COMPRAM Method

In chapter two we discussed some of the differences between the problems investigated by cognitive psychology and the subject of our study. In this chapter, some similarities between the problems investigated by cognitive psychology and complex interdisciplinary societal problems will be discussed. We discuss the process of human problem handling, and in the process we will answer the remaining part of research question 1b:

1b .....what are the similarities (...between the problems dealt with by cognitive psychology and complex interdisciplinary societal problems...) that are relevant for analyzing and defining complex interdisciplinary societal problems (expectation 1)?

The chapter begins with a description of the phases distinguished in the two sub-cycles of the problem handling process. After discussing the phases in detail in order to emphasize certain similarities between the problems cognitive psychology deals with and complex interdisciplinary societal problems, we shall, in discussing scenarios, discuss research question 2:

what is the relation between a scenario representing a complex interdisciplinary societal problem and reality?

The problem handling phases identified by other researchers will then be discussed, followed by a discussion of rationality in problem handling, and the knowledge and data needed for handling complex interdisciplinary societal problems. The discussion ends with some remarks about problem handling techniques.

Handling a problem is to act with the intention of changing the problem in the hope this will increase knowledge about the problem which, in turn, may contribute to reduce the problem. The problem handling process can be roughly divided into two sub-cycles. The aim of the first sub-cycle is to obtain a good view of the problem: to describe the problem, identify the phenomena which are involved and the relations between the phenomena. The first sub-cycle is complete with the formulation of the conceptual model of the problem, by which the problem is defined. In this sub-cycle the emphasis is on analyzing the problem, which can be done by thinking, reading, discussing and asking questions, and may include all kinds of observation, different methods of interviewing and all kinds of data searches. The aim of the second sub-cycle is to change the problem in the hope to reduce the discrepancy between the actual or future state of the problem and the desired state, at least in the view of the problem handler(s). In the second sub-cycle, in addition to thinking, reading, discussing and asking questions, the emphasis is on describing, performing, implementing and evaluating interventions. The second sub-cycle is a mixed process of thinking and acting. The second sub-cycle starts with is initiated by fleshing out the conceptual model with empirical data, as a consequence of which the conceptual model becomes an empirical model. On the basis of on the empirical model of the problem, interventions can then be proposed, selected and evaluated. After this, interventions can be carried out and their effect on the problem evaluated.

The two sub-cycles in the problem handling process can be divided into phases. Although this division of a continuous process into phases is more or less arbitrary, nevertheless such phases do usefully provide check points to facilitate and guide the process of problem handling. The following account of the process of problem handling is based on an ideal process. In reality the process will be more diffuse, and more disorganized than this theoretical description[1], which suggests a linear process. Although in reality the process is necessarily to some degree a linear one in that a person proceeds from a state of knowing something to a state of knowing more, beginning with phase one and ending with the last phase, the phases are part of an iterative process which reverts to former phases in the problem handling process whenever this is necessary[2]. However, since it is unwise to skip phases in the first walk through the phases, to that extent the process is linear. In each phase one or sometimes two activities can be distinguished. When there are two activities they are in close interaction with each other and are in themselves iterative.

The aim of the first sub-cycle is to define the problem. Although the process from having a (vague) mental idea of a problem to forming a conceptual model of a problem is a gradual one, several phases can be distinguished. We will discuss each phase in detail.

Phase 1.1 Awareness of the problem and forming a (vague) mental idea of the problem

The first phase that can be distinguished is the phase of awareness of the problem. This might be the result of a direct conscious act based on a suspicion that there could be a problem in a certain area, or it might be the result of a vague unconscious feeling that something is wrong or missing, or a feeling of anticipation, a feeling which gradually becomes more clear. There must be some sign that something is not right, that there is a problem, before the problem can begin to be dealt with. A problem is only a problem when it is recognized or experienced as such. Awareness begins with hearing about a problem or hearing about something and suspecting that there is a problem.

In describing the phases of problem handling many researchers skip this first phase of awareness of the problem, or only mention it briefly. One of the important differences between the phases we distinguish and the phases identified by many other researchers[3], is that they often begin with the phases after the problem has been noticed and presented as a problem. In this study we start at the beginning with the phase of becoming aware of a problem.

