Xi’an Lectures It is an enormous honor to be invited to Xi’an Jiaotong University, and be invited to give two public lectures, each an hour long, with questions. I thank you for your invitation and your hospitality. This invitation results from contacts that were made through the Santa Fe Institute for Complexity Studies and through common interests in the study of social networks. I was asked to give a first lecture on social networks and a second on complexity. Because these two are intimately connected, I address the first in terms of foundations and the second in terms of conclusions. The foundations are those of the human sciences, or better, humanizing of science, with a focus on the role of Anthropology as providing integrative perspectives. The conclusions are those of networks and complexity as a central integrative perspective. Lecture I. Foundations (society and networks) The foundations I will speak about, then, are those of Anthropology as a discipline, in relation to the other disciplines; and the contribution of anthropological (w)holism to human knowledge and problem solving. Anthropology is holistic in the integrative sense, across disciplines, as well as across the various aspects that constitute human beings – physical and material, biological, historical, linguistic, cultural, and social. At a larger scale, this plays out as integration of disciplinary perspectives between the nomothetic sciences (searching for regularity) and the idiographic humanities (concerned with interpretive validity). Nomothetic Holistic Idiographic Physics Sociology Anthropology History Ethnography Generality Rigor Integrative Process Meaning Primary <-------------- Integration --------------> Deutero- (second-order) <-------------- Integration --------------> <-------------- --------------> Networks Networks Networks Networks Networks Theory Theory Theory Theory Theory Figure 1: Integration across disciplines from a Human Complexity Sciences Anthropology Figure 1 is a crude representation of the kind of integration I am interested in and its relation to illustrative disciplines from the perspective of anthropology as a Complexity Science. These compartments are not airtight and should not be seen as separate. Each, for example, studies networks, albeit in a different way. Each has its own theory, albeit different. Figure 2 shows how another discipline, Sociology, might relabel these columns differently. Nomothetic Social Physics Holistic Sociology EthnoSociology Idiographic Historical Sociology Ethnographic Sociology Primary <-------------- Integration --------------> Deutero<-------------- Integration --------------> <-------------- --------------> Networks Networks Networks Networks Networks Theory Theory Theory Theory Theory Figure 2: Integration across disciplines from a Human Complexity Sciences Sociology One of the challenges for interdisciplinarity to succeed is to find common languages in spite of differences in perspective, differences in terminology, differences in methodology, differences in modeling, and differences in theory. To bridge some of these differences at the personal level, as a network researcher, I might call myself an anthropologist but also an ethno-sociologist. I can easily call my discipline ethno-sociology. Relabeling of this sort might help to establish an interdisciplinary identity when giving a paper at a conference or teaching in a summer school, but it does not solve the problem of common languages across the sciences and humanities. You can see why I do not want to give a public lecture on social networks or to identify myself as a social network researcher. You can also see why I do not believe that it is useful to create a taxonomy of networks, organized by disciplinary definitions, as in Figure 3. Taxonomy? Physical Networks Biological Networks Social / Kinship? Networks Historical Networks Personal? Networks Figure 3: A Spurious Taxonomy of Networks: NOT a Complexity Sciences Perspective The fields of networks, bottom-up or adaptive simulation1 and complexity were already building common languages across disciplines in the 1960s (the period of my early 1 Agent based modeling (ABM), Complex adaptive systems (CAS), parameterized analytical simulations of emergent phenomena, and related approaches. 2 professional life), each on the basis of interdisciplinary research communities. Integration between these approaches, however, had already begun. Ithiel de Sola Pool, for example, made major contributions, during WW II, to quantitative analysis of networks and complexity in his studies of communications content and political elites, to “computer simulation of social processes, including the first computer simulation of decision making in international crises”2 and did “the first major computer simulation of the American electorate based on public opinion data and used to advise President Kennedy's campaign for the Presidency in 1960” (Etheredge 1997:1, see also Deutsch 1971). Stanley Milgram’s (1971) small-world problem is a good example of a general concept spoken in a common language, across disciplines.3 Network concepts, measures, processes, and models have become part of a larger scientific common language in the last few decades, and they are now widely studied not only empirically and theoretically but used in simulation studies and recognized as an essential