumstances can result in reorganization Sensitivity to initial conditions is a up during the study. They will lead you to better respond to those changes. “signature,” as they say, of a CAS. What to ways of learning and documenting Methods “evolve” as local information does this have to do with ethnography? that you had no idea existed when you about how to do a study accumulates. A lot. Consider the problem of eth- first started the study. You will learn Ethnography does this. Traditional re- nography by different people on the how to ask the right question of the search prohibits it. “same” community. The problem is right people in the right way using So if we think of ethnography as a part of the early history of American knowledge you didn’t know existed. CAS, we get some understanding of why anthropology. The famous Redfield- You will see that certain kinds of data we look so different from other social Lewis debate, for example, centered on belong together in ways that you would sciences. And it turns out the differ- whether villagers in a Mexican pueblo never have imagined until you’d ences are an advantage, assuming that a lived in a fundamentally harmonious or worked on the study for awhile. researcher wants to improve methodol- a conflict-ridden world. In fact, many of No ethnography will lie at the max- ogy with experience in the course of a these debates—Benedict-Bennett and imum algorithmic complexity end of study. With ethnography, he/she can Mead-Freeman are other examples— the scale. It would actually be funny to maintain flexibility and creativity to pivot in large part on just this difference write a parody that represented such a adapt method to unforeseen circum- (For a review of the classic restudies study, something like an extreme Being stances, and he/she can reorganize issues, see Ref. 11). Was the village a There, the novel by Jerzy Koszinski. But methods to adapt to research problems place of self-interest and competition or any ethnography will be more complex, the likes of which were unknown when a place of altruism and cooperation? algorithmically speaking, than any the proposal was written or, for that Any “village” is of course both. And it is study done with traditional social sci- matter, until well after the study was interesting that just this issue—compe- ence methods. already underway. tition vs. cooperation—is a centerpiece There are a couple of conclusions to Did we know all that already? Sure. of recent CAS research as well, but foreground here. The first is, algorith- Did we know how to say that already? that’s another story [12]. mic complexity helps understand why Not very well. Did we have a transdisci- The anthropological debate took more the concept of “methodology” was and plinary framework as a foundation for political shape after the 1960s. In this still is so slow to develop in anthropology. the knowing and saying? No. ongoing dispute, the question takes a “Methodology,” as understood in social couple of different forms. One is, can an science in particular, science in general, SENSITIVITY TO INITIAL CONDITIONS ethnographer who is not a Y ever do an is all about neatly compressed algo- So say ethnography is a CAS, as indicated adequate job of studying the Y, or can rithms that work to generate data and by its algorithmic complexity. Let me now onlyaYdosuch research? A second form analysis, whatever the problem, what- move to one consequence, the famous is, given of the A’s that ever the context. But the idea of a “sensitivity to initial conditions.” In pop- have always and only been done by B’s, “methodology” section—algorithms ular discourse this is called the “butterfly what might the B’s have missed or mis- that generate a study—is anathema to effect,” meaning that a butterfly flaps its understood that would show up if eth- ethnographers. They know that a study wings in one part of the world and even- nographies of the A’s were done by C’s? always develops, methodologically tually the disturbance becomes a major There are enough writings and de- speaking, in ways unforeseen at the be- storm in some distant location. The bates to fill an encyclopedia. One con- ginning. In fact, “methodology,” many phrase “iron butterfly” comes to mind. cern centers on identity issues: feminist, would argue, is something best told in The example is a little melodramatic, but minority, gay/lesbian. Another centers the course of telling the study, not as a it just means to show that a little differ- on ethnographers from locations that separate section. The story of method- ence in initial conditions, when you turn have been traditional destinations for ology is the story of the study. This is a on a CAS, can make a big difference after European and American ethnogra- reasonable argument given a relatively it runs for awhile. phers, the so-called “indigeneous eth- high degree of algorithmic complexity. We need other dramatic phrases, like nography.” The question here is, “how A second conclusion: Traditional so- a “bureaucrat effect,” meaning no mat- have we been (mis)represented by eth- cial science methodology lies low on the ter how much energy you put into nographers not of our identity, and how algorithmic complexity end of the scale. changing initial conditions you get the might we see things in a particular eth- Things like nonlinearity and researcher same result. And we need a “goes nography, of them or anyone else, that influence and path-dependence are around, comes around” effect, meaning people not of our identity would miss?” problems to be eliminated, not features that you change initial conditions and However the problem is phrased— to be celebrated. The problem here is get a result that is pretty much what you science and/or politics—ethnographers that we’ve designed a research animal deserve. All these are possible CAS pro- have a long tradition of wondering why

