©Copyright JASSS

JASSS logo ----

Michael Drennan (2005)

The Human Science of Simulation: a Robust Hermeneutics for Artificial Societies

Journal of Artificial Societies and Social Simulation vol. 8, no. 1

To cite articles published in the Journal of Artificial Societies and Social Simulation, reference the above information and include paragraph numbers if necessary

Received: 17-Feb-2004    Accepted: 07-Dec-2004    Published: 31-Jan-2005

* Abstract

The inability to verify simulation behavior limits the veracity of our claims to the theories and assumptions underlying the design of artificial societies. Those theories in turn suffer from their own cultural preconceptions, such as the location of agency at the cognitive level. The following essay highlights these concerns and points to a solution. I contend that artificial societies are subjective exercises in imagination, our description of their dynamics on par with the "thick descriptions" of cultural anthropology. Hermeneutics, the interpretive methodology employed in that discipline, can assist designers to negotiate the interstices between micro- and macro-level perspectives on agency. The resulting interpolation of theories reduces the impact of observer bias, giving rise to robust descriptions of agent behavior. I address the computation of hermeneutics through its resonance with the philosophy of William Wimsatt, and treat Laitin's work in identity construction, through model complication and "docking", as an example of robust hermeneutics.
Hermeneutics, Validation, Artificial Societies

* Introduction

Artificial societies have subtly shifted the kinds of questions one might ask in the social sciences. Gilbert and Conte first noted a decade ago one may longer simply inquire "'what has happened' or even 'what might have happened' but rather 'what are the sufficient conditions for a given result to be obtained'" (Conte and Gilbert 1995: 2). As Epstein and Axtell phrased it in introducing us to their Sugarscape, artificial societies change the meaning of "can you explain it" to "can you grow it (Epstein and Axtell 1996: 20)." For Jim Doran this shift has had a concrete impact on his work integrating agent technology into anthropological research. Unlike social simulations artificial societies offer "insight into the relationship between micro-level cognition and macro-level social behavior" that "avoid problems of validation (since there is no specific target system to be validated against) (Doran 2000: 98)."[1] This freedom to visualize the ramifications of social theory beyond empirical constraints has given scientists of disparate disciplines a means to communicate. Lustick sees agent-based models as a means to formalize social constructivist theory, thereby "attracting a wider audience in political science, sociology, psychology, and anthropology to the potential of agent based modeling (1999: 3-4)." Axelrod views it as an "effective research tool" for the social sciences, allowing scientists to communicate "in spite of their differences (1997: 206)." For Conte, Hegselman and Terna artificial societies promise to capture the interaction between top down panoptic theories of social action (ie the Macro) and bottoms up emergent "Micro" theories of individual action (1997: 14). This and the explanatory power of emergence can set the stage for social scientists to confront the division of their disciplines. (1997: 8).[2]

Recent Challenges to Simulation

While social scientists may conjointly share those visions imparted by their simulations, translating the intuition such visions impart into substantial social science research remains limited. It requires the ability to explain the virtual emergence of social dynamics to social scientists untrained in programming, and, for those modeling culture, scientists who have eschewed formal modeling entirely. The inability to validate the performance of such simulations complicates translating their emerging visions beyond modeling circles. Any substantial exploration of simulation results would be seen to proceed not from a continuing theoretical discourse grounded in valid findings, but from the cultural preconceptions under girding the simulation. Duffy in reaction to his own query as to the source of the micro-rules determining agent behavior, wondered "why characterize an approach as bottom-up when the rules that agents must adhere to are assigned in a seemingly top-down fashion by the research himself (1998: 793)?" Among those preconceptions informing artificial societies, emergence would stand front and center. For Gilbert (1995, 1996, and 2000) the danger and saving power of artificial societies in sociology derive from the emergence of social facts from the agency of actors. Most exercises to simulate "abstract social processes" identified by Gilbert fail to address a key element in social systems; the emergence of agent behavior based on known social facts. (2000: 367) In addressing the need for a social constructionist simulation program, he warns that unless scientists also model those social processes of negotiation and interaction between individuals "(simulations) is likely to be doomed to play a role no more important than for example, sociobiology." (2000: 367) Informed by cultural preconceptions that reduce macro level social behavior to the individual, and bereft of real world populations for validation, these societies currently exist as little more than quantified metaphors, fantasies or toys.[3]

Placing Artificial Societies in context of the Human Sciences

Artificial societies put assumptions tested previously in substantive research to play, not to avoid the consequences of such play for theory development, but to welcome them. In this most rigorous of scientific adventures, where political scientists rub shoulders with engineers, cognitive scientists with archeologists, simulations assist us to imagine what could be. Ironically the emergence of social processes in such simulations plays a similar role to fieldnotes and interviews in anthropology. Ethnographers use this substantive material as grist for "thick descriptions" of culture. The validity of such descriptions, and the impact an author's scientific or cultural bias may have on their inscription have been integral to the continuing development of ethnographic techniques in anthropology. [4] Ethnographers rely on the intuition granted by such descriptions to forward their research, but at a price. They have learned that telling stories about a persons cultureÄö regardless of how factually based that story may beÄö make them participants in that culture, and accountable to it. The resulting discourse between scientist and subject has served to keep ethnographers honest about their fictions, and even heighten the communicative power of these inscribed visualizations. It has also become a hallmark of the human sciences.
In light of this irony, this essay considers the design and analysis of artificial societies as an inherently human science. Recent critiques of simulation practice mirror ongoing concerns in the fields of history and ethnography; notably "truth and its social location; imagination and formal problems of representation; domination and resistance; (and) the ethical subject and techniques for becoming one." (Rabinow 1986: 236) Scientists in these fields have wrestled with questions of ontology, reflexivity and validity since Dilthey questioned the science of history almost one hundred and fifty years ago. The resulting discourse may assist simulation designers with issues of observer bias, and suggest ways to co-operate Micro- and Macro- descriptions of culture. But we have to listen closely for the answers.
Hermeneutics, a longstanding method for the systematic explication of literary and social texts provides a practical guide for reconciling disparate cognitive and social theories underlying our metaphorical subjects. The typical association of hermeneutics with the holistic description of culture tends to obscure the systematicity of the method in its reductions, and its ability to reduce the subject to multiple organizational levels. Most important the discursive tactics of relying on multiple sources, and the ongoing critical dialectic central to the methods application make it a natural choice in addressing observer bias. How better to share your visualization of the subject than to make it intelligible to a diversity of scientific perspectives?
I argue here that hermeneutics can help simulation designers reduce observer bias by providing a framework for negotiating cognitive and socially oriented views of the subject. To introduce this method I follow something of a hermeneutical circle. I begin by discussing the practical aspects of validating artificial societies, move on to consider the reflexive nature of simulation design, and how the scientific and cultural perspective of the designer inevitably locates agency towards the lower levels of any experiential ontology. I return the dialectic along its trajectory, first by introducing a classical form of hermeneutics to describe how scientists may negotiate an agency distributed between the social and cognitive levels. Placing cognitive and social theories of self in critical relation to each other I then argue foregrounds their role as conceptual and practical tools, the instruments of science, without losing them to a navel-gazing reflexivity.
I finish the dialectic by introducing robustness, the comparison and interpolation of probabilistically independent heuristics to generate a view of the subject, as a suitable replacement for validity. A hermeneutics of simulation capable of generating robust descriptions of agent behavior would enable scientists to envision social dynamics across a broad range of disciplines. This review of discussions in the disparate fields of social simulation and the philosophies of science and technology presents the barest outlines of a method. Despite re-visioning Laitin's work in identity construction (Laitins 1998) following a robust hermeneutics I remain firmly grounded in the philosophical. I cannot claim the hermeneutics described here will heal those divisions constituting the social sciences; at least I hope not. I propose it here rather as a modality for their negotiation.