way to formulate, compare, and analyze basic processes in the complexity sciences. For my part, it would be incorrect to say that I simple study “social” networks, as a part of social reality and theory, as in Figure 3. Rather, networks are part of my conceptual language for analyzing phenomena of all sorts, across the spectrum in Figures 1 and 2. This specialized common language is also fit – well suited – for historical, humanistic (e.g., text-based), and ethnographic studies. Network concepts also translate into several branches of mathematics. They complement and relate to other branches of mathematics – such as probability theory and calculus and differential equations among many others – that are used in scientific analyses and theory. Testable models and theories in my ethnography with Ulla Johansen, Process models of a Turkish Nomad Clan, for example (now being translated by Xi’an Jiaotong Professor Haifeng Du), are stated in a language of network concepts that connect to empirical social networks as understood and recorded through participant and systematic observation, on the one hand, and, on the other, to graph-theoretic models and mathematical theorems that provide the basis for testable theoretical explanations, as in Figure 4. Ethnographic data <--------------> Network measures <--------------> Formal theorems e.g. e.g. e.g. e.g. nondeterministic probabilisitic complex multivocal dynamical nondeterministic probabilisitic complex multivocal dynamical Figure 4. Explanatory Models, but not based on (Referential) Correspondence Theory What is interesting about the mappings in Figure 4 is that they are not based either on the classical theory of Aristotle where words and things are in 1-to-1 (referential) 2 Simulating the processes involved in the outbreak of World War I ("The Kaiser, the Tsar, and the Computer"). Sola Pool also contributed to the study of contact networks and influence and the impacts of new communication technologies. 3 The fields of networks, adaptive simulation, and complexity rediscovered themselves in an increasingly interdisciplinary complex systems science and in the late 1990s rediscovered the links between physics and social sciences when so many physicists were attracted to the parameterized small-world problem of Milgram as posed by Watts and Strogatz (1997). 3 correspondence, nor on deterministic physical theory, nor are they based on equation-based deterministic modeling. The links between data, measurable quantities, “entities” and theory in this case are often nondeterministic, complex, multivocal, and dynamical, and may also be probabilisitic. I can best illustrate this through my own work. Two contributions to network theory and research that I am known for are: a. Abstract role theory and the identification of multivocal roles in networks.4 b. Structural cohesion theory and multilevel cohesion in networks.5 These are models, respectively, that provide generally applicable identification of roles and groups in networks, be they in physical phenomena, biology, sociology, kinship and marriage (classical domains of anthropology, history, ethnography, and “network studies” proper, whether in technology (the Internet, power grids), biology (protein interaction), or culture (networks of meanings in texts). Note that meanings are parsed from their contexts by a set of inferences which do not include the idea that a single word has a single referent. Rather, what appear to be the same words (spelling, pronunciation), where they occur in mutually exclusive and contrastive contexts, are initially assumed to have different meanings. In general, it is relationships (i.e., networks) that define contexts. The extent of equivalence of positions of elements in networks – their abstract role (as above) – is determined by whether elements hypothesized to occupy the same role do in fact occur in the same contexts, and whether those that occupy different roles occur in the different contexts. Figure 5 shows how regular role relations are not necessarily unique but admit of multiple contradictory models.6 Figure 5: The Non-Uniqueness of Regular Roles 4 White and Reitz (1983), Reichardt and White (2007, 2008). Moody and White (2003) 6 Figure 5 is from White and Reitz (1983:210). 5 4 In Figure 5, we start with twelve people in the graph (network) to the upper left, each person or node with a kinship link R to a relative on their right, and with a sexual link S to a partner on their left. Here, there are two possible role structures. One is the structure-preserving map to the 4-role structure on the upper right, where each of the upper and lower nodes has a link S on one side and a link R on the other. We cannot map the upper and lower node together in a 3-role structure or get a 2-role structure because each would then be represented as committing incest, with R and S links to occupants of the same role. These R and S links could then be linked simultaneously to the same person, but this does not occur in reality. A mapping of the same quality – an accurate mapping of roles – occurs in the bottom graph, but now we have people assigned to six roles that are inconsistent with those in the upper right. Persons 1 and 3 are in the same role in the first role model but in different roles in the second. Thus there is no Aristotelian necessary referential 1-to-1 correspondence