© 2004 Wiley Periodicals, Inc. COMPLEXITY 19 two different studies should come up system that results from all those inter- viewee interrupted him. The interviewee with different results. If we take seri- actions over a period of time, the sys- said, “You know, this is all very interest- ously the idea of ethnography as a CAS, tem will produce some characteristics ing, but can we talk about me for awhile?” this “problem” should not be a surprise. that were not foreseen when the model CAS offers an ironic combination of In fact, it’s not a “problem” in the sense started to run—the unexpected emer- the poststructural and the scientific, a that something has gone wrong that can gent properties. framework that accepts the heresy of be repaired. Instead, it’s a normal part Once again, this CAS framework will researcher influence but then deals of a CAS—sensitivity to initial condi- sound familiar to ethnographers. You, with it in a systematic fashion. Some of tions. The idea that a study is a process the ethnographer, aren’t just watching the consequences of this “distributed that can have only one outcome is old the ethnography run. You are in the authority,” as the computational types science. The idea that a study can take ethnography as well, part of the ma- call it, haven’t been well worked out yet different shapes depending on initial chinery that makes it run. Hence the for debates about “authority” in anthro- conditions is CAS. Two ethnographies, oxymoron “participant observation.” pology. But placing the ethnographic considered as CASs, may well produce You are one, but only one, of the agents. “agent” in the story does recognize that different results, even if applied to the As just described in the previous sec- authority. On the other hand, because “same” pheonomenon. tion, you are part of the initial condi- the ethnographer is only one among What rescues us from epistemologi- tions to which the CAS will be sensitive. many, he/she is less significant overall cal anarchy is this: The fact that more But whatever the story of the ethnogra- than he/she probably thinks. In fact, than one ethnography of the X is possi- phy will turn out to be, you are also an during most of the ethnographic story, ble doesn’t mean that any imaginable active agent in its construction. he/she probably has little if any author- ethnography of the X is acceptable. In For years social science has obsessed ity at all. He/she certainly isn’t respon- fact, before I began my nonlinear read- about the “Hawthorne effect,” named af- sible, alone, or even mostly, for emer- ings, it was clear that a methodological ter the research project conducted for gent properties that result. problem for ethnography was figuring Westinghouse by the Harvard Business CAS brings agency and distributed out the edge between credible and in- School. When the researchers made authority and reflexivity into the equa- credible, between studies (note the plu- changes in a wiring room, productivity tion, three issues that have driven eth- ral) that made a convincing case and went up. Funny thing was, it didn’t mat- nographic debate since the 1980s. An other studies that shouldn’t be believed. ter what changes were made, even ethnographer is part of the study, and To put it in CAS terms: For any given changes that undid a previous change. he/she and “the others” collectively cre- community, network, location, event, The reason productivity went up was be- ate the story that will later be told as an or whatever focus an ethnography cause morale went up due to the atten- ethnography. Different ethnographers might take, what does the “attractor tion given to the workers, mostly immi- will produce ethnographies that will dif- space” of possible ethnographic results grant women. The method produced the fer from each other. That is to be cele- look like? If we could know that space, results. Never mind that Heisenberg was brated because they can be compared we could say in the end that given the figuring this out for at roughly the to obtain a larger perspective than just constraints of an ethnographic research same time. The Hawthorne effect was an one alone can provide. The fact that a process, and given the different points evil to be avoided in social research. Czech feminist critical theorist and a of view that can be represented at that Ethnographers think—at least this Nigerian masculinist methodological historical moment on the part of an eth- one does—that if you believe you’ve individualist differ in their ethnography nographer, here’s the space that defines eliminated the Hawthorne effect, you of the White House does not mean that the acceptable ethnographies that have probably smoked too much for any ethnography of the White House is might emerge. Not much of a clear so- breakfast. An ethnographer has to ac- credible. But more than one is possible, lution, I’ll admit, but an interesting