* Validity, Reflexivity, Ontology; towards a different science


Operationally agent based models and artificial societies are very similar, borrowing the same techniques from system dynamics, cellular automata, genetic algorithms and distributed agent systems (Moretti 2002: 53) They differ in the target system simulated, and of course in design of the research program. Simulations with real world populations as a target must include some means of validating results. In econometrics and parts of political science and sociology data sets for verification are plentiful. Other fields, mainly anthropology suffer a lack of relatively clean data. The relative supply of data sets however are a secondary concern. The principal difficulty comes in matching data sets to agent architecture. Studies focused primarily on the cognitive roots of social theory are a case in point. Doran finds the relative complexity of simulation architecture, especially the cognitive structure of agents, a significant hurdle for validation. "In fact, it becomes near impossible." (Doran 2001: 63) Assigning measurement values for simulation variables grounded in the theory of the model, rather than empirical reality presents another hurdle.[5]
Research programs in artificial societies avoid these difficulties by focusing instead on simulations capacity to engender new theoretical inquiry. This shift however does not entirely rescue artificial societies from the question of validity. Running iterations free from comparison to the behavior of real world populations simply means we had better have tested the performance of our theories before digitizing them. Even then the design and analysis of models around accepted theories harbors plenty of opportunities to inform our otherwise objectively derived propositions with cultural preconceptions. A scientific team may run the model hundreds if not thousands of times to test a vast range of parameters. "The more complex the agent cognition, the more adjustable parameters there are." (Doran 2000: 98) The resulting parameter space would require evaluative heuristics extremely sensitive both to system characteristics arising from low level implementation decisions and the analytical and cultural preconceptions of the scientists.


Our preserved theories and the world fit together so snugly less because we have found out how the world is than because we have tailored each to the other. (Hacking 1992: 31)
To avoid such preconceptions Doran suggests two routes. We can follow the hard sciences and mathematics and "make our assumptions precise and low level, and then observe what consequences flow from them." (Doran 2000) Complex behavior could arise unbeholden to the preconceptions of the involved through a careful examination of all possible trajectories to complexity. Not simply would their determination prove immensely laborious. The task would involve both the assumption that "relatively simple regularities (are) waiting to be found," and that a "preexisting conceptual repertoire" exists to capture those regularities. "To believe that (such repertoires) are sufficient, is to fall into the exact trap we are trying to avoid." (Doran 2000: 100) David Lane in discussing the deeper insights derived from the subtexts of model performance, essentially describes the same dilemma. The presuppositions we use to wring meaning from artificial societies are best accessed "through the models themselves, even though the very meanings of these models is derived from the presuppositions, whose meaning is apprehended through the models (Lane 1994: 2)."
Reflexivity, the rebounding relation of cultural, analytical and technical preconceptions with the subject, becomes the principal challenge underlying artificial societies. Woolgar defines reflexivity as "the intimate interdependence between representation and represented object... such that the sense of the former is elaborated by drawing on knowledge of the latter, and knowledge of the latter is elaborated by that which is known about the former." (1988: 32) These rebounding relations move from the technical level, with specific choices in parameter settings and modeling paradigm (system dynamics or genetic algorithm for example), to the rationale for the study, and the accompanying theoretical and professional issues the study was meant to address.
Professional assumptions and cultural preconceptions informing concepts of agency and social dynamics often lie at the very root of a social scientific perspective and beyond. Emergence, the appearance of orderly behavior at the macro-social level arising from the micro-level activity of its constituent members has been an organizing paradigm in the field of simulation science for over a decade. While commonly viewed at least until recently as means to demonstrate that simple causes underlie complex social phenomena, it often simply ends up reifying the behavior of precedent models of Homo Oeconomicus. Hayles (1999), described the difference between emergence and analytical reduction this way.
Instead of starting with a complex phenomenal world and reasoning back through chains of inference to what the fundamental elements must be, they (simulation designers) start with the elements, complicating the elements through appropriately nonlinear processes so that the complex phenomonal world appears on its own (232).
Authors in sociology and anthropology have developed ways to handle reflexivity in their fieldwork and thick descriptions. Karin Knorr-Cetina explained her management of reflexivity to Callebaut this way:
You have to show that you are reflexive about the reflexivity issue, that you know that your own account is also just an account and not reality as it really is, and you have to invent devices to show that, like including dialogue in your presentation instead of just telling a straight story and thereby making clear that you are aware of the interpretive flexibility of things-issues like that. Callebaut 1993: 116)
Helmreich (1998) achieves something like this by describing in detail the academic contexts and personal development that brought about his ethnography of A-Life developers at the Santa Fe Institute. By recounting verbatim the harsh critique of his work delivered by Tom Ray, inventor of the Tierra simulation, Helmreich reflects on a crucial weakness of cultural studiesÄö the loss of the historical individual to their cultural contexts (1998: 240).
Doran's alternate route for salvaging objective insights from simulation performance would imbue agency with reflexive behavior. In considering different structures for agency in simulations, he recognizes "agents may have a view of themselves... designed according to an egocentric perspective, but with a sociocentric view of themselves."[6] Gilbert, working from Giddens' Structuration Theory, that "people are routinely capable of detecting, reasoning about and acting on the macro level properties (the emergent features) of the societies of which they form part," has long argued for the construction of self reflexive agents (1995: 151). In artificial societies he characterizes this dynamic as "second order emergence... in which 'the system (i.e. The individual) is able to detect, amplify, and build upon emergent behavior" (Gilbert 1999: 366 quoting Steels 1995: 90, quoting Baas in turn).[7] Steels sees this behavior in the emergence of new, qualitatively different obstacle avoidance strategies in robots, where "a variety of factors, some of them related to internal structures in the agent (in this case associative learning networks), some... to the properties of certain sensors and actuators, and some...to the interaction dynamics of the environment" play a part. (Steels 1995: 101) The multiplicity of factors giving rise to second order emergence in robot behavior mirrors the complexity of factors comprising a self-reflexive agency, including the role of context dependent meaning in constituting social action, the kinds of interactions between agents, and the negotiation of social identities within specific contexts. Second order emergence then effectively inverts the reduction to the individual implicated by first order emergence, while remaining dependent on it. This may involve the distributed agency or subagent format mentioned by Doran, featuring the constant negotiation of egocentric and sociocentric loci.


