Mohammadreza Zolfagharian, Reza Akbari, Hamidreza Fartookzadeh

Transcription

1 Theory of Knowledge in System Dynamics Models Mohammadreza Zolfagharian, Reza Akbari, Hamidreza Fartookzadeh Abstract Having entered into the problem structuring methods, system dynamics (SD) is an approach, among systems methodologies, which claims to recognize the main structures of socio-economic behaviors. However, the concern for building or discovering strong philosophical underpinnings of SD, undoubtedly playing an important role in the modeling process, is a long-standing issue, in a way that there is a considerable debate about the assumptions or the philosophical foundations of it. In this paper, with a new perspective, we have explored theory of knowledge in SD models and found strange similarities between classic epistemological concepts such as justification and truth, and the mechanism of obtaining knowledge in SD models. In this regard, we have discussed related theories of epistemology and based on this analysis, have suggested some implications for moderating common problems in the modeling process of SD. Furthermore, this research could be considered a reword of system dynamics modeling principles in terms of theory of knowledge. Keywords: System dynamics, Theory of knowledge, Justification, Belief, Truth

2 1 Introduction In any discipline, there are always a number of underlying philosophical predispositions in the projects of scientists the reflection on which may enlighten the debates on the concepts and the processes in the discipline, as a result, it will ensure its vibrance in overcoming new problems and improving its capability against potential competitors. Furthermore, it is essential to comprehend the implications, the limitations, and the scope of our very scientific practice (Olaya 2009: p. 9057). System dynamics (Hereafter SD) as a systems methodology for modeling socio-economic problems is not an exception. The necessity of the explanation of these issues in SD will be more apparent if we take a look at the increasing number of modelers who have SD in their toolkits without any awareness of the potential effects of philosophical background on model conceptualization, formal model building, validation and policy design. Furthermore, remaining un-answered questions about these predispositions for developing this methodology neccesiate a more profound and a broader research on this scope in order to reinforce protective belt of SD against the existing criticisms. However, in this regard, during more than 50years since the emergence of SD, efforts have been made and different philosophical approaches have been proposed as theoretical foundations of SD. Sometimes, for instance, it has been suggested that certain varieties of moderate realisms such as that of Popper s falsificationism1 or critical rationalism, 2 could fit into SD procedures. Since causal models in SD offer clear test points by which problems could be solved and also theory could be advanced (Bell and Bell 1980; Bell and Senge 1980). Other times, it has been maintained that a certain practical relativism, 3 or a contextualist and pragmatist philosophy of science along the lines of Kuhn, would offer an adequate framework for the justification of the claims of SD models (Barlas and Carpenter 1990). Internal realism 4 of Hilary Putnam is another philosophy, whose proposers have believed that it can be adapted to the roles of mental models5 and, in general, conceptual problems in SD models (Vazquez et al. 1996). Recently there have been some efforts for using Constructivist of John Searl6 and Expressivist of Robert Brandom7 for explaining philosophical foundations of SD (Vazquez and Liz 2007). They have argued that a certain combination of the perspectives of Searle and Brandom could be very useful to achieve a reflective understanding of SD modelling. [Vazquez and Liz] applied constructivist and expressivist views in the discussion of three crucial problems concerning the validation of SD models: the ontological problem of realism with respect to the structures postulated in SD models, the epistemological problem of the explanatory value of SD models, and the methodological charge of merely producing a kind of patchwork. Finally, [they] generalized that constructivist and expressivist approach analyzes the interrelations between mental models, social systems and SD models. (Vazquez and Liz 2007: p. 18). Furthermore, there were some efforts for proposing a theoretical framework that can convey all SD activities (e.g. Lane 1999 and Pruyt 2006). In this regard, Lane unearths different forms of SD practices in Burrels and Morgan framework for explaining social theories of system dynamics and Pruyt classifies different SD practices in a paradigmatic classification framework on the basis of their basic assumptions, namely, Positivist system dynamics, Post positivist System Dynamics, Critical pluralist system dynamics, Pragmatist system dynamics, Constructivist system dynamics, Transformative Emancipatory, Critical system dynamics. Although, all of these articles have some points about the philosophical issues of SD and as a result, explicitly or implicitly, indicate notes on the epistemology or the nature of knowledge in SD models, but just a few of them consider the knowledge elicited from SD models justified true belief (the standard