way cept that he or she is part of the data, or, and it is likely that those two ethnogra- to better understand the problem, given to return to the CAS version, that he or phers would prove the point in ex- that we think of ethnography as an CAS. she is an agent in the simulation. Tell- tremely interesting ways. ing a story that you were part of makes By the rules of old science, the Haw- THE ETHNOGRAPHIC AGENT more sense than telling a story and pre- thorne effect is an embarrassing problem. Another way that CAS and ethnography tending you weren’t there. By the rules of CAS, it is an aspect of an link: Consider the thought experiment This can go to extremes. In one no ethnography to be expected. It is a feature laboratory that CAS has spawned, doubt apocryphal account, a former pos- of a research process to be celebrated for agent-based modeling [13]. Such models itivist on a panel at an anthropology its realism, flexibility, and adaptability, put a number of autonomous agents meeting said he became a born-again not something to try and conceal. into a simple world. The agents know poststructuralist. He decided that he’d some things, know how to do some lean into the study and make his pres- FRACTALS things, and they and the world change ence known. After about an hour of Let’s take a look at another age-old eth- as a result. When you take a look at the interviewing in this new mode, his inter- nographic process. I’ve used a frame-

20 COMPLEXITY © 2004 Wiley Periodicals, Inc. work since the 1990s [2], one that de- the development of the circulatory sys- levels of scale. It can appear in a conversa- scribes a core process of ethnographic tem or the growth of trees or the forma- tional moment, a family dispute, the me- research. The process cycles as long as tion of mountains and coastlines. They dia, public policy, or even be used as a label the study continues. It goes, in outline, have been applied in economics, as in to characterize the entire city of Vienna. like this: Ethnography can be described the way stock prices squiggle in similar Once I worked out an outsider’s under- as translation writ large. It bridges the ways at different time scales. They have standing of “Schmaeh,” many pieces of the gap in understanding between the been applied to produce those interest- Austrian puzzle fell into place. world of an audience that the ethnog- ing visual images that now adorn the Call it a “theme,” call it a “cultural rapher means to address, including sides of coffee mugs. And they form the pattern,” call it what you will. It acts like a him/herself, and the world that is being basis of Wolfram’s book, cited earlier, fractal, in the sense of being an algorithm researched. Ethnography, in other where simple algorithms, iteratively and that applies iteratively and recursively to words, is a construction that shows how recursively applied, generate patterns of create patterns at different levels. social action in the context of one world amazing complexity, the basis of his But are these fractals—methodolog- can be understood as coherent from the “new kind of science.” ical as discussed earlier and thematic as point of view of another. Hold in your mind, for a moment, the just described—are they really like the Much of what an ethnographer does is idea that ethnographic research rests on a neat math that explains trees and blood focus on the differences that he/she no- fractal-generating process. One rich point veins and the stripes on Waldo the cat? tices, what I have called “rich points.” leads to another, more detailed, rich They don’t just reproduce identical pat- Rich points are moments when some- point. Things are not so neat as in the terns. But then that is common in na- thing doesn’t make sense, ranging from a versions, but I want to ture as well. As the old saying goes, no complete surprise to a departure from the describe a second resemblance to fractals two snowflakes are alike, yet the algo- expectations that the ethnographer before dealing with that issue. rithms that produce them are the same, brought with him/her. Such rich points The second resemblance is actually a the variations being a product of con- signal differences between the two worlds core concept of American cultural an- tingencies and environment. “Context” and, as such, constitute the basic prob- thropology. One clear example of this we would call it. lem of an ethnographic study. core concept lies in the title of Ruth But is the rich point cycle or the In pursuit of this problem in under- Benedict’s classic book, Patterns of Cul- Schmaeh-like attitude really an algo- standing, an ethnographer engages in a ture [14]. Another example, one that I rithm? Neither is a “finite procedure, cyclical process, modifying and trying used long ago, is Morris Opler’s idea of written in a fixed symbolic vocabulary, out frameworks that initially didn’t “themes” [15]. Numerous others exist. governed by precise instructions, mov- work until they work so well across so What is this idea about? This is about ing in discrete steps 1, 2, 3,…, whose many kinds of data that they become differences between worlds that are so execution requires no insight, clever- candidates for an ethnographic conclu- fundamental that they appear at many ness, intuitions, intelligence, or perspi- sion. The processes of modification and different levels of scale and across many cuity, and