Managing reflexivity to reduce observer bias involves addressing the ontology of our simulations. As Doran previously charged we tend to locate simulation dynamics in individual agents. Some of the first major simulations of culture, notably the Sugarscape of Epstein and Axtell (1996) and Axelrod's Social Influence model (1997), reduce cultural dynamics to processes occurring within individual agents. The dynamics of identity construction in Lustick's PSI toolset (2002) adverts agents to their social surroundings without consideration for their psychological makeup. Conversely Haddadi's work modeling speech acts (1996) or Doran's modeling of collective misbeliefs (Doran 2000) both focus on the beliefs, desires and intentions of the agent. This reduction of simulation dynamics is hardly new or surprising. The in principle reduction of systemic behavior to one level, be it the micro or macro, has long been the hallmark of empirical modeling. When operated exclusively in an individualist or holist fashion however it tends to alienate those very scientists who might assist in complicating model behavior.
Designers can reduce the effects of cultural bias that accompany the reduction of agency, but it requires company. It would involve not simply a distributed agency between micro and macro, but the know-how to formalize theories informing those agents. To properly relate perspectives on the ego and sociocentric one would need cognitive scientists, sociologists, and anthropologists expert in their fields, and reflexive about their practices and perspectives. Thus a reflexivity in simulation design would require locating different scientists, theories and ideas on agency in a cooperative research ontology. The types of confrontations that would arise from such an exercise, micro vs. macro, individualistic vs. holistic, explanatory vs. interpretive, have already been forecast in the literature, as has the solution (Conte, Hegselmann and Terna 1997). Democratizing simulation design would reduce bias in an avowedly descriptive enterprise. Just getting cognitive scientists or economists working from a rational choice perspective to fruitfully discuss social simulation with social anthropologists accustomed to narrating thick descriptions of culture would prove difficult. Either side would need a common conceptual framework through which to approach such discussions. Given that two such groups could agree to the substantive study of a commonly viewed subject, the question becomes how you structure the resulting negotiation between the disciplines, how you visualize the discourse.

* Hermeneutics

We will try to find a point of view which acknowledges the disciple of logic without fallng into other paradoxes of logicism and which acknowledges the constructive role of concepts and the influence of the scientific community both on their origins and how they are employed without slipping in the nihilism of post-modernism.

Harre (2002: 219)

Casting artificial societies as a method more at home in the human sciences simply recognizes what many in the field have long known. The inability to verify simulation behavior with real world data renders those simulations purely descriptive. Unlike the "thick descriptions" of ethnography however artificial societies can describe multiple outcomes through strictly formal means, making their outcomes amenable to quantitative analysis and statistical application. [8] These two aspects alone have brought many scientists to examine and play with artificial societies; the method can already boast of multidisciplinarity, at least between scientists of kindred perspectives. I suggest in the following that a hermeneutical perspective can build on these strengths while reducing cultural bias through the reflexive management of our reductions. To that end I will rely heavily on several authors: F.D.E. Schleiermacher for his universal and mechanistic form of hermeneutics; Don Ihde and Patrick Heelan for their advocacy of the hermeneutic method in science, and William Wimsatt for a compatible realist philosophy.[9] Ihde's Instrumental Realism demonstrates a deep resonance between realist authors like Hacking and Wimsatt and phenomenological hermeneuts like Ihde, Heelan and Dreyfus, one untainted by cross citation (Ihde 1991: 97). An ontologically rich, multidisciplinary and reflexive construction of artificial societies needs a practical guide for computing agency. Wimsatt proposes a complex ontology, an instrumental approach to theory building, and a robust, collaborative approach to verification for just such causal mechanisms.
Having completed one part of this hermeneutical circle, from validity, to reflexivity to ontology, I will now retrace my steps through avowedly interpretive territory. A robust hermeneutics would describe a rich ontology for our simulations, locating agency at both the micro and macro levels. The resulting confrontation can work for us, should we render our theories and models instrumental in their use, metaphorical in relation to the subject, and successional in their piecemeal improvement. A truth claim of sorts, notably the robustness of diverse theories in dynamic relation would emerge from such a collaboration. I consider two different means of modeling Laitin's work on ethnic identity and change (1998) to demonstrate how a scientific community might bring the robust hermeneutics described here into play.