3 analysis of knowledge in epistemology) 8 and based on that establish their theoretical analysis, which could open a new arena for the epistemological analysis of SD. Therefore, in this paper, falling back upon to the point that any model offers some knowledge around a problem, an epistemological analysis of SD models, based on justified true belief concept, is discussed. This will bring significant implications for the structure and validation of SD models and could moderate the debates of researchers about the common problems in the justification of SD models. To this end, beginning with a brief account of SD, and theory of knowledge and its core concepts (namely, justification and truth) and debates concerning them, it goes on to describe these concepts in SD models. It proceeds to suggest some implications for acquiring stronger and more valid knowledge following each category. 2 System Dynamics System dynamics is an approach, which has entered into the scientific and practical discussions in 1950s by J.W. Forrester, the professor of Massachusetts Institute of Technology. He saw the emergence of SD for university research in management as the same opportunity to assume aggressive, innovative leadership that we have seen in the great schools of medicine, science, and engineering. (Forrester 1961: p. 360). From the late 1950s to the late 1960s, SD was applied almost exclusively to corporate and managerial problems and its conceptual and practical framework was explained in two books of Industrial Dynamics and Principles of Systems. In 1968, SD has started experiencing new major projects such as Urban Dynamics and World Dynamics in different versions that drew world-wide attention toward viewing complex problems through the lens of SD. Nowadays, researchers from different fields have examined the capabilities of SD in facing the complexities of diverse problems and building simulation models (including quantitative and qualitative practical, Theoretical, interactive and learning models9) in the areas including business, economics, society, politics and culture and etc System Dynamics Modeling Process In describing modeling process, system dynamicists have organized stages varying in label and arrangement but the activities considered along the different stages remain fairly constant across them.10 Sterman (2000), in his explanation for modeling process, stated five iterative stages, namely, problem articulation, dynamic hypothesis, formulation, testing and policy formulation and evaluation that ensure two main functions of SD as the explanation of the structure of complex socio-economic problems and policy making for improving the current situation of the surveyed problem (Fig. 1). Figure1 It could be said that the most important claim of SD is mapping and changing the mental models of the clients (or the problem owners) in the process of modeling. However, this process could lead to the learning (with the aim of supporting reasoning, team learning, and nurturing system thinking and planning based on scenario-making) and improving some undesirable performance patterns. In this section, it will be tried to explain these stages, briefly, so that it prepares a common context for discussing epistemological issues of SD. 1. Problem articulation (boundary selection) The most important step is complex problem articulation. It proceeds from clarifying the clients problem by going through the evidences and accessing to the stems of the real problem that brings about a vicious

4 whirlpool in the activity boundaries of the problem owners. This will continue by eliciting and structuring the mental models of the group to define a problem. Establishing behavior reference modes11 and setting appropriate time horizon are two most useful processes, which modelers try to use for this end in SD modeling. 2. Formulating a Dynamic Hypothesis After problem articulation, modellers should start developing a theory about the problematic behavior that can be obtained in conversation and interaction with clients. This theory shows the process which is rendered by the problem and usually, it is provisional, because it may be changed (by expansion, revision and contraction of variables or loops) as you learn in the modeling process. In early stages of formulation, modellers should capture mental models of clients and develop maps of causal structure to recapitulate them by facilitation tools (such as Model boundary diagrams, Subsystem diagrams, Causal loop diagrams, Stock and flow maps, and Policy structure diagrams) for endogenous explanation of the complex problem. (Sterman 2000: pp ). 3. Formulation of a Model Until now, the reasoning base of SD has been established, and to continue, modelers should try to complete designed mental models via mathematization of variable relations and simulation facilities on the behavior of the proposed structure system in various scenarios. This step helps the modeler and clients to understand the flaws of proposed behaviour structure and see their endogenous explanation profoundly and insightfully. 4. Testing Like all other models, system dynamics functions would be valuable if it is validated correctly. Testing SD models is a continuous process which should begin in the early stages of modelling for building confidence in models (Barlas and Carpenter 1990). In summary, it could be said that there are some structural and behavioral experiments that modellers set to test the model. Absolutely, this process will be different, depending on the selected validation approach. 5. Policy Design and Evaluation From this step, the modeler tries to go far from the current situation by scenarios and policies that include the creation of entirely new strategies, structures, and decision rules. In policy making, the robustness of policies and their sensitivity to uncertainties in model parameters and structure and mutual effects of alternative policies must be assessed to propose best ways for improvements (Sterman 2000: pp ). 3 System Dynamics and Theory of Knowledge Epistemology, or its contemporary equivalent theory of knowledge, includes the most pivotal and central philosophical issues that are concerned with the nature and the scope (limitations) of knowledge. Much of the debate in this field has focused on analyzing the nature of knowledge and how it relates to connected notions such as truth, belief, and justification. It also deals with the mechanisms of producing knowledge, as well as skepticism about different knowledge claims. Epistemologists have distinguished some species of knowledge, including: propositional knowledge (that something is so), non-propositional knowledge of something (for instance, knowledge by acquaintance, or