that sooner or later comes to validation are too complicated to detail different domains of experience. An ex- an end” [17]. Clearly they don’t fit this here, but the critical point for now is ample that I’ve used elsewhere [16]: Liv- definition, though there is a family re- that they typically require several cy- ing and working in Austria over the cles. They are iterative. years, I became fascinated with the con- semblance here. A better social inter- And while in the middle of working cept of Schmaeh. I heard it and read it pretation of algorithm lies in Bourdieu’s on one rich point, another often comes all the time and native speakers gave a “dispositions,” or perhaps some sche- up. What is an ethnographer to do? He/ bewildering variety of definitions, some ma-theoretic constructions like the de- she now applies the same process in saying it couldn’t be defined at all. fault value, but I think those attempts which he/she is currently engaged, only Here is a gloss on what I eventually just hide the problem under a new set this time to a rich point that appeared came up with: The term encodes a gen- of ambiguities. as the process was ongoing. In other eral principle of irony plus humor. How are rich points and Schmaeh words, the cycle isn’t only repeated over Nothing is what it seems, what it actu- like algorithms then? They are certainly and over again. It is also applied within ally is, is worse, and you might as well finite and symbolic, but they also re- itself. The process is also recursive. joke about it rather than getting upset. quire rather than exclude “insight, clev- This description bears an astonish- There’s a joke I heard all the time, in erness, intuitions, intelligence and per- ing resemblance to the idea of fractals various forms, with a particular punch- spicuity.” They can be learned by in CAS (see Ref. 4. for the classic intro- line: The German says, “The situation is modeling them as, “instructions,” but duction). In its pure mathematical serious but not hopeless.” The Austrian those instructions will seldom be pre- form, a fractal is a simple algorithm ap- replies, “No, the situation is hopeless, cise or move in discrete steps. And in plied over and over again at different but not serious.” what I jokingly call “the ethnography levels of scale. Fractals have been used What I’ll call a “Schmaeh attitude” ap- halting problem,” it’s not clear that they to describe natural formations, such as pears across many domains and at many ever end.

© 2004 Wiley Periodicals, Inc. COMPLEXITY 21 Perhaps it would help to think of an Now the space is wide open. Ethno- A representation must certainly be area rather than a path. The computa- graphic results are presented in ways intersubjective. It is intersubjective be- tional notion of algorithm, at least the other than texts—museums, films, tween ethnographer and “others,” the one cited above, is like a path with a clear plays, computer programs, executive people whose point of view an ethnog- beginning and end, a narrow strip cut summaries, program designs—virtually rapher is exploring. It is also intersub- through thick woods. The ethnographic any way that study results can be “re- jective between an ethnographer and notion of algorithm is more like a cleared presented” in another format. And con- an audience of the representation. ski area, with a large mountain face sider the ways that a representation Whatever it looks like internally, the bounded by limits, but with plenty of might be constructed just in the textual representation must link, at some point, room for a skier to carve an infinite num- realm—anything from a formal mathe- with available representations for the ber of paths from peak to base. My read- matical analysis to a book of poetry. ethnographer, for the people studied, and ings on fuzzy logic [18] suggest that the As far as how to build a specific rep- for the consumers of the representation. notion of a “fuzzy algorithm” might help. resentation, then, I don’t think CAS can It has to make sense of things for all those I can’t resolve the issue here. But be of much help, because we live in a points of view individually, and it has to however it is eventually resolved, the historical moment when the possibili- enable common action for any two of method fractals and the results fractals ties are overwhelming. When so many them talking or working together as well. don’t stand in a one-to-one relation- experiments are underway, it is not at Let’s call this the triad constraint, triad as ship. The method fractal produces data all clear which of the new ethnographic in, the representation has to make sense in a fractal-like way, but it doesn’t nec- “species” will survive into the new era. of the C from three different points of essarily—or even usually—produce But framing things in terms of “sur- view A, B, and C itself. This constraint limits the possible fractal-like themes at the same time. As vival” does lead to an CAS type of ques- shapes a representation might take. In an ethnographer works on a rich point tion. We know that many new represen- fact, much ethnography fails the test, and then another rich point surfaces, tation “species” are out there now. because many ethnographers just write the second rich point may well belong Maybe this suggests thinking about rep- for a few others like themselves. People to a different thematic pattern