between person and role. Rather, there is a higher order correspondence based on structural or regular equivalence. This disjunction of role models – each with a different irreducible roleperspective – occurs once we admit that role relations that may be composed of multiple relations (so that you and I may be friends, relatives, or both friend and relative, with each possibility distinguished). This 1-to-many correspondence is unstable, and our perception of roles could shift between one and the other and both be correct but their intersection incorrect.7 Similarly, but for a different reason, there is an unstable ambiguity about the concept of cohesive group, which is another area of my contribution to network theory and analysis, along with important empirical discoveries where structural cohesion makes important empirical predictions (Powell et al. 2005). Johansen and I showed in our book, Network Analysis and Ethnographic Problems: Process models of a Turkish Nomad Clan, an ethnographic case where our nomads use the same word, “kabile”, in different contexts, for extended patrilineages, intermarried lineages forming a clan, and intermarried clans when they form a single tribe or subtribe. As you might guess, it is the notion of intermarriage as forming larger cohesion social units – although variable in size – that constitutes the common but sliding signification. So context itself is not always an absolute criterion for meaning. In this case the contextual variable of size is ignored. But a speaker might use the word kabile extended to larger units when stressing integration, and choose instead the word “aile”, meaning family, when the intent is to signal that the interests of “smaller” families might be opposed. Here, the conceptual structure underlying the use of words is fluid and shifting, not arbitrarily but signaling differences between integration and opposition, and reflects a larger social fields of potentially shifting alliances. The formal definition of structural cohesion at a given level k – subnetworks that cannot be disconnected without removal of k or more members – works perfectly in this case of shifting alliances, and works in many (indeed most) other kinds of applications of network theory because it intrinsically defines a determinate and powerful measure of how well a unique and Moreover, we could imagine the males in Figure 5 as exclusively on the “outside” in these role structures or on the “inside,” and still both perceptions of the role models would be correct. 7 5 perfectly defined maximal group of elements hangs together through bonds of positive relations. If people refer to their kabile at large, medium and smaller scale (tribe, clan, lineage; depending on context), moreover, they are technically correct that if each unit is connected cohesively (structural cohesion level 2 = biconnected, where every pair of couples has two or more connecting kinship paths) they are in one sense equivalent, but in another sense, the difference of scale in these three types of units will differ by the average length of the paths by which people are multiply connected in a structurally cohesive kinship and marriage group. Every such group, however, is uniquely defined by a chosen network of genealogical relations, with a determinate boundary that we call that of structural endogamy. (White 1997). Fast and Frugal (FF) Heuristics and Ecological Rationality I spoke of intention in how we use words and how we see patterns and showed the potential for complex interactive stability or instability that needs to exist in our models when we use networks, simulations, probabilies, or complexity in our empirical studies. Still, even with these instabilities, we can – and must – talk about causality through the study of dynamics. Rather than Aristotelian entities that are always fixed in nature, we have entities that are emergent. Human beings possess emergents that evolved from their adaptive histories. Here is an example of a FF heuristic for individuals to compute their membership in structurally cohesive groups. People can easily consider and experience socially how they are connected in two different ways to other people. If I have two different paths to person x in my social environment, it is likely that I will meet at one time person z who connects to x by one path, and at another time or even the same time meet w who connects to x by another path. In my world I can meet some distant x and in our discussion it turns out that we both know w and z, and we exclaim “Small world!” (even before or without knowing about Watts and Strogatz 1998 or Travers and Milgram 1969). People thus can easily identify many of those in their network bicomponent, their “social group” with reference to some relation. It is an easy theorem to prove that if everyone in a subset of people in a given radius (defining a network perimeter by geography or network distance across contexts or settings) belongs to ego’s “bicomponent” and this is true for most egos within this social perimeter, then it will contain a network “bicomponent” that is mostly valid for each and every person within it. This is a variant of the theorem that states: in a network connected by positive and negative ties, if each ego can trace only cycles of connections that are “balanced” (product of signs is positive, where ++ and – are balanced, and -+ +- are unbalanced), then the entire network is divisible into two sides, each with only positive internal connections and only negative intergroup connections. Going back to if we allow the perimeter for bicomponents to vary from person to person, of course these bicomponents will overlap and form larger communities. 