5 by direct awareness), empirical (a posteriori) propositional knowledge, non-empirical (a priori) propositional knowledge, and knowledge of how to do something. Despite controversies over distinctions between such species (namely, the relations and the viabilities among them) (Moser and VanderNat 2002: p. 3), it could be said that the epistemologist s primary interest is propositional knowledge (or knowledge of truths) (Klein 1998: p. 2005). The generally accepted meaning of propositional knowledge (Hereafter knowledge) among Western epistemologists, from Plato to the midst of twentieth century, was justified true belief, which is, sometimes, called the standard analysis of knowledge. In epistemological analysis of SD modeling, we could find the tracks of all aforementioned knowledge in different phases of the modeling process. For example, the sort of knowledge of the problem owners, used in the process of modeling especially in the problem articulation, could be taken into account as knowledge of things, knowledge of that and knowledge by acquaintance. On the other hand, the knowledge proposed for improving actions and decision is a kind of knowing-how. According to system dynamicists, SD models are as the reflections of our beliefs about the networks of causes and effects, and each model offers some knowledge around a problem; in other words, SD modelers look for the justification of beliefs in variables about a specific problem to reach knowledge with some levels of truth.12 Therefore, it could be said that although all types of knowledge elicited or obtained in the process of SD modeling are important, the explanation of propositional knowledge in SD models needs about more analysis that could bring more distinctive theoretical and practical implications for system dynamicists and epistemologists. In continue, we will concentrate on the mechanism of obtaining this type of knowledge according to the SD modeling process. The rest of the paper proceeds to explore the concepts, namely, justification structure and truth for rewording system dynamics modeling process in term of theory of knowledge, according to discovered similarities.13 4 Justification In epistemologists view, knowledge is related to how our beliefs are possible or how they are formed. In response to this question, the theories of justification14 comes to the debates for analysis of knowledge. These theories could be classified in the following categories according to the structure and the content of knowledge they propose: 1. Based on structure, there are three distinct justification theories: a) Foundationalism Aristotle, St. Thomas Aquinas and Rene Descartes are classical foundationalists. According to the classical foundationalist philosophers, we have infallible, direct, self-evident and incorrigible knowledge of the first principles or basic statements from which other knowledge is elicited from them.15 You can see a schematic view of this kind of structure above (Fig. 2): Figure2 As it shows, through taking infallible principles, which are comprised of the first principles or intuitions, we can build a strong structure of knowledge. Despite some of the structural benefits of foundationalism, some problems challenge this structure, such as the short contents of knowledge that can be produced by infallible statements. This point can lead to the skepticism, which may cause serious problems for the house of knowledge. It is worth mentioning that foundational processes work on logical, perceptual, memorial, and testimonial beliefs (Pojman 2001: p. 127).

6 b) Coherentism Coherence theories of truth state that truth resides in the Absolute System of Knowledge. This kind of justification structure predominates in theoretical beliefs such as natural, religious, political, social, economic, and metaphysical beliefs (Pojman 2001: p. 127). To explain it, two most basic theories of coherentism as linear and holistic will be detailed. Based on linear16 coherentism, a belief (B1) gets its justification from belief B2, which, in turn, gets its support from other beliefs, namely, B3, B4,..., B,,, which, itself, will be justified by B1. It is clear that the regress problem and the circular scheme remain in this structure because none of them has any positive epistemic warrants (Fig. 3). Figure3 Considering holistic coherentism, one belief does not receive its positive epistemic status only from other beliefs; rather its status is related to the role it plays in a total system of beliefs. Suppose a three-dimensional network in which each belief is a connecting node, held together by other beliefs. As opposed to foundationalism, justification and protecting beliefs are multi-sided and concurrent, and each belief may be supported by many beliefs, not just one belief occurring in the linear structure. In the following, you can see the symbol of this kind of justification structure. As it is clear in the graph, a variable such as A1 is justified by A6, A7, A2 and the A1, itself, justify A3, A4 (Fig. 4). Figure 4 Against the foundationalism, in which the process of eliciting knowledge from the principles is obvious, in this structure, a question that may be asked is whether or not coherence is a necessary and sufficient condition for justification. In literature, some classical criticisms have been offered against the sufficient condition, the most important of which are indicated in the following17: 1. The Alternative System Objection This problem is related to the truth connection of coherent systems. Coherence, according to Rescher (1973), pays attention solely to the strictly internal relationships of implication that obtain within a set of mutually cohering propositions. So, it is problematic, to say the least, to show that a relationship obtains between this feature of such propositions and their actual truth status. Therefore, there is no criterion for preferring alternative systems of beliefs that are coherent. 2. The Input Objection This objection which is, also, called the isolation objection said that coherentism, by itself, does not have any criterion for relating it to the external world. Laurence BonJour (1985: p. 108), clearly, formulates this objection as follows: Coherence is purely a matter of the internal relations between the components of the belief system; it depends in no way on any sort of relation between the system of beliefs and anything external to that system. Hence if, as coherence theory claims, coherence is the sole basis for empirical justification, it follows that a system of empirical beliefs might be adequately justified, indeed might constitute empirical knowledge, in spite of being utterly out of contact with the world that it purports to describe. Nothing about any