com- resentations in another way. What kind about whom a book is written find it pared with the original rich point. No- of environment are they competing in? boring and irrelevant, and the public ticing a problem, then working on the To what conditions must these compet- can’t make much sense out of it either. problem, and noticing a new problem ing representations adapt if they are to Here’s another constraint a represen- embedded within it—this does not survive and flourish? tation has to live with: It has to have some guarantee that the two problems no- Here we enter the realm of another kind of a “zoom” function, like Mapquest ticed will also express the same theme. CAS concept, that of a fitness landscape.A or a video camera. The idea is that you The rich point cycle, in other words, fitness landscape is a picture of the world can focus on a particular rich point, and doesn’t also do the thematic analysis. It within which agents are dying, surviving, then either zoom in for more detail or produces what it is that needs to be or flourishing, as the case may be. It zoom out for broader context around it. analyzed by other means. shows what agent characteristics will The question here is, will the representa- have a higher survival chance than oth- tion look similar at different levels as we LANDSCAPES ers. Often the landscape is visualized as zoom in and out? Will those levels scale? What of all the talk about ethnographic hills and valleys, with the higher altitudes Recall the earlier discussion of fractals, representation over the last couple of representing better adaptation. Fitness both the substantive kind called “themes” decades? What, if anything, might CAS landscapes can vary—there might be one and the methodological kind produced help with here? high mountain, or they might be a lot of by rich points. Given these two built-in The representation debate has low ones scattered about. And of course scaling processes, we can say that eth- opened up a confusing space of possi- the landscape, like everything else in CAS, nography “wants” to find scaled patterns. bilities. In the old days, an “ethnogra- can change with time. It’s a unique strength, but also a strong phy” was a piece of writing, usually a For present purposes, we just need to bias that must be watched carefully. In book-length thing. It had certain char- understand this: “Fitness landscape” sug- the end, though, scaled representations acteristic sections, like the arrival scene gests we think about representations in are probably one of the most important to establish that the ethnographer had terms of what they accomplish in the results we offer. Let’s call this the scaling come a long way and actually was world. In other words, can we imagine constraint. “there.” It had certain topic require- why some representations would do bet- Here’s another constraint: Many ments—social organization, religion, ter than others, no matter how they were have written about narrative as a natu- etc. It reflected certain writing conven- built by an ethnographer? Any answer to ral representation for nonlinear dynam- tions—minimization of the ethnogra- this question is of course open to debate. ics given that ethnographers, and histo- pher’s presence through passive voice, But it occurs to me that a few character- rians, tell a story of contingencies and for example. istics are at least worth considering. connections and how they interact over

22 COMPLEXITY © 2004 Wiley Periodicals, Inc. time. But a representation will have to 1. The triad constraint: A representa- how well the representation flourishes in contain at least two narratives. One tells tion’s boundary must connect with the fitness landscape. the “backstory” of what is being shown boundaries in at least three different The constraints set out a sort of in the representation itself, the story of points of view—subject, researcher, “management by objective” research where the main features of a represen- and audience. goal. Ordinary social science is more the tation came from. A second tells the 2. The scaling constraint: The represen- old hierarchical “command and con- story of how the representation itself tation will allow for granularity and trol” model. was constructed, the story of the study, scale in a fractal-like way, at least in It’s not hard to imagine that a book telling and showing the results at the key areas. or a movie or a museum exhibit could same time. This double narrative will be 3. The double narrative constraint: It will be built from the same research project a part of a representation, but not nec- contain a double narrative with itself and that any of the three—in spite of essarily—not usually—all of it. Let’s call as the end point: one narrative telling their differences—could fit the bill. And this the double narrative constraint. the history of the differences that the it’s clear that a movie, to take one ex- Another constraint: A representation, representation focuses on, and the ample, could be structured around plot- by its nature, will be fixed and static, be- other telling the history of the con- linked close shots or it could be a post- struction of the representation itself. cause it is an end product presented in a modern collage loaded with tracking 4. The dynamic constraint: It will con- book or film or museum exhibit. But even shots and jump cuts. tain dynamic features, such