"Although there appears to be a growing interest in problem finding ......in the majority of studies, however, the traditional starting point for research is the given problem." (Brugman, 1991, p. 212, 213)

Brugman (1991) acknowledges that some researchers in the period 1926-1933 emphasized the phase preceding the 'solving of a problem', what we indicate in our model as (the first phase of) the first sub-cycle. Researchers like Dewey (1933), Claparède (1933) and Rossman (1931) paid attention to this phase. Dewey recognized two steps: a state of doubt, cognitive confusion, frustration or awareness of the problem and an attempt to identify the problem, involving a global awareness of the aims to be achieved and the gap to be bridged in the particular situation (Brugman, 1991). Wallas (1926) stated that no adequate response can be expected to any problem unless it is clearly formulated in advance.

Becoming aware of a problem is called 'problem finding' (Brugman, 1991). Problem finding is the discovery that there is a problem. Recent researchers with some interest in problem finding include Getzels & Csikszentmihalyi (1976), Perkins (1981), Dillon (1988), Gardner (1984) and Sternberg & Smith (1988). Getzels & Csikszentmihalyi (1976) studied the relation between problem finding, problem solving and artistic success.

The implication of the idea of problem finding is that the sooner one is aware of the problem, the sooner the problem can be anticipated and the greater the chance of intervening in order to prevent the problem from becoming too severe and causing too much trouble. Becoming aware of a problem in time can be very important, particularly for complex societal interdisciplinary problems, although the matter may be different at the level of individuals where, when problems are sometimes found for which there is no adequate intervention available, it may be that not knowing the problem in advance is better than knowing[4].

Hearing about a problem does not always lead to realizing that there is a problem. One can also interpret the news as just the information and leave it at that. In this case there is a problem, but one does not recognize it and as a consequence leaves the problem as it is.

Being aware does not imply that one starts handling the problem automatically. The problem can be neglected, or it can be decided not to handle the problem, or to circumvent the problem. However, when one decides to handle the problem, the problem has first to be defined.

Newell & Simon (1972) regard problem handling as an information processing and goal seeking activity. To them problem handling is a special way of information processing in which one tries to reach goal(s) without knowing directly how to do this[5].

Norman & Bobrow (1976) consider information processing as a linear process as follows:

Experimental psychological research often begins with this information processing model (Sanders, 1967; Norman, 1969; Broadbent, 1971). With Duyne (1983) we share the opinion that this model is too limited. Research by Selz (1922) and De Groot (1946/1965)[6], research in the field of language development (Lenneberg, 1967; Morgan, 1971; McNeill, 1970; Sinclair-de Zwart, 1973) and research on how people observe (Gibson, 1977) make it plausible to assume that a human being is not a passive receiver of information input, but collects information based on his or her own mental ideas. People have (commonsense) knowledge and use this knowledge to interpret new information. Spilich (1979), in his research on memory, states that based on commonsense knowledge, a person is able or unable to observe certain things (Spilich, Vesonder & Chiesi, 1979, p. 33). The idea that memory consists of pictures, 'mental images' (Paivio, 1969), has been doubted by Pylyshyn (1973) and Banks & Flora (1977). This idea of mental images has been superseded by the scheme theory of Bobrow & Norman (1975) and Neisser (1976). Memory functions far from perfectly, as is shown by research on memorizing chess positions (De Groot, 1965), buildings (Norman & Rumelhart, 1975), stories (Rumelhart, 1977) and the recognition of criminals and remembering of offences (Wagenaar in Wagenaar & Loftus, 1991). Rumelhart's research shows that no visual copy or literal report could be produced bu an observer, but only schemata that were meaningful for the observer. These schemata may be compared with a kind of 'map', a 'street map' or an 'internal map' that can be filled independently of the situation in which a person is located. Downs & Stea (1973) call this a cognitive map. For instance, when somebody has lost something in the house and wants to find it again, the whole scheme of the house does not have to be recalled, but only the sub-scheme of that part of the house where the lost object is supposed to be. A mental scheme can be written down in a report, a paper or in some notes. Miller (1960), among others, speaks of such a scheme as an acting plan that is the basis for concrete behavior. The term 'scheme' has already been used by Selz (1922) and Barlett (1932) in their research on problem solving.

A general definition of a scheme, according to Rumelhart (1977), is: a scheme is an abstract representation of a generic concept for an object, event or situation.