7 requirement of coherence dictates that a coherent system of beliefs need receive any sort of input from the world or be in any way causally influenced by the world. 3. The Infinite Regress Objection The summary of this critique is that in order to justify one belief, it should be justified to person S and its consistency should be surveyed with all the beliefs of person S, and this survey will lead to the infinite regress objections as follows: A: S is my belief set and B1 is coherent with S. A1: S is my belief set and A is coherent with S. A2: S is my belief set and A1 is coherent with S. A3:... An:... In response to these objections, coherentists proposed some points, as mentioned in the following: For the first objections, it is said that, although we cannot prefer between two different coherent systems, we can find out which of them is closer to the truth. Of course, this reasoning is a kind of circular one, since for realizing more coherentism, they use reasoning and vice versa. Also, there are some answers to the lack of input one of which is Bonjure s statement: If Finally, in response to the third objection, it is said this confuses the process of discovering whether a belief is justified or not, with the status of actually being justified (Pojman 2001: pp ). In view of these criticisms and such responses from coherentists and also short contents of knowledge in foundationalism, many philosophers give up pure foundationalism and coherentism and choose moderate positions. As a result, it is rare to find some philosophers located in the extreme of this spectrum Based on the content, there are two distinct theories of justification: internalism and externalism. a) Internalism From the point of view of internalists, justification is relied upon as the essence of internal beliefs and the intuitive relationship of beliefs, or a combination of the essential and structural characteristics of beliefs and non-beliefs internal processes. Of course, internal justification has an epistemological meaning, namely, epistemic access that means the falsehood and truth of beliefs are intuitive, and are only affected by the structural characteristics of beliefs, their internal relations, and mental and cognitive states. In this category, we could see foundherentism and classic foundationalism and coherentism and later on, we would explain foundherentism: a-1) Foundherentism: Foundationalism and coherentism are not exhaustive of possible structures of theory of epistemic justification. In Evidence and Inquiry Haack argues for a theory of the justification of empirical beliefs, namely Foundherentism, which includes elements from both foundationalist and coherentist intuitions. According to this theory, in order for experience to play a role in justification (that is also, ignored in

8 coherentism), it is not required that the basic belief or beliefs be completely justified independently of any further belief (Haack 1993:p. 43.). All that is required is that the basic belief or beliefs acquire partial justification via experience and that this partial justification occurs independently of the support of any other beliefs (Clune 1997: 461). It should be noted that Hack uses the phrase of truth indicative against truth conducive for her theory of justification, which insures that the subject is ultimately justified in virtue of the justifiers experiences (as an ultimate source of empirical information available to him (Clune 1997: p. 218). b) Externalism An externalist sees the justification dependent on external and objective criteria. In the following, we will review theories, which are based on externalist view: b-1) reliabilism The most prominent form of externalism is reliabilism, which, in its most common form, holds that a belief is justified if and only if it was produced and or sustained by a reliable belief-forming process (Goldman 1986). b-2) Naturalism of Quinn Most of today s naturalized epistemology originates from Quine s (1969) Epistemology Naturalized. He treats the object of epistemology as a natural phenomenon, and calls for an experimental method in the study of epistemology to provide an account of a certain natural phenomenon, namely, knowledge itself. (Kornblith 1999: p. 161). Naturalist and non-naturalist divide over whether or not continuity between epistemology and science exists. Naturalists differ among themselves over what form this continuity should take. (Maffie 1990: p. 281) b-3) Warrant of Planting After proposing Gettier counterexample (Gettier 1963) against Justified True Belief, Alvin Plantinga s version of externalism is offered as an account of warrant, defined as whatever it is that distinguishes knowledge from mere true belief. He rejects the term justification as a label for this quality, on the ground that it is biased in favor of internalist conceptions of what is required for knowledge Besides these theories, there are two more approaches: a) Contextualism According to DeRose (1999: p.187), Contextualism refers to the position that the truth-conditions of knowledge ascribing and knowledge denying sentences... vary in certain ways according to the context in which they are uttered. 21 b) Pragmatism In summary, pragmatism implies that a proposition is acceptable to us if and only if it is useful to us besides its approved truth, that is, it is useful to us to accept the proposition.