as oscil- though it is fixed, it will contain attractors But notice that the constraints do limit lators, attractors, and “edge of and mixes of order and chaos, so in this the possible shapes the representation chaos” descriptions. sense it will also be dynamic. It will dis- can take, whatever it is one decides to 5. The pragmatic constraint: The repre- play the dynamics learned during the pe- build. Imagine the number of possible sentation will submit to test in prac- riod of research. So the representation representations before the fact, and the tice, in the sense that it will enable will have the odd feature of being a but- number is for all practical purposes infi- comprehension and action across terfly on the wall with its wings still flap- nite. Add a few constraints of the type triad differences. ping, together with the story of how it outlined here and the number drops dra- landed there and what it might do to get matically. Add a few more and you’re un-stuck and where it then might go. We CAS leads us to think in a different down to a few choices. Perhaps fitness can call this the dynamic constraint. way about a representation. Rather than constraints are a more powerful way to foreground how to build it, CAS empha- The representation must also link with tell “when an ethnography is done right” sizes conditions it has to meet in order for actual practice, the things that people say than any description of a methodological it to serve its purpose. Put in terms of the and do, because understanding that flow recipe. This is an odd conclusion for an fitness landscape metaphor, a represen- was the point of building it in the first author who has written dozens of meth- tation has to evolve and survive in a world place. The methodological test for an eth- odology pieces over the decades. that sets constraints on what will work. A nographer is, can he/she use the repre- There are many more issues around good representation will co-evolve—that sentation as a vehicle to get from con- the problem of an ethnographic repre- is, a good representation will in turn scious change to habitual new pattern? sentation and what if anything CAS modify the very landscape within which it Can he/she now explain exactly what ini- might help us think through. For now, it took shape. But in this article the focus is tially was the signature problem of eth- is enough to leave with the already odd on building, and you can’t play in the nography—two people communicating idea that a representation is best evalu- co-evolution game until you’ve got some- with each other with the result that what ated in terms of the fitness landscape to thing to play with. they said and did didn’t make sense to an This is an interesting twist on meth- which it must adapt and within which it outsider? And then can the ethnographer odology in general, and on my favorite must survive. And that a few major convey that transformation from rich question from the uninitiated in particu- landscape constraints specify limits on point to coherence to an audience, so lar—“How do you know if an ethnogra- what we produce. And then to conclude that an audience can take a rich point phy is any good?” Traditional social sci- that the major issue for ethnographic example, learn the representation, and ence requires an answer in terms of research might be the outcomes a rep- then see the rich point as coherent where sample size, instrument, statistical analy- resentation must achieve rather than they couldn’t before? And can someone sis, the usual benchmarks of the tradi- any cookbook for its construction. from the researched group look at the tional process. The CAS emphasis on fit- Clearly there are methodological issues representation and feel a shock or recog- ness landscapes argues otherwise. In fact here, including the possible use of tra- nition, perhaps an “aha,” because it at its most extreme, the argument would ditional social research methodology. makes the familiar explicit? Let’s call this be that it makes no difference what pro- But the fitness landscape notion subor- one a pragmatic constraint. cesses were followed to build the repre- dinates the methodological issues to So far we have several constraints on sentation, nor does it matter what it was larger goals of ethnographic research, the representation: made out of. What makes a difference is raising the possibility that traditional

© 2004 Wiley Periodicals, Inc. COMPLEXITY 23 Continuing this theme of ethnogra- ethnographic research within main- The CAS nature of ethnography pher presence, CAS helped see how the stream social science have been, for the also foregrounds its adaptive ethnographer is one among many agents most part, like discussing computer ca- strengths, because it can change with an active role in the research pro- pacity in cubic feet. You can do it, but based on what has been learned cess. This acceptance of researcher influ- does it make any sense? as research progresses. ence, compatible with CAS but dissonant with traditional social research method- ACKNOWLEDGMENTS social research will play only a partial ology, allows poststructural issues of Support by NIDA DA-10736 is gratefully role, or perhaps no role at all. agency, distributed authority and reflex- ivity to be systematically included. acknowledged. CONCLUSION The next section of the article looked This article asks if any value