Restructuring Inquiry

(Charles) Taylor observes, 'the meaning of a word depends... on those words with which it contrasts, on those which define its place in the language... on those which define the activity... in which it figures.' Consider the following modification. the meaning of the datum depends on those data with which it contrasts, on the territory that defines its place within the larger dataset, on the context to which it relates...(he adds) the problems of data interpretation parallel those of textual interpretation. (Kritzer 1996: 4)
Hermeneutics can create holistic interpretations of social and literary texts, but only through the careful reduction of elements to their respective levels. Like a social system comprised of nested and interlocking structures from the macrosocial to the individual a text features a similar layering and correlation of parts at various levels: chapters, sections, paragraphs and individual phrases. Working at both synchronic and diachronic analysis of words and sentences, "every effort to determine and fix the meaning of particular passages by interpreting them separately should be part of a cumulative process that ultimately determines the precise meaning of any particular passage in terms of its context." (Schleiermacher 1997 [1977]: 71) Understanding emerges from the analysis of sentence structure in relation to the paragraph, the work and other works of the author. As with the individual, where one might examine the beliefs, intentions and attitudes constituting their behavior, one may continue a literary analysis both into the morphology and the syntax of individual sentences. "Complete knowledge always involves an apparent circle, that each part can be understood only out of the whole to which it belongs, and vice versa. All knowledge which is scientific must be constructed in this way" (Schleiermacher 1997 [1977]: 113).
Wimsatt describes a similar ontology of complex systems for causal mechanisms. The reduction of causal mechanisms to one or more levels depends on the system in question. Physical systems may be explained through a one to one reduction of parts and relations. (Wimsatt 1974, 1987) Biological systems being more complex may require scientists to exercise a complete or partial one to one reduction, or a many to one reduction between increasingly more diffuse levels. Eventually however systems (physiological, behavioral, environmental etc.) become too complex for simple reductions. Wimsatt accounts for observer bias at this stage by referring instead to perspectives, "quasi-subjective cuts (relative to the observer, the techniques and technology employed) on the phenomena characteristic of a system, which needn't be bound to given levels." (1994: 222) Sciences like anatomy and physiology, and more loosely economics and political science, that analyze system properties as descriptively complex in their parts and interactionally complex in their relations (i.e. reducible to multiple levels) fall under this rubric. (1974: 70)
While economics and international relations may loosely carry the title "perspective," for most human activity the explanatory power of a single perspective simply will not do. Immensely complex issues like ethnic identity rely on a multitude of causal interactions ranging from the neurophysiological to the psychological, linguistic, social, economic, historical, cultural, religious environmental etc. The limitations automatically built into each perspective make it difficult for them to "go-it-alone" in observing this phenomenon. These "causal thickets" require the coordinated operation of several different perspectives for identification and explanation. (1994: 273)

Building a Reflexive Scientist

The resulting negotiations to define the common subject would foreground the applied theories and models as useful descriptions rather than possible representations of social dynamics (Heidegger 1997 [1962]).[10] To that end Patrick Heelan suggests a different semantic role for our models, "that of helping to create new descriptive concepts by organizing new sets of perceptual profiles." (Heelan 1980: 21) Models in this sense serve "as a kind of quantified metaphor...suggest(ing) avenues for systematic perceptual inquiry."(Heelan 1983: 22) Heelan's phenomenological view of modeling comes directly from Mary Hesse, who, borrowing in turn from Max Black (1962) re-cognizes models as "metaphoric redescriptions of the explanandum." (Hesse 1980: 111) Working from a philosophy of ordinary language, scientific models borrow elements from the observation sentences they subsequently reorganize, thus becoming more dynamic in their application.
To render these metaphoric redescriptions both reflexive and corrigible Ihde (1991) considers our models and their constitutive theories instrumental. Scientific instruments, as theories and assumptions technically and contextually embodied provide a more economical, intuitive means of representation. This holds especially true of instruments that provide for an isomorphic relation with the subject, such as telescopy or telephonics. The closer one's models come to re-presenting the explanandum in the abstract, either through sonar, magnetic resonance imaging, or DNA testing, the greater the number of assumptions informing them, the more we experience the subject with our tools, rather than through them. We move from an embodied relationship with the perceived subject to a hermeneutic relationship with our "'textlike visualization'" of the proposed subject. (Ihde 1991, 1998: 166) Tackling artificial societies instrumentally would foreground the operative assumptions of a theory in light of the data sets, subjects, historical parameters, logical constitution, immediate professional contexts and more mediate cultural preconceptions constraining it.
Rendering such instruments computable means seeing them as something more than metaphors, and something less than theories. In a universe that requires cause and effect, Wimsatt suggests using heuristics as a convenient means of conceiving the subject without losing it to "sufficient conditions." Heuristics are a "cost effective" means of transforming a problem into a "non-equivalent, intuitively related problem" that "make no guarantees they will produce a solution or the correct solution to a problem." (Wimsatt 1985: 295) Unlike theories and models exercising a covering law capacity heuristics tend to locate properties wholly within a system and/or reduce properties to the next lowest system. (1985: 301-303) Models of emergence for example exercise a heuristic of conceptualization, that descriptively locates "a relational property as if it were a monadic or lower order relational property," and a heuristic of model building, that "look(s) for an intrasystemic mechanism rather than an intersystemic one to explain a systematic property." In so doing it "(simplifies) the description of the environment before simplifying the description of the system." Such reductions are necessary for useful model building. Scientists can reduce the negative impact of the above heuristics by co-operating their respective self reflexive agents. This careful comparison of heuristics for stepwise refinement within bounds of the models design defines their reflexive character.