6 In both these cases, bicomponents and polarized social worlds, people can solve a complex assignment problem by using only local knowledge by using a fast frugal (FF) heuristic, and their solution will match the solution of this problem by optimization methods applied to the whole network, but using many many more steps in the computation. The problem of finding all k-components of a rather dense network is maximally difficult computationally, but a problem that has fairly obvious and intuitively accurate solution for individuals in communities. Two competing paradigms of human sciences: Network Complexity versus Rational Choice So, it is time to talk about the alternative paradigm: the illusion that it is only “unbounded rationality” that is truly rational when there are FF heuristics for many problems that humans face recurrently where they can obtain solutions equally good or better using the human adaptive toolkit of FF heuristics. The study of the contexts in which FF heuristics outperform computer optimization methods is the field of ecological rationality studied by Gert Gigerenzer (2007) and researchers at the Max Planck Center for Adaptive Behavior and Cognition. They investigate reasoning and decision making under uncertainty at the levels of both individuals and social groups and find a large proportion of problems where humans outperform optimization algorithms.8 What the research on FF heuristics (ecological rationality, Gigerenzer, 2007) refutes is the idea that the theory, methods and models of “Rational choice”, including dynamical Game Theory, “Optimality”, and classical microeconomic optimization, which is “unbounded” constitute the only valid formal definition of rationality (e.g., Simon argues that people who make actual economic decision do not meet all the assumptions that the model requires, including perfect information and unbiased processing of the imperfect information they have)9, a premise that until recently has dominated mathematical economics and utility theory, especially after the advent of formal economics, Von Neumann and Morgenstern’s (1953) invention of Game Theory, and modern computational methods. These are useful “ideal models” of something, but that something is retrodictive, asking: Was a given behavior unboundedly rational, computationally optimal, as defined in the theory of “Rational choice”, even if bounded rationality is better at solving the problem? Network complexity, in contrast, is one of the areas where local behavior in the adaptive behavioral toolkits that have evolved, not only those of humans but of other organisms, yield in many cases to simpler, faster, and more efficient solutions to complex problems of perception and choice, and often better optimization that formal computational optimization. That is, the best decision algorithms are often not those of “Rational choice”. What his implies is that we should study actual networks, actual behavior, and actual heuristics used in cognition, those called, until now dismissively, “gut feelings” or 8 This also suggests that we would do well to put FF heuristics with contingent strategies for local behavior (depending on the problem) in our simulations if they are to be useful in understanding real-world behavior of humans and other living organisms. This is a goal of contemporary work in digital evolution models, formerly know as “Artifical life.” 9 Leaf (2008:6). 7 “intuition”, if we want to understand causality, the dynamics of complex interactions, make better predictions, design better policies, and design better solutions to human and ecological problems. Implications for studies of networks and society Society is not a self-organizing, stable, self-contained system, but like migration, politics, economics, and any other area of life, consists of networks of processes, internal and exogenous, in which there are many dynamically reactive processes. If used wisely, network studies include the time dimension and try to focus on those problems and areas where the data on the networks studied are taken to represent channels of actual interactive effects on other outcomes. With refinement of our methods of testing causal models, we might be able to identify causal interactions. But if we fit all of our data to a given model, even if fit to the is perfect, we do not have evidence of causality. Only if we set half our data aside to build a model, for period 1, and test the predictions of the model for period 2, and do this over many different contexts, times and places, will we have evidence of causality. To do this we can usefully combine the approaches of physics, concerned with generality, sociology, concerned with reliability and validity, anthropology, concerned with a holistic view of how pieces of the puzzle fit together and how to reconceptualize our research problems, history, concerned with valid inferences from data in space and time about actual temporal processes, and ethnography, concerned with what human life looks like in the full dimensions in which it