comes from at how algorithms produce patterns at REFERENCES looking at ethnography through a non- different levels of scale. This “fractal” 1. Bernard, H. R. Research Methods in Cul- tural Anthropology; Sage: Newbury Park, linear dynamic lens. I think it is fair to concept applied twice over. On the one 1988. hand, methodological fractals result conclude that recent work in this “new 2. Agar, M. The Professional Stranger: An In- kind of science” offers marginal ethnog- from the “rich point” cycle that describe formal Introduction to Ethnography; Aca- raphy, the step-child of U.S. social re- a core of ethnographic methodology. demic Press: New York, 1996. search, substantial intellectual backing. On the other hand, age-old concepts in 3. Agar, M.H. How to Ask for a Study in Qualitatish. Qualitative Health Research Ethnography, on the other hand, offers cultural anthropology, like “theme,” 1999, 9, 684–697. also label a fractal-like concept, sub- CAS, a form of social research compat- 4. Gleick, J. Chaos: Making a New Science; ible with its assumptions and objec- stantive rather than methodological. Penguin: New York, 1987. tives. If CAS means to investigate and Important here was the fact that the two 5. Waldrop, M.M. Complexity: the emerging theorize the social world in any serious kinds of fractals are not produced by the science at the edge of order and chaos; Simon and Schuster: New York, 1992. way, ethnography is the kind of re- same algorithms. Fractal-finding is a 6. Gaddis, J.L. The Landscape of History: How great strength, and strong bias, of eth- search it will have to do. Historians Map the Past; Oxford University Looking at ethnography through a nographic research. Press: Oxford, 2002. CAS lens also casts some age-old prob- Finally, the CAS concept of “land- 7. Macy, M.W.; Willer, R. From factors to ac- lems in a new light, adds clarity, and scape” added a new twist to issues of tors: Computational sociology and agent- based modeling. Annu Rev Sociol 2002, 28, suggests solutions that are beyond the representation. At a time when experi- 143–166. ments and innovations are encouraged, bounds of what traditional social re- 8. Lansing, J.S. Complex adaptive Systems. search frameworks can provide. are there any constraints on ethno- Annu Rev Anthropol 2003, 32, 183–204. First of all, we saw that algorithmic graphic representation? “Landscape” 9. Epstein, J.M.; Axell, R. Growing Artificial complexity is higher for ethnography by called attention to “selective pressures” Societies: Social Science from the Bottom Up; MIT Press: Cambridge, 1996. comparison with most social science from the world in which a representa- 10. Wolfram, S.A. New Kind of Science; Wol- methodologies, the first indication of its tion “struggles for survival.” The discus- fram Media: Champaign, IL, 2002. fit with CAS. This fit allows us to see the sion led to the strange idea that a good 11. Naroll, R. Quality control in cross-cultural sur- “unscientific” nature of ethnography as representation was one that got the job veys. A Handbook of Method in Cultural An- in fact the presence of characteristics done, rather than one that was con- thropology; Naroll, R.; Cohen, R., Eds.; Colum- bia University Press: New York, 1970. that are the normal focus of CAS. The structed in a specific way out of specific 12. Axelrod, R. The Complexity of Cooperation: CAS nature of ethnography also fore- materials. The idea hardly fits tradi- Agent-Based Models of Competition and grounds its adaptive strengths, because tional social research concepts of stan- Collaboration; Princeton University Press: it can change based on what has been dardization of procedure as the ulti- Princeton, 1997. learned as research progresses. mate benchmark of a good study. 13. Agar, M. Anthropological problems, com- plex solutions. Human Organization 2005, Next, we explored “sensitivity to ini- In the end, I think CAS promises an forthcoming. tial conditions.” This robust concept of interesting conversation with ethnogra- 14. Benedict, R. Patterns of Culture; Houghton CAS helps understand decades of con- phy to the benefit of both. This article is Mifflin: Boston, 1959. troversy over the influence of an eth- only a first step; it just begins to touch on 15. Opler, M.E. Apache Odyssey: A Journey nographer’s identity on study results. similarities among contemporary ethno- Between Two Worlds; Holt, Rinehart, and Winston: New York, 1969. Such influence is to be expected, given graphic issues and CAS frameworks. 16. Agar, M. Language Shock: Understanding that ethnography is a CAS. Rather than What I hope is clear, though, is that CAS is the Culture of Conversation; Wm. Morrow: aspiring to a delusional “objectivity,” a a transdisciplinary research framework New York, 1995. better goal is the development of an where we ethno-types can find some in- 17. Berlinski, D. The Advent of the Algorithm: attractor space of acceptable study out- teresting company, more interesting than The Idea That Rules the World; Harcourt: New York, 2000. comes. The variations in that space, due our traditional “social science” classifica- 18. Kosko, B. Fuzzy Thinking: The New Science to ethnographer differences, will in fact tion has allowed in the past. Unfortu- of Fuzzy Logic; Hyperion: New York, 1993. strengthen overall study results. nately, discussions about problems in

24 COMPLEXITY © 2004 Wiley Periodicals, Inc.