9 4.1 Justification in System Dynamics With respect to the role of mental models in the modeling process and the structure of causal loop diagrams and its direct relation to our beliefs, it will be clear that the dominant structure in these models is coherentism. If, in dynamic hypotheses of SD models, in which there are no stock and flow variables, and all variables are considered as beliefs (experimental or logical), we can see linear coherentism among causes and effects of closed loops. For a better and complete understanding, let us review the famous market growth of Forrester (Forrester 1978). This model is illustrating causal loops, which show a set of relationships often encountered in the growth of a new product. In the phase of problem articulation, the modelers reached a behavior reference mode that reflected a fall in the market growth after exponential mode. Then, the dynamic hypotheses were constructed in term of causal loop diagrams one of which is as following. This is attained by mapping the mental models of problem owners, which is made of a set of related beliefs that produce knowledge around the problem. Justification of beliefs in the sales growing loop is depicted in the Table 1, as can be seen, i.e. belief in increase (or decrease) of budget as a cause variable Justifies a belief in increase (or decrease) of sales budget as an effect variable. Table 1 It is clear that beliefs-protecting in this loop is similar to the justification structure of linear coherentism (Fig. 5). Figure5 However, justification structure in SD models does not finish at this point and the modeler (justifier) combines these loops and creates a system and network of coherent beliefs, which is likened to the holistic coherentism. As can be seen in the following model, beliefs justification and protection is multi-sided and each of them is justified by some belief. As an example, the order backlog variable which is protected by just one belief in the last model, is protected by variables such as sales force, sales effectiveness and support variables such as revenue and the quantity of desired production (Fig. 6). Of course, you can find larger and more complicated SD models in the papers and projects where multi-sidedness of their variables is more evident. Figure6 In the following, implications of regarding SD models as justification structure through the comparisons of coherence theories criticisms in SD models will be discussed: 1. The alternative system objection in SD models occurs in a way that we cannot easily prefer between two models that survey the dynamics of a specific problem and have equal structural coherentism. Formal and informal processes of validation in SD models seek to decrease this problem, but it seems that because of the kind of SD s justification structure, these models will suffer from this challenge forever. 2. The input problem in SD models is not relevant because of their empirical beliefs. In reality, each belief of the models (causal variables) is elicited from the experiment and the sense that is according to the

10 real world (although we can find some instances of lack of firm relations between beliefs). Nevertheless, there are some models, which are designed based on the subjective reality. In this manner, they have some challenges from this critique The infinite regress objections are also not relevant to the SD s holistic coherentism structure, since, before the creation of the coherent network of beliefs, some variables have been chosen and the consistency of the selected and relevant beliefs will be tried to survey. It is worthy to note that identifying the model s boundary and the endogenous, exogenous and excluded variables, is the first and key stage of mapping the system structure that could remove the mentioned critique. Now, the insufficiency of foundationalism structure is also apparent because following this structure in social systems, we couldn t build wide house of knowledge and skepticism is a main threat. All in all, for moderating these challenges in SD models, a combination of coherentism and foundationalism for the SD model s justification is suggested (like what Hack proposed as foundherentism, especially for resolving the first problem). As a result, (in addition to the dominant linear and holistic coherentism), some initial and inevitable principles based on past experience of best practitioners will be accepted as foundations, as a kind of archetypes or even causal relations of beliefs (in variables). These foundations, considering the high development of SD, could be prepared in cooperation with the specialists of SD and the specialists of socio-economic fields as a SD reference handbook. In the third step of modeling process of SD, the modelers formulated the causal relations among variables, quantitatively, which is beyond our discussions. We will proceed to the forth step, validation, according to the truth concept in epistemology and give up the policy making for future research, since it is a kind of prescriptive knowledge. 5 Truth In the tripartite analysis of knowledge, truth is one of the necessary conditions of the beliefs in order to count them as knowledge. Examining beliefs and identifying truth or untruth of beliefs is one of the longlasting concerns of epistemologists. It should be noted that there are differences between the nature and criteria of truth, while most challenges are related to the criteria of truth. There are four important truth theories (correspondence theory, coherence theory, pragmatic theory, and relativist theory). In the following, we will explain these theories and their criticisms, briefly, and finally, truth theories in SD models, which are related to the validation debates, will be discussed. 5.1 The Correspondence Theory of Truth This interpretation is, likely, to be as one of the oldest general theories about truth and of course one of the most well-known theories that was accepted among Western and Eastern philosophers. In this view, truth consists in its agreement with (or correspondence to) reality. On the other hand, the correspondence theory captures our Common-sense intuition that truth depends on something objective (or mind-independent) in the world that makes it true (Pojman 2001: p. 5). This theory has attracted many proponents but, traditionally, some kinds of criticisms have been leveled at correspondence theories, primarily that all of them are the starting points of the tendency to create other