is experienced. Formal modeling and mathematics has a vitally important place in understanding networks and society, or social processes. Mathematics – from graph theory to differential equations, and all the rest – has a role equally important to that of data collection and analysis because inductive inference alone cannot provide causal explanation. There are aspects of causation and explanation that are reached by deductions from axioms to consequences, and if these can be proven a priori they do not require empirical evidence. Rather, if the consequence does not follow in the empirical world to which the theory is applied, it must also be the case that the axioms did not apply in this case. The strategy I have used in my work – like that of measuring cohesion or role structure – is to weaken the axioms of model that might lead to explanatory causality until they can be weakened and generalized no further and still help to explain the phenomena. References Deutsch, Karl W., Platt, J. & Senghaas, D. 1971. “Conditions Favoring Major Advances in Social Science.” Science 171(3970): 450-459. Etheredge, Lloyd S. 1997. What Next? The Intellectual Legacy of Ithiel de Sola Pool http://web.mit.edu/comm-forum/papers/etheredge.html MIT Communications Forum. http://web.mit.edu/comm-forum/papers.html 8 Gigerenzer, Gerd. 2007. Gut Feelings - The Intelligence of the Unconscious. New York: Viking Press. Moody, James, and Douglas R. White. 2003. Structural Cohesion and Embeddedness: A Hierarchical Conception of Social Groups. American Sociological Review 68(1):1-25. http://www2.asanet.org/journals/ASRFeb03MoodyWhite.pdf Outstanding Article Award in Mathematical Sociology. American Sociological Association, 2004. Powell, Walter W., Douglas R. White, Kenneth W. Koput & Jason. Owen-Smith. 2005 Network Dynamics and Field Evolution: The Growth of Interorganizational Collaboration in the Life Sciences. American Journal of Sociology 110(4):901-975. Viviana Zelizer Best Paper in Economic Sociology Award (2005-2006), American Sociological Association. http://www.journals.uchicago.edu/doi/abs/10.1086/421508 Reichardt, Jörg, and Douglas R. White. 2007 Role Models for Complex Networks. European Physical Journal B 60: 217-224. http://intersci.ss.uci.edu/wiki/pw/ReichardtWhite.pdf http://eclectic.ss.uci.edu/~drwhite/pub/RW_6Page_JournalFormat.pdf Selected for Europhysics News 8(1) 2008 “Highlights”. Simon, Herbert A. 1957. Models of Man. New York: Wiley. ____. 1986. Rationality in Psychology and Economics. The Journal of Business 59;4, part 2, The Behavioral Foundations of Economic Theory, pp. S209-S224. Travers, Jeffrey & Stanley Milgram. 1969. “An Experimental Study of the Small World Problem.” Sociometry 32(4): 425-443. Von Neumann, John, and Oscar Morgenstern. 1953. Theory of Games and Economic Behavior, Third Edition, Princeton University Press. Watts, Duncan J., and Steven H. Strogatz. 1998. “Collective dynamics of 'small-world' networks”. Nature 393: 440-442. White, Douglas R. 1997. Structural Endogamy and the Graphe de Parenté. Mathématiques, Informatique, et Sciences Humaines 137:107-125. http://eclectic.ss.uci.edu/~drwhite/pw/strendo.pdf White, Douglas R., and Ulla Johansen. 2006. Network Analysis and Ethnographic Problems: Process Models of a Turkish Nomad Clan. Boston: Lexington Press. 2006 Paper http://eclectic.ss.uci.edu/~drwhite/pub/PMContFwd01.pdf White, Douglas R., and Karl P. Reitz. 1983. Graph and Semigroup Homomorphisms. Social Networks 5:193-234. http://eclectic.ss.uci.edu/~drwhite/pub/whitereitz.pdf 9 Lecture II. Conclusions (networks and complexity) I had six very good teachers of complexity. One’s interest is directed toward those approaches that inspired you. Each of these teachers pointed toward the links between networks and complexity. 1. Buckminister Fuller, Engineer and Designer, taught me the relation between dynamics and structure: the dymaxion house, the fuller dymaxion dome, dymaxion transportation systems, dynamical earth. At the heart of Buckminster Fuller's Dymaxion concept is the idea that rational action in a rational world demands the most efficient overall performance per unit of input. His Dymaxion structures, then, are those that yield the greatest possible efficiency in sustainable (hence alternative) technology. His belief was that several hundred years would be required for these advances to come forth into practice. Dynamical equipartition and utilization of energies were the basis of dymaxion designs. The heart of complex, efficient, sustainable structure and dynamics is simplicity in design. 2. Wikipedia:Adamson Hoebel, process models, cooperation and conflict, and an introduction to the Manchester School of Max Gluckman and his students and colleagues (Clyde Mitchell, Elizabeth Colson, Thayer Scudder) 3. Eric Wolf, the best of my seminar teachers, among many of the greats at the University of Michigan: “Anthropology is both a science and a humanity” ('Anthropology 1964 Englewood Cliffs: Prentice-Hall). 4. Edward Tufte, who taught the principles of visualization, visualizing quantitative information, one-day seminars. 5. Arthur Iberall, Engineer and Physicist, designer of the space suit for the first American space voyage, developed homeokinetics as a theory of causal complexity 6. Frank Harary, my teacher of Graph Theory (University of Michigan) and a delightful coauthor … 10