Rumelhart (1977) concludes from his research on remembering short stories, that the stories are reconstructed by the reader out of an 'abstract scheme' drawn from the main idea of the story according to the scheme used by the reader. Different mental schemata of a story can lead to departures from the original story. Spiro (1977) and Black, Turner & Bower (1979) show that these faults, omissions, additions and distortions are usually not coincidental. They result rather from the need to make the reconstruction fit (logically) and acceptable for the person. In experimental research it can be seen that information is processed to produce a piece consistent with a person's own scheme. This research was based on reflecting on a person's own information process (Markus, 1977).

Observation is also based on schemata. People are only able to observe something because they already have some schematic knowledge, some foreknowledge (Jorgenson, 1978). In most cases a person knows what to expect and the relevant knowledge schemata are activated before he or she looks at something (Duyne, 1983, p. 35). Selz (1922) talks about 'anticipation schemata', while Tsujimoto, Wilde & Robertson (1978) also confirm the relevance of the scheme theory to the observation and memorizing of social events. Thorndike (Thorndike & Hayes-Roth, 1979) too confirmed the plausibility of the scheme-theory with his experiments.

Kintsch & Van Dijk (1978) confirm the ideas of Rumelhart (1977) on remembering the story scheme and the details of the story. Some details of the story are remembered, others are not. In the author's view the memory is not a filing cabinet into which all kind of things are put, but should rather be considered as an active medium (Norman & Bobrow, 1976; Cofer, Chmielewksi & Brockway, 1976; Hertel & Ellis, 1979). Memory processes are not only active when a person searches for things from the past. Memory processes are not dissociated from the continuous stream of everyday activity, such as looking, listening and acting. Memory processes are always active.

For Bruner (1957) the interaction between information and changing the mental scheme is the most important aspect of human information processing.

Information is processed in the context of a certain expectation of what to see. This does not mean that a person can only see what is expected. When the external stimuli are strong enough to override the expectation, it is possible to overrule the inner scheme (Mynatt, Doherty & Tweney, 1977), and by means of confirming or changing schemata new information is processed. Neisser (1976) speaks about a cyclic process of information processing (see Duyne, 1983, p. 36).

New information becomes active in the field of old information, as is demonstrated in the reading of a story by Riskey (1979). In some sense the reader is already constructing his or her own story during the reading process, not simplt in a later reconstruction of the story (Kintsch, 1976). Misrepresentation can also arise when things are reconstructed by memory.

According to Duyne (1983) people continuously process information during problem handling[7]. A person cannot solve problems without processing information. Duyne points out that an important feature of verbal information in general is that this information can be interpreted in different ways, so that processing of verbal data as well as of written data has a certain range of freedom of interpretation. Duyne speaks about information-narrowing[8] when certain information is deliberately omitted from the mental scheme. On the other hand, information-extension[9] is used when a person adds something to the information.

Many terms have been proposed to characterize the intermediate step between the initial perception of a problem or an issue and the fully developed idea a person has in mind: 'frame' (Minsky, 1975), 'conceptual model' (Checkland, 1981; Bots, 1989) and 'mental model' (Johnson-Laird, 1980, 1983; Brewer, 1987; Boden, 1988, 1990). The term 'schema' (Brewer, 1987; Rumelhart & Norman, 1978; Rumelhart, 1984; Boonman & Kok, 1986) also closely approximates this idea, as also does the term 'mental' or 'cognitive map' (Sevón, 1984). Although these terms differ slightly from each other in their precise meaning, they all refer more or less to the same issue: the representation of something in a person's mind.

"A schema is the primary meaning and processing unit of the human information processing system. A schema is considered as active interrelated knowledge structures, actively engaged in the comprehension of arriving information, guiding the execution of processing operations. In general, a schema consists of a network of interrelations among its constituent parts, which themselves are other schemata." (Rumelhart & Norman, 1978, p. 41)

We suggest the use of two terms to indicate the representation of something in a person's mind: 'mental idea' and 'conceptual model'. Using two concepts enables us to make a distinction between a vague or general idea of something: the mental idea, and a more structured idea, more clearly defined idea about something: the conceptual model. These two terms are needed to describe the process of problem defining.

In this study we indicate the rather vague and diffuse idea of a problem, which is formed directly after becoming aware of a problem, as the mental idea of the problem, and we reserve the term 'conceptual model' for the definition of the problem. In a conceptual model the main phenomena and the relations between the phenomena that influence the problem are more clear and developed than in the mental idea. We are aware that the distinction between the two concepts is sometimes a gradual one and that the border between mental idea and conceptual model is not sharp. It is more or less a matter of subjective interpretation at what point a mental idea has developed into a conceptual model. The mental idea and the conceptual model are two extremes on the same continuum (see figure 2).