11 truth theories. Most of the objections to these theories are related to the ambiguities and difficulties occurring in confirming and verifying beliefs regarding facts. 5.2 Coherence Theory The term coherence in the phrase coherence theory of truth has never been very precisely defined. The most we can say by way of a general definition is that a set of two or more beliefs are said to cohere if they fit together or agree with one another (Kirkham 1998a). Different versions have been proposed in regard to coherence theory; one of the most important belonged to the idealists that emphasized on subjectivity rather than objectivity in discovering truth of beliefs. This theory has two principles (Pojman 2001: p. 7): 1. The doctrine of internal relation 2. The doctrine of the degrees of truth According to the doctrine of internal relations, any subset of a structure has the necessary consistency with the other subsets of that structure, and the result will identify the structure s truth. The doctrine of the degrees of truth says that if we suppose pure truth as a whole, it has never been attained (because of human cognitive limitations) and just comes in degrees. In this manner, structures will be contextually true and not purely true. The most common objection to the coherence theories is that these theories make truth as a matter of a truth bearer s relations to other truth bearers rather than its relations to reality and the conditions they place on truth are too weak (Kirkham 1998a). 5.3 The Pragmatic Theory of Truth Under the name of pragmatic theory, we can see two distinctive types of truth. First, the consensus theory of C.S. Peirce that says a true proposition is one that would be endorsed unanimously by all persons who had had sufficient relevant experiences to judge it. Second, there is the instrumentalist theory associated with William James, John Dewey, and F.C.S. Schiller, according to which truth of a belief is identified by its usefulness. A proposition counts as true if and only if behavior based on a belief in the proposition leads, in the long run and all things considered, to beneficial results for the believers (Kirkham 1998b). The main criticisms of the pragmatic theory of truth are related to ambiguity and difficulty in understanding the truth, a point such that, in Pojman s opinion, these criteria may lead to a form of relativist theory of truth. On the other hand, reaching consensus as a criterion for truth cannot be convincing, since there are many examples that are true at the time of proposing but there is no agreement on them. 5.4 The Relativist Theory of Truth This theory seems to be new, but it originated in Protagoras (c.490-c.420 BC) famous formula: Man is the measure of all things, of things that are, that they are, and of things that are not, that they are not. 23 There are many interpretations of this statement but, in one related to relativist theory of truth, it is said that personal view is the criterion of falsehood or truth of a proposition.

12 In other words, according to relativist doctrine, truth always really means true-by-the standards-of-x, where X is some individual or group. Such a theory allows that truth may differ from group to group or from person to person, since social conditions and individual psychology may affect the satisfaction to be had from a given belief (Craig 1998). Proponents of this kind of theory face various problems. Lack of conceptual scheme, paradox in some statements, and anarchism (which is the result of this theory) are just some of important objections to the relativist theory of truth (Harre and Krausz 1996: pp ) (Table 2). Table Truth Theories in Validating System Dynamics Models As it is mentioned at the start of this discussion, truth theories are related to the validation phase of SD modeling. All the philosophies which authors indicated about SD models have some key points for the validation phase. For example, when Barlas and Carpenter addressed a relativist philosophy of science for SD models, their idea is that knowledge is confirmed and accepted through social interaction and it is relative to a context and so on. However, our idea is a straight-forward one because we count all possible theories and the one, which is remained, is comparing and analyzing them according to SD models. Before entering the discussion, it is necessary to note that truth-testing in SD models is procedural and, like quality assurance, is gradual and reciprocal from the first to the last stages of the modeling process or even after that. This issue makes truthtesting a dynamic process toward gradually attracting confidence and validation between modelers and clients. The dynamic process of truth making as an outstanding property of SD modeling could be explained in term of the Bayesian epistemology24 (such as conditionalization principle). In this regard, knowledge refinement after entering new information in a causal Bayesian learning network will be similar to the iterative process of constructing dynamic hypothesis for a complex problem by beliefs in variables changes25. However, by looking at SD models and the quotes of the pioneers of the field (despite of lack of consensus in the issue), it could be said that the nature of truth in SD models is correspondence theory.26 Based on this theory, all variables of models can and should have quantitative and conceptual correspondence in the real system. However, the claim of correspondence theory regarding the nature of truth in applied situations reveals the existing difficulties in this theory, especially in socio-economic fields; therefore, system dynamicists emphasized other theories as a criterion of truth. The pragmatism, relativism and coherence theories (and even correspondence theory) are the most indicated theories that are counted as truth theory in SD. Considering SD s justification structure (coherentism), and the pivotal role of mental models in the modeling process, it seems that the coherency can be the most suitable standard in the validation of SD models. Thus, for SD, the truth has collective not individual essence, so that it will be earned through the inter-subjective space and challenges among mental models. As such, two doctrines of internal relations and degrees of truth are dominant in SD models. First, each of these beliefs (in variables) should have necessary coexistence with other beliefs (in variables) and, as a part, have consistency with the total network or causal systems (the doctrine of internal relations). Second, attributing truth (validation) to the models is not a complete truth, but it is contextually true. It is clear that