Turing validity upside down

Thus different schools so to speak will rise among the masters, and different parties among the audience as followers of those schools, and even though the method is basically the same, different translations of the same work, undertaken from different points of view, will be able to exist side by side; and we shall not really be able to say that one is, as a whole, more or less perfect than another; only that certain parts will be more successful in one version and others in another, and not until they are all taken together and related to each other... not until then will they completely exhaust their task, for each one in itself will always be of relative and subjective value only (Schleiermacher 1982: 19).
Following the circle we have almost returned to Doran's conundrum: how can we avoid our cultural preconceptions in designing and analyzing artificial societies? Reconceiving our theories and models as heuristics both instrumental and metaphorical renders them corrigible to scientists of different viewpoints within an ontologically rich simulation. Rather than claiming these simulations represent real world dynamics, cooperating scientist would instead claim to have described possible realities for substantive exploration. I think this cooperation sets the stage for an alternate truth claim, robustness, to become available. Wimsatt puts forward robustness (1994: 210) as a criterion "working scientists" can fulfill well enough for the stated aims of a particular program of substantive research.[11] Something is robust if it is "accessible (detectable, derivable, definable, producible, or the like) in a variety of independent ways." (Wimsatt 1994: 210) He includes observation and measurement, mathematical and logical derivation as well as experimental manipulation as means of access to a phenomenon, "with many stops in between." Research then becomes an issue of matching results from these different means of access as well as the scientific assumptions informing the means. This process can first establish the existence of the studied phenomena, and report differences between the means to tell us more about the object. (215) Though the heuristics involved do not have the same force as deductive analysis, their mutual coordination and interpolation reduces the possibility for error in measurement.
If the checks and means of detection are probabilistically independent, the probability that they could all be wrong is the product of the probabilities that each could be wrong, and this declines very rapidly as the number of means of access increases even if the individual means are not very reliable. (Wimsatt 1994: 214)
Perspectives provide the context then for the evolution of heuristics in rebounding relation to the subjects they inform. The mutual reduction and improvement in heuristics operated robustly points for Wimsatt to a piecemeal development of theories he would term "successional." (Wimsatt 1976: 682)
Ihde in generating a strong program of hermeneutics in science recommends an application of multiple instruments to uncover regularities in the perceived phenomena. (1998: 172) To achieve this he describes several "multiple instrumental relays. These include the congruent emergence of scientific objects with the development of new instruments, the 'robust' discovery of an object using different instruments, and the application of different instruments to reach agreement on given aspects of a system (1998: 175) Regardless of which approach they take scientists need a "process (that) will necessarily include critical means for sorting out falsehoods, 'dead ends' deceptions and the like from within its 'texts.'" (1998: 185) In terms of eliminating artifacts from scientific data Ihde recommends a robust approach certainly in keeping with Wimsatt -
If one type of technology creates an 'artifact,' it is unlikely that the two or three or more other instruments which are aimed at the same phenomena will duplicate the instrumental artifact. Similarly a multivariant set of measurements can be regarded as more rather than less accurate when these different measurements converge. (1998: 186)
Thus objectivity derives not from the close match between covering laws and simulation output, but rather, as Heelan puts it. From a "reciprocity of perspectives" (Heelan 1983: 186) where concepts are "intersubjectively testable by normal adults who share a world." (1983: 189) This "intersubjective" sharing may occur between "compatible contexts of reality and discourse" or, as is more often the case between theories and methods extending across multiple professional and temporal contexts, between incompatible viewpoints. In this case those points shared between contexts may be considered "complementary statements." (1983: 179)
A robust hermeneutics of simulation design then would provide an instrumental and reflexive framework for social scientists to negotiate a common subject for general modeling purposes. Like the micro-rules governing agent cognition and behavior, scientists would cooperate diverse theories of individual and social behavior to lay the groundwork for subsequent design. Not simply would the resulting model exhibit less cultural bias but should also prove intelligible across several disciplinary perspectives. Far from limiting the descriptive power of simulations, their ability to visualize what could be, the contextualization and reflexive negotiation of their construction would make such artificial societies intelligible even to those of a pure interpretive vein.

* Possible applications

To visualize a robust hermeneutics in action I turn to the political scientist David Laitin. His Identity in Formation (1998) employs several theories and methodologies from different ends of the social scientific spectrum. The unique blend of colonization and nation building that characterized the evolution of the Soviet Union saw the development of a sizable Russian speaking population in the various republics. After the fall of the Union and the establishment of strict language laws in many of the ex-republics, these populations were faced with a terrible choice; return to a Russia many never knew, or adopt a foreign language (and thus culture) as their own. Laitin, through historical review, formal modeling, participant observation, linguistics experiments and statistical surveys found many of these populations had a third choice however; to become a member of a Russian speaking population in their new countries. (Laitin 1998: 33) Laitin describes resistance and assimilation to cultural change through Harre's (1984) theory on identity construction . The actual mechanism for identity change he posits to Schelling's (1978) tipping game model. Despite the diversity of methods and theories employed by Laitin his study reduces the social, linguistic and historically constrained behavior of his subjects to the dynamics of the tipping game.

Two Approaches

Following a robust hermeneutics Laitin would reexamine the relationship of identity construction and personal choice in terms of an agents internal states and social contexts. In formalizing this new study Laitin would have two routes of action open to him. He could work with corresponding scientists and engineers on designing a model of ontologically distributed self reflexive agency. He could also adopt previously examined models through alignment or "docking" to simulate possible outcomes. Each approach has its advantages and drawbacks. A single more complex model would have greater explanatory power in that agencies at different levels would have clearly discernable relations. The parameter space would be enormous. The micro-rules informing agent behavior would also clearly display the assumptions employed in patterning relations between agencies at the personal, social and even environmental levels. The study would need a section to explore the immediate technical and theoretical constraints of those assumptions, and another section to discuss the mediate professional and cultural constraints as well. The further complexity of model design and the development of a shared theoretical standpoint would assuredly place additional financial and time constraints on the study.
To conserve financial and professional resources scientists could also employ several previously developed models in relation to each other. Axelrod developed model alignment to acheive a "'domain of validity'" similar to that found in mathematized theories (Axelrod 1997: 184). Aligning simulation mechanics to compare the results supports two "hallmarks of cumulative disciplinary research: critical experiment and subsumption." Axelrod explains that "If we cannot determine whether or not two models produce equivalent results in equivalent conditions, we cannot reject one model in favor of another that fits data better; nor are we able to say that one model is a special case of another, more general one." Laitin experimented with docking his cultural influence model (called the ACM) to the cultural dynamics found in the Sugarscape of Epstein and Axtell (1996) to establish the one as a specific case of the other.
The alignment process poses many unique opportunities to explore the similarities and differences between models, returning a "robust" portrait of agent dynamics. Instead of developing a single model with complex agent dynamics Laitin could design and build a model from composites generated using Doran's SCENARIO-3 and Lustick's PS-I toolkits to see if the composites acted as special instances of each other. But the subsumption or reduction of one simulation as a more specific instance of another sacrifices theoretical features instrumental in describing real world dynamics potentially important to the study. Indeed Axelrod, Axtell and Epstein could only establish complete distributional equivalence of statistical results between the simulations once the micro-rules determining agent behavior in the hybrid model were practically identical to those in the ACM. Burton considered the process of "docking" a powerful but rare opportunity to increase the validity of a simulation, and generally favors the construction of "new and more complex models." (Burton 1998: 226)[12] He would consider the more traditional literature survey in conjunction with thought exercises to compare simulation results as sufficient for establishing validity. (1998.: 227) This comparative approach would keep intact those theoretical elements instrumental to each model. Either "docking" or the comparison of simulation results would cost less than generating a new model whole cloth, and would take less time too.[13] The sacrifice of simulation elements important for "docking," and the inability to measurably relate different agencies through comparison of simulation results reduces the explanatory power of this approach.

