Petra Ahrweiler
Abstract
Applying computer simulations in Sociology: history and controversies
After sketching out the origins and the history of computer simulations in Sociology (focussing on digital simulations only, see for the broader the history of social sciences modeling Séror 1994), this contribution will outline the disciplinary controversies provoked by the application of simulation techniques using a few examples. Introducing the main line of discussion within these debates the following two views can be isolated:
The Con´s: Considerable parts of modern Sociology accuse the attempts for "social simulation" to be a contradictio in adjecto. Simulations have to rely on adequate "representations" of their target; but just this reliability has been dismissed by social constructivism. This dismissal includes advices how to avoid shortsighted perspectives: "Any study of mathematics, calculations, theories and forms in general should [...] look at how observers move in space and time, how the mobility, stability and combinability of inscriptions are enhanced, how the networks are extended, how all the informations are tied together in a cascade of re-representation" (Latour 1987: 246f). For social constructivists there is no way to model these overlapping processes of continuous compositions and de-compositions, of differentiations and fusions. What really happens cannot be objectivated in modeling "objects" and the influence of society on them. The only software package most sociologists would accept is the non-computational, implicit creation mode of society itself.
The applicability of computer simulations implies on the one hand a "realistic" perspective on reality: the elements to deal with are given as "datum". On the other hand it implies an intrinsic "order" of reality: the given elements act on each other causally. The relations given as equations or qualitative rules really exist, the objects force us to rationally accept their existing laws that science has to discover. The disciplinary identity of many sociologists questions these implications and, in doing so, doubts the applicability of simulation techniques. Max Weber's theory of concept formation as a basic epistemological foundation specified the respective notions and relations of history and social laws in a way that the following statements became shared belief in sociology's methodological repertoire: "objects" of the social sciences are not independently given as "data" from the outside, but are constituted within a selection process (realistic perspective) or a construction process (constructivist perspective) out of an immense variety of potentialities. These constitution processes are carried out by "observers" (cf. Luhmann 1990), who are themselves tied up in and influenced by the identified social environments. The basic thought of "object formation" epistemologically locates the relation of theory and reality/objects in a way that sociology's notion of "causality" cannot be but an ascription of theory: even describing the smallest piece of reality can never maintain to be complete. The number of causes for a certain event always is infinite.
The "objects" do not force the shaping of the theory - neither with respect to categories of description nor at all "its laws". If terminologies do not function to mirror social reality, if no causality and no "laws" can be discovered, there will be no plausible support for formulas of description and "computation". Even using modern simulation techniques like AI or NN would not make a difference for this position: whether reality is described as information processing of symbol systems or as activation patterns in network structures – this likewise implies "realism" and "intrinsic order". There is no "input" for (autopoietic) social systems, and the selfreferential reproduction modes of society cannot be formalized (Maturana 1991, Luhmann 1990). Modeling society (creating itself and its environment) is a "contradiction per se" (Schefe 1990).
The Pro´s: The second position introduced in this contribution does not accept the accusation to idolating undercomplex but managable models of society and pretending that they are the double of society. Actually, the protagonists of social simulation share a certain starting point with the position just mentioned. For them, investigating society means to cope somehow with the immense complexity of the target. Various changing parameters, changing combinations, growing and declining relations: what method of sociological research could be fit for investigating this area? To answer this question, social simulation is presented as providing the adapted methodology. Although "scientists involved in creating computer simulations of complex adaptive systems know it´s a painstaking, complicated process" (Features 1993), they offer to formalize social construction, to implement constructivist concepts like "operational closure", and to model something like "evolution" on a computer. Representatives of this position try to follow basic concepts of modern (constructivist) Sociology: for example, simulation pleatforms (e.g. Swarm) often contain a de-centralized collection of relatively autonomous learning agents who interact with each other and with a permanently changing environment. Typically, there is no central agent who dictates the behavior of all the others: each individual is able to choose its own behavior according to its own concepts, its inner state, and its evaluation of the "world". This takes place through communications with other agents due to the fact that communication events are the key elements of such systems in simulating processes of interaction between the agents. Furthermore, these systems perform the possibility that their agents take actively part in construction processes of system events, for example in carrying out simulations themselves in order to perform anticipatory actions. This position reacts on the challenges of social constructivism by the statement that it is not sufficient to speak about fragile and uncontrollable processes of growth and declination in social systems. Instead, chances of intervention and reasonable options for involved agents must be optimized. Therefore, we have to learn about the mechanisms, life cycles, and options of social networks. This task requires intense sensitivity analysis, "playing around" with parameters, in short: the setting of an artificial social experiment, which cannot be carried out in reality.