13 by accepting this theory, the problem of preference among alternative models remains. This is because there may be two or more different causal systems for one problem, 27 which have equal internal relation or intersubjectivity agreement, but refer to different degrees of truth. Under these circumstances, we can set criteria (such as the degree of goal attainment) as a preference criterion of models or belief networks (systems) to reduce this problem. It is proponent to the statement of Forrester (1961: p. 129) that said: knowledge of all forms can be brought to bear on forming an opinion of whether or not a model is suitable to its particular purpose. However, as it is mentioned in justification structures, this type of justification has unique merits (such as holism, two sided relations, etc.); nevertheless it always suffers from the lack of a stable basis or criteria for reasoning. Therefore, a strong and flexible basis should be established in order to be effective in system s truth and preference. Based on aforementioned explanations about truth theories, in the following, the feasibilities of other theories as a validation theory in SD are discussed. Pragmatism theory cannot be the correct one, due to the abstract focus on results and the lack of commitment in the SD s justification structure. However, efficiency and usefulness can be counted as modeling goals besides the criteria of coherence theory. Relativism theory, because of lack of coherent conceptual scheme and anarchism, cannot be very useful for SD, and using this theory may lead SD to the mentioned outcomes of relativism thought. Finally, Correspondence theory, also, does not have sufficient criteria in identifying truth, since there are difficulties in adopting variables with their counterparts in reality as Forrester (1971: pp. 3 4) also emphasized that we can never prove that any model is an exact representation of reality. All in all, we can conclude that the essence of truth in SD is correspondence theory, while the criterion of truth is coherency in addition to the level of model s goal attainment that could include efficiency and usefulness of the model. Of course, according to these discussions, the goal should be set on attaining the Correspondence concept and improving the degree of truth. Thus, validation includes Correspondence theory (even partially) between real system components and belief systems (Table 3). 6 Conclusion Table 3 The aim of this paper was to explore the essence of knowledge and the knowledge production mechanism of SD models. In other words, in this article, we have tried to translate the principles of the field in terms of epistemological expressions and concepts; subsequently we have revealed the analysis s implications, limitations and suggested ways for reducing possible criticisms. However, this kind of research may be considered a new kind of philosophical exploration of SD since it started from the structure of models, which is the center of the debates in the field and then by concepts such as truth extended the border of discussion. What we have argued here is that there are strange similarities between justification structures and SD models that will be followed by some implications. Maybe the most important of them is that we cannot easily prefer among SD models (a strong problem in validation debates); however, in this paper we have suggested some ways for decreasing this inevitable structure s effect. However, our claim is not that we have detailed all the facets of this conformity completely; we have merely scratched the surface of theory of knowledge in SD models, and more work remains to be done about these

14 concepts, especially where their practical implication is an issue. On the other hand, exploring the peculiarities of SD in contrast with other ways of obtaining knowledge could be one of these niches, which remains for interested researchers. All told, we hope that this paper can help to reinvigorate the advocates of methodological issues of SD by opening new and firm arenas for more discussions. Endnotes: 1 A useful definition of falsificationism appears in the Dictionary of Jargon (Green 1987, London: Routledge and Kegan Paul): (Sociology) a doctrine which claims that scientific advance can only come through testing and falsifying hypotheses, which are then replaced by new hypotheses to be tested and falsified in their turn; one can only falsify, never ultimate verify. 2 Critical rationalists believe scientific theories, and any other claims to knowledge, can and should be rationally taken to task, and (if they have empirical content) can and should be subjected to tests which may falsify them. 3 By relativism, these authors mean some sort of relativism, in which the epistemic justification of models is relative to the interests and purposes of the participants in the model building process, not an extreme relativism that assumes knowledge is an objective representation of reality and that theory justification can be an objective, formal process. 4 For IR, it does not make much sense to suppose a passive reality, one which is ready-made and structured independently for our knowledge and actions, from which our models can at best be copies. The same is true of SD, where knowledge from mental models is such an important part of the building and justification of the final SD models. (Vazquez et al. 1996). 5 A mental model of a dynamic system is a relatively enduring and accessible, but limited, internal conceptual representation of an external system whose structure maintains the perceived structure of that system. Doyle and Ford (1998). 6 According to Searle, social and institutional phenomena are construed through the recursive iteration of three basic mechanisms: collective intentionality, the assignment of functions and systems of constitutive rules. These phenomena are epistemologically objective but ontologically subjective and, in general, we only construe them implicitly. (Vazquez and Liz 2007: p. 18). 7 According to Brandom, logic does not describe or represent any ideal realm. It has an expressive role linked to what is implicit in our inferential practices. (Vazquez and Liz 2007: p. 18). 8 Refer to: Olaya 2009: pp For more details about the kinds of SD models, refer to: (Barlas, 1996, pp ). 10 For review of The SD modeling process across the classic literature, see: Luna-Reyes and Andersen (2003). 11 A set of graphs and other descriptive data showing the development of the problem overtime.