Exploration, Negotiation, Change

Whether building a more complex model to explore cultural identity and change, or cooperating various models Laitin would engage in the same circle of exploration, negotiation and change. The exploration of daily instances of cultural resistance and assimilation, and the choices people made in those instances would naturally lead to an exploration of possible relations between the applied theories. Negotiating the exact relations of personal choice with identity construction would proceed to a formal stage in designing a self-reflexive agency. This singular agent, or distributed "agency" extant at two ontic levels would negotiate responses to its environment from an internal structure of beliefs desires and intentions on the one side, and a socially defined repertoire of identities on the other.[14] One could then design identity entrepreneurs as agents with a greater view of its surrounding neighborhood, and thus a larger identity repertoire. At each stage of design and analysis Laitin would work with the research team from his study, cognitive scientists and simulation engineers to negotiate the technical exigencies and theoretical ramifications informing their agent. At each stage members of the group will have to explicitly address, if not the mediate professional and cultural contexts informing their theories, than certainly the immediate contexts of theory production. Agreements in views of the subject will simply bolster the robustness of the resulting simulation. Differences between team members and their theories will point to avenues of further, fruitful research.

* 5. Conclusion

While addressing the polarisation of the social sciences Conte, Hegselman and Terna found a number of phenomena would bring hidden similarities between the disciplines to light, and lead, if not to a reunification of the social sciences, than at least the confrontation of those similarities (1997: 9). A robust hermeneutics of model design meets the criteria for supporting those phenomena. It would support the cooperation of previously exclusive theories. Though differences between agencies at different ontological levels would remain intact in such a program, working respective reductions of the subject would generate more complex, compelling descriptions of social dynamics by generating agents who learned in both a recursive and discursive fashion. The foregrounding of theoretical and disciplinary structures in such an exercise would reduce the inescapable effects of cultural bias. Whether in the creation of more complex artificial societies or the correlation of results and dynamics from existing societies a robust hermeneutics would free participants from epistemological differences and constraints of validity to pursue a common vision of the target system. And a robust hermeneutics finds a place for discussions of the cultural production of scientific knowledge, one from which the involved scientists may draw practical guidance; one that renders the creation of artificial societies a work of community.

Reservations regarding a robust hermeneutics

This essay proposes a hermeneutical framework for addressing philosophical issues of validity, reflexivity and ontology in the design of artificial societies. The role of simulation in a hermeneutics of science however has received scant attention in the literature. Though instrumentation provides an effective means for relating theories across disciplinary bounds Ihde limits its use to "contemporary imaging processes which make 'natural' things into scientific, and thus 'readable' visual objects (Ihde 1998: 181)." He specifically excludes computer models from his program. Given the space necessary for a phenomenological treatment of simulation I can understand his hesitation. This shortcoming does not jeopardize the proposed hermeneutics. The heuristics of William Wimsatt, though limited in their corrigibility to the immediate contexts of knowledge production in science would admirably serve the purpose suggested here. An analytical treatment demonstrating the rebounding relations between technical constraints, immediate theoretical and methodological concerns, and more mediate professional and cultural contexts, will have to await the phenomenologists.[15]

Fiscal constraints and boundary issues

More practical constraints on a practice of robust hermeneutics emerge from technical factors, notably the construction of models that can handle both micro level cognitive processes such as those explored by Doran and macro level social and institutional actors. As discussed here and elsewhere such multilevel modeling may have to await tools that can integrate agent based modeling with other simulation technologies.[16] Financial considerations also apply. Funding the travel, housing requirements and technical exigencies a small community of software developers, cognitive scientists, economists, and sociologists (not to mention the occasional anthropologist or philosopher) might face differs from the fiduciary constraints on a single researcher. And there is a social dimension to consider. On the way to building simulations of ontologically diverse, self reflexive agents scientists and designers would face multiple feats of translating terms and viewpoints into commonly shared ideas. The larger the community, the greater the task. Given a group of scientists committed to the exploration of simulations potential for the social scientific imaginary, an agreement on the basic issues constituting philosophical differences between scientists, and a mode for their negotiation, such challenges should prove surmountable.

Hermeneutics and the robustness of a "silicon second nature."

The Santa Fe Institute has devoted considerable attention to how systems survive and incorporate exogenous shocks in several disciplines, including molecular biology, ecology, and problem solving. (SFI 2003, 2000) With regard to social processes researchers at SFI plan to explore robustness through two complementary perspectives: as a dynamic system "with feedback across multiple scales and in multiple dimensions on multiple networks, and as a distributed information-processing system with feedback control."(SFI 2003) Through exercising these perspectives jointly while pursuing robustness in decision making, cultural traditions, economies and other areas scientists expect to incorporate useful aspects from engineering and ecology along with cognitive science and anthropology. At Santa Fe we see the stage set for the confrontation at hand. For social scientists to achieve a true entente regarding the micro macro distinction, where complementary agent mechanisms associate at different levels, they will first have to reflect on their assumptions and preconceptions informing their theories and methodologies. To render their simulations of robustness truly robust, scientists at Santa Fe will first have to place their science in a broad perspective, one that considers the professional and cultural contexts of its production.
By locating our selves in the immediate professional and more mediate social and cultural contexts that constrain our work, by recognizing our formal agencies as corrigible descriptive models for improvement, rather than attempts to model the dynamics of the real world, we make our agencies, theories indeed our worlds available to one another. Abbott in his fractal analysis of the intellectual and institutional distinctions informing the sciences suggested that "breaking those affinities (micro/macro, empirical/interpretive, realist/constructionist informing the disciplines) is the most powerful mechanism for knowledge change in social science (Abbott 2001: 29)." Placing artificial societies as a human science, akin to thick descriptions of culture, and amenable to a robust hermeneutics, increases the power of simulations to visualize our in silico civilizations. The democratization of assumptions, theory and evidence implicit in the proposed hermeneutics not only increases the designers ability to explain the dynamics of social processes to a wider audience, but addresses and improves on the systematic biases of their simulations as well. I am not proposing a "transdiscipline" to pave over methodological and theoretical differences between social scientists. Neither the philosophy expressed here, nor the technology it addresses can erase the divisions informing the social sciences. Rather I argue here for a practical means to negotiate the crossing of those divides.

* Notes

1 For the subtle difference between social simulations and artificial societies see Moretti (2002), Macal and Sallach (2001), Gilbert (2000) and Lane (1994).

2 The authors, in sketching the polarization of the social sciences, listed several assumptions the sciences unwittingly held in common, such as "the study of collective action can do without a model of agents internal states(,) one should take either a holist or individualist view (of the subject) (and that) social institutions and organisations are products of social action (and so cannot address their implementation onto individual action)" (Conte, Hegselman, Terna 1997: 8)

3 Timothy Kohler (2000: 1)admitted that his models of Anasazi settlement were "just a game... I'm happy to admit that it is, so long as our definition of of games encompasses child's play — which teaches about and prepares for reality — and not just those frivolous pastimes of adults, which release them from it.". Conte, Hegselmann and Terna (1997: 14) admonish designers of artificial societies to "render our fantasy controllable, public and explicit." In reviewing critiques of social scientific modeling, Gilbert characterized the disdain for such simulations thus: "Clearly, the whole enterprise is just an excuse for playing around with computers." (2000: 356)

4 See Sanjek (1990) for a discussion of ethnographic methodology. Nor would any mention of thick descriptions pass without some acknowledgement of Geertz' Interpretation of Culture (1973)

5 For example, when we define a variable that represents the prestige of a political community we can decide that it should assume a real value between 01 and 1. How can we interpret the values assigned to this variable during the simulation? What does a prestige value of .4 mean , in a dynamic system?" (Moretti 2002: 54)

6 Here Doran (2000: 101) operates a distinction between views of the individual first suggested by Mageo (1995).

7 Steels drew this phrase from a conversation with Baas, who in turn may have been referring to the "second order cybernetics" of Von Foerster (1960). Ironically this insight comes by way of N. Katherine Hayles (1999: 10).

8 Lustick (1999, 2002) found artificial societies the perfect means to formalize theories of social constructivism that hitherto had received only a qualitative treatment.

9 Ihde and Heelan are not alone in advocating the utility of hermeneutics for understanding the technics and praxis of science. See the Heelan Fetschrift edited by Babich (See Harre 2002) and the proceedings of the Hungarian Academy of Science edited by Feher, Kiss and Ropolyi (See Eger 1998) for other discourse on the subject.

10 I paraphrase here his "fore-structure of understanding", where the presuppositions underlying the construction of theories, and the theoretical advantages and technical limits of their use remain in view. For an update see Latour and Woolgar (1986)

11 Wimsatt attributes his development of the robustness criterion to Levins (1966), Campbell and Fiske (1959) and Campbell (1966) This is not the place for a thorough exploration of the emergence of robustness. However the resonance between robustness, schools of feedback thought described by Richardson (1991), the learning machine model outlined by Hesse (1980: 182) and the hermeneutic circle outlined here by Schleiermacher, Hegel Dilthey and others deserves further treatment.

12 Conte, Hegselman and Terna (1997) suggest the same.

13 Axelrod (1997) demonstrated the efficacy of model alignment by efficient management of project constraints and good use of electronic communications. He found the most time-consuming aspect to be debugging the hybrid model for the alignment runs.

14 Assuming here that one has designed a model that incorporates the AI of Haddadi's agents with the social repertoires of Lustick's. While perhaps the ramblings of an amateur philosopher unversed in the difficulties of simulation design the further incorporation of environmental elements such as those found on the Sugarscape (Epstein and Axtell 1996) would add an additional fold to the complexity of agent behavior.

15 Wimsatt has examined and designed agent based models to address issues of nonaggregativity in emergence, developmental constraints, and the units of selection controversy in evolutionary biology. His approach recognizes constraints on model building, but from an instrumental viewpoint; how can we build successively more powerful models to explain the data? He has not addressed the impact of social and cultural constraints exogenous to contexts of model construction.

16 Hans Scholl (2001) at the University of Albany has already outlines operational similarities between ABMS and Systems Dynamics techniques. Troitzsch et. al. (1996: 341) in applying a multilevel modeling software to questions of technological innovation and diffusion, saw the integration of agent based models the next step in developing their own more classical approach .

* References

ABBOTT, A. (2001) Chaos of Disciplines Chicago IL: University of Chicago Press.

AXELROD, R. (1997) The Complexity of Cooperation Princeton: Princeton University Press 181-195.

BLACK, M. (1962) Models and Metaphors Ithica NY: Cornell UP.

BURTON, R (1998) "Validating and Docking: an overview, summary and challenge" Simulating Organizations: Computational Models of Institutions and Groups ed. J. Prietula, K.M. Carley and L. Gasser, Menlo Park: AAAI Press.

CALLEBAUT, W. (1993) Taking the Naturalistic Turn or How Real Philosophy of Science is Done Chicago IL: University of Chicago Press.

CAMPBELL, D.T. And D. FISKE (1959) 'Convergent and Discriminant Validation by the Multitrait - Multimethod Matrix' Psychological Bulletin 56 81-105.

CAMPBELL, D.T.(1966) 'Pattern Matching as an Essential to Distal Knowing' The Psychology of Egon Brunswik K.R. Hammond ed., New York: Holt, Rinehart and Winston.

CONTE, R. AND N. GILBERT(1995) 'Computer Simulation for Social Theory' Artificial Societies: The Computer Simulation of Social Life Gilbert, N., R. Conte, eds. London: UCL Press

CONTE, R. HEGSELMAN and P. TERNA (1997) 'Introduction: Social Simulation: A New Disciplinary Synthesis' Simulating Social Phenomena Conte, R., Hegselmann, R. and Terna, P. (eds.). Springer

DORAN J.E. (2000) 'Trajectories to Complexity in Artificial Societies' in Dynamics in Human and Primate Societies Kohler, A., G. Gumerman, eds. Oxford University Press.

DORAN J.E.(2001) 'Can Agent-Based Modelling Really Be Useful?' Cooperative Agents: Applications in the Social Sciences Saam, N.J., B. Schmidt eds., Dordrecht: Kluwer 57-81

DUFFY J. (1998) 'Growing Artificial Societies: Social Science from the Bottom Up' Southern Journal of Economics 64: 3, 791-794.

EGER M. (1998) 'Language and the Double Hermeneutic in Natural Science' Hermenueutics and Science Feher, M., O. Kiss and L. Ropolyi ed., Dordrecht: Kluwer 265-281.

EPSTEIN, J. AND R. AXTELL (1996) Growing Artificial Societies: Social Sciences from the Ground Up Cambridge: MIT Press and Washington: Brookings Institution Press

GEERTZ C.(1973) The Interpretation of Culture New York: Basic Books.

GIDDENS A. (1979) Central Problems in Social Theory New York: Macmillan.

GILBERT N. (1995) 'Emergence in Social Simulation' Artificial Societies: The Computer Simulation of Social Life" Gilbert N. and R. Conte eds., London: UCL Press.

GILBERT N. (1996) 'Holism, Individualism, and Emergent Properties' Modeling and Simulation in the Social Sciences fro the Philosophy of Science Point of View Hegselmann, R., Mueller, U. and K.G. Troitzsch eds., Kluwer Academic.