15 12 Causes and Effects in SD models are variables, and a causal connection is the relation between them. In rewording SD models to theory of knowledge expressions, as it will be explained, beliefs in variables is intended, and subsequently, we count causal connection among variables as justification (Although justification has a meaning beyond causal connection, but it could be said that any causal connection of beliefs of knowledge is a kind of justification). 13 We should declare a point of our epistemological analysis of SD models in order to decrease, somewhat, misunderstanding about the precise similarities of justification and SD modeling. The similarity which we use in our paper may be construed as a kind of argument from analogy, whereby perceived similarities are used as a basis to infer some further similarity that has yet to be observed. Of course, the argument doesn t state that the two things are identical (or interchangeable), but it says that they are only similar. The argument may provide us with good evidence for the conclusion and explanation, but the conclusion does not follow as a matter of logical necessity. Accordingly, we don t have authority to use this argument as a pretext for complete substitution of expressions and concepts as they are originated in different systems and their implications are different, in essence. Let us to clarify this position by an example from mathematics: equivalence of numbers, say number 2, in two-dimensional space is the size of a line equal to 2, while we couldn t never substitute 2 in two dimensional space with a line and use them instead of each others. 14 It should be noted that here, we mean a wide meaning of justification, which covers descriptive and prescriptive ones. 15 In a classification, beliefs are divided into basic and non-basic beliefs. A belief is basic for a person at a time just in case it is not based on any other belief for the person at that time. Basic beliefs are at the foundations of a person s doxastic structure (Quinn 2002: p. 526). Or in other words, beliefs which are fully justified independently of the support of any other beliefs. 16 This linearity is not against the nonlinear attribute of complex problems. Otherwise, it is defined as circular structure, which is named linear against holism in coherentism debate in theory of knowledge. 17 There are some other problems and challenges regarding the necessity of coherentism in belief justification that say, under special conditions, we may have some justified beliefs that are incoherent with other beliefs. The lottery paradox is an example of this kind. Assume a fair lottery with a thousand tickets in it. Each ticket is so unlikely to win that we are justified in believing that it will lose. So, we can infer that no ticket will win. However, we know that some ticket will win. 18 Explanation of this answer is out of the scope of this paper, for more information, refer to: (Pojman 2001). 19 For example, Audi (1993) and Alston (1976) have tried to establish an approach, namely, moderate foundationalism. 20 Plantinga offers the summary statement of his account of warrant in the following reference: Plantinga, Warrant and Proper Function [hereafter WPF] (New York: Oxford University Press, 1993, pp ). 21 see also DeRose 1992: p. 914.

16 22 Holon dynamics and Modeling as radical learning are two types of SD models that could be covered in this critique: Holon Dynamics (or HD) is an envisaged form of practice grounded in the interpretivist paradigm. With HD the notion of model building as a social process is embraced and models are nominalist representations, useful devices which help human agents to create their social worlds via debate and the construction of shared meaning. (Lane 1999: p. 517). Modeling as Radical Learning implies the use of SD modeling to further communicative competence within groups. (Lane 1999: p. 518). 23 Protagoras (c.490-c.420 BC) Fragments, in H. Diels and W. Kranz (eds) Die Fragmente der Vorsokratiker (Fragments of the Presocratics), Berlin: Weidmann, 7th ed, 1954, vol. 2, pp (The standard collection of the ancient sources both fragments and testimonia, the latter designated with A, includes Greek texts of the fragments with translations in German.) Routledge Encyclopedia of Philosophy, Traditional epistemology, with its focus on the analysis of knowledge, is relatively silent about the questions of belief dynamics. If there is talk about belief change, it is generally assumed that it takes place on the basis of learned evidence that is certain. Traditional epistemology shares this assumption with logical theories of belief revision such as the AGM theory (Gardenfors and Rott 1995). However, Jeffrey taught us that learning often does not come in the form of certainties. To address these cases of learning and belief change, philosophers as well as researchers in artificial intelligence have formulated new updating rules (such as Jeffrey conditionalization) and developed powerful tools such as the theory of Bayesian networks (Neapolitan 2003). (Dancy et al. 2010: pp ). 25 There are three different belief changes as: belief expansion (when we learn something new), belief revision (in the light of evidence that contradicts what we had earlier mistakenly accepted) and belief contraction (when we discover that the reasons for some of our beliefs are invalid) that is usual in modifying the primary hypothesis about a complex problem in SD. 26 Forrester (1961: p. 60) argues in this respect that [a]ll constants and variables of [a system dynamics] model can and should be counterparts of corresponding quantities and concepts in the actual system. Also, Sterman (2000, 517) said that: All variables and relationships should have real world counterparts and meaning. 27 There is no privileged single model for a complex behavior and we could discover structures, which are generating the same behavior with no primary evidence for preferring among them (Olaya 2009: 9067).