GILBERT N. (2000) 'Modeling Socialty: The View from Europe' Dynamics in Human and Primate Societies Kohler, A. and G. Gumerman eds., Oxford University Press, 354-367

HACKING I. (1992) 'The Self-Vindication of the Laboratory Sciences' Science in Practice and Culture A. Pickering ed., Chicago: University of Chicago Press

HADDADI A. (1996) Communication and Cooperation in Agent Systems: A Pragmatic Theory , Lecture Notes in Aritificial Intelligence #1056, a Subseries of Lecture Notes in Computer Science, Springer-Verlag

HARRE, R. (1984) Personal Being: a theory for individual psychology. Cambridge: Harvard University Press

HARRE, R.(2002) 'Science as the Work of a Community' Hermeneutic philosophy of science, Van Gogh's eyes, and God Babich, B. ed., Kluwer Academic.

HAYLES, N.(1999) How We Became Posthuman Chicago: Chicago U.P.

HEELAN, P. (1983) Space Perception and the Philosophy of Science Berkeley U.P.

HEIDEGGER, M. (1977 [1962]) 'The Question Concerning Technology' Basic Writings New York: Harper and Row.

HELMREICH S. (1998) Silicon Second Nature Berkeley: U. of California Press.

HESSE M. (1980) Revolutions and Reconstructions in the Philosophy of Science Bloomington and London: Indiana University Press.

IHDE, D. (1991) Instrumental Realism Bloomington: Indiana University Press Indiana Series in the Philosophy of Technology.

IHDE, D. (1998) Expanding Hermeneutics: Visualism in Science , Northwestern University Studies in Phenomenology and Existential Philosophy, Evanston: Northwestern University Press.

KOHLER, T.A. (2000) 'Putting Social Sciences Together Again: An Introduction to the Volume' Dynamics in Human and Primate Societies Kohler, A., G. Gumerman eds., Oxford University Press

KRITZER, H. (1996) 'The Data Puzzle: The Nature of Interpretation in Quantitative Research' American Journal of Political Science (40) 1: 1-32.

LAITIN, D. (1998) Identity in Formation: The Russian Speaking Populations in the Near Abroad London: Cornell U.P.

LANE, D. (1995) 'Models and Aphorisms' Complexity 1; 2

LATOUR, B. and S. WOOLGAR (1986) Laboratory Life: The Construction of Scientific Facts Princeton: Princeton University Press.

LEVINS, R. (1966) 'The Strategy of Model Building in Population Biology' American Scientist 421-431.

LUSTICK, I. (1999) 'A Constructivist Agent-Based Model of Collective Identity and Identity Change.' Unpublished electronic mail received by Peter Schuh from Professor Lustick Feb. 22

LUSTICK, I. (2002) PS-I: A User Friendly Agent Based Modeling Platform for Testing Theories of Political Identity and Stability https://www.jasss.org/5/3/7.html

MACAL C, AND D. SALLACH (2001) 'Introduction: The Simulation of Social Agents' Social Science Computing Review 19: 3, 245-248

MAGEO, J.M. (1995) 'The Reconfiguring Self' American Anthropologist 97(2) 282-296.

MORETTI, S. (2002) 'Computer Simulations in Sociology: what contribution?' Social Science Computing Review 20: 1, 43-57

RABINOW, P. (1986) 'Representations are Social Facts: Modernity and Post-Modernity in Anthropology' Writing Culture: The Poetics and Politics of Ethnography Clifford, J., G. Marcus. Berkeley: California 234-261

RICHARDSON, G.P. (1991) Feedback Thought in Social Science and Systems Theory U. of Pennsylvania Press

SANJEK, R. (1990) 'On Ethnographic Validity,' Fieldnotes: The Makings of Anthropology R. Sanjek ed. London: Cornell U.P. 385-413.

SCHELLING, T. (1978) Micromotives and Macrobehavior. New York: Norton.

SCHLEIERMACHER, F.D.E (1982) 'On the Different Methods of Translation' Trans. A. Lefevere, German Romantic Criticism Ed. L. Willson The German Library Vol. 21, New York: Continuum.

SCHLEIERMACHER, F.D.E. (1997) Hermeneutics: The Handwritten Manuscripts H. Kimmerle ed., J. Duke and J. Forstman trans., Atlanta: Scholars Press AAR Texts and Translation Series #1.

SCHOLL, H.J. (2001) Agent based and System Dynamics Modeling: A Call for Cross Study and Joint Research 34th Annual Hawai'i International Conference on Systems Science IEEE: http://csdl.computer.org/comp/proceedings/hicss/2001/0981/03/0981toc.htm P. 3003.

SFI (2000) A Founding Program in the Study of Robustness Santa Fe Institute Bulletin 15(2) http://www.santafe.edu/research/publications/bulletin.php

SFI (2003) Robustness and Social Processes Santa Fe Institute Bulletin 18(1) http://www.santafe.edu/research/publications/bulletin.php

STEELS, L. (1995) The Artificial Life Roots of Artificial Intelligence Artificial Life: An Overview Langton, C. ed., Cambridge: MIT Press.

TROITZSCH, K. and K. BRASSEL, O. EDENHOFER, M. MOHRING (1996) 'Economic Development, Opinion Formation, and Technological Change in a Multilevel Simulation Model' Modeling and Simulation in the Social Sciences from the Philosophy of Science Point of View Hegselmann, R., Mueller, U. & K.G. Troitzsch ed., Kluwer Academic.

VON FOERSTER, H. (1982) Observing Systems Intersystems.

WIMSATT, W. (1974) 'Complexity and Organization' Philosophy Society of America: 1972 R. Buck, S. Cohen ed., Boston Studies in the Philosophy of Science Dordrecht: Reidel 67-86.

WIMSATT, W. (1976) 'Reductive Explanation- A Functional Account' Philosophy Society of America: 1976 Boston Studies in the Philosophy of Science, Dordrecht: Reidel 671-710.

WIMSATT, W. (1985) 'Heuristics and the Study of Human Behavior' Metatheory in Social Science: Pluralisms and Subjectivities D.W. Fiske, R. Schweder ed., Chicago: Chicago University Press

WIMSATT, W. (1994) The Ontology of Complex Systems: Levels, Perspectives and Causal Thickets Canadian Journal of Philosophy Supplementary Volume 20: Biology and Society: Reflections on Methodology, 207-274.

WOOLGAR, S. (1988) Science: The Very Idea Key Ideas Series ed. P. Hamilton Chichester: Elias Horwood.


ButtonReturn to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, [2005]