Exactly how the formation of a conceptual model from a mental idea takes place in a person's mind is not known. In terms of the schema theory, it involves filling in one or more variables of the scheme. However Boonman & Kok (1986, p. 30) state in criticism of the schema theory that the term 'schema' is not univocally defined and that it is not explained how cognitive processes activate and control schemata. The schema theory does not clearly explain the origin of knowledge either.

We limit ourselves to a description of the activities that lead from the mental idea to a conceptual model. On hearing about a problem for the first time, a person forms a mental idea of the problem. This idea consists often of a (vague) idea of the phenomena involved, how they are related, about the impact of the problem, about the causes of the problem, how these causes are related and about the effect of the problem. This mental idea can be very vague, or more clear, depending on the complexity of the problem. Sometimes the mental idea is not at all vague and there is a clear, intuitive picture of the problem[10] (Snoek, 1989, p. 450). Snoek discovered that experienced neurologists often diagnosed a disease on the basis of intuition. However, it may be that a person is not sure whether this intuitive idea is correct or not.

The mental idea of the problem will develop on hearing more about it. When a person finds the problem sufficiently relevant to devote some time to it, he or she can elaborate his or her knowledge about the problem by hearing, thinking, talking, reading and asking questions about it. This process can be an overtly conscious, half conscious, intuitive, explicit or implicit process or any combination of these[11]. Development of the idea can be assisted by data collecting based on experiments, measurements, observations, interviews and literature study. The search for data is selectively based on the mental idea a person already has of a problem. Also the questions that are asked and the answers attempted are based on the mental idea of the problem. It is an iterative process of hearing, thinking, talking, reading, asking questions, searching for data based on a continuously developing mental idea of the problem. This mental idea will not return to the same state, however, but will be reformulated at a new, higher, level where the mental idea of the problem becomes more specific. The mental idea becomes more precisely defined step by step adding more information and by rethinking the problem. On the basis of the renewed mental idea, new questions and data can be searched for (see figure 3).

A mental idea as such, is indifferent to whether it is a correct or incorrect idea of the problem. A mental idea can be wrong, partly right, or can be a correct representation of the problem. The mental idea can be a more or less correct representation of reality, or can have almost no relation with reality at all.

As the mental idea of the problem develops, the first hypotheses about what the problem looks like can be formulated. The ideas and hypotheses about the way the phenomena are related with each other can be based on well-known elaborated theories, on hypotheses, on assumptions, on experience, or on intuition or any combination of these. Intuition can be the start of a process of formulating hypotheses on the basis of a theory or the beginning of a theory on how the phenomena are related to each other. The ideas and the hypotheses have to be confronted with supportive as well as with refutative empirical data, in the course of which the mental idea of the problem can become more specific.

Phase 1.2 Extending the mental idea by hearing, thinking, reading, talking and asking question about the problem

When one is aware of the problem, one forms an idea about the problem. This can be a vague mental idea or it can be immedialtely as clear as the conceptual model. This will depend on the kind of problem. In the beginning one has a vague and sometimes even incorrect idea of many complex interdisciplinary societal problems. It is likely that after hearing of a problem and if one is interested in the problem, one will think, read, talk and ask questions about the problem, and this information will elaborate the mental idea of the problem. Thinking, reading and talking may be conducted at random, incidentally or structurally. It can also be based on specific questions and specific investigations for answers.

In collecting data, people tend to neglect these data that do not support the mental idea and to overestimate the value of data that support the mental idea. This is the conclusion of the problem handling analysis of domain related undefined problems in the medical field and in the field of law (Crombag, de Wijkerslooth & Cohen, 1977). Crombag (1978, 1984) indicates that medical diagnosis of family doctors is not the result of a long and difficult process of collecting data, thinking and deciding which diagnoses should be made. A very limited sum of possibilities appears at a very early stage in the thinking process. Elstein (Elstein, Schulman & Sprafka, 1978) have shown that physicians tend to gather information of symptoms that confirm their hypothesis, while they neglect non-confirming information, such as symptoms that conflict with the hypothesis. This is confirmed in the research of . The physician constructs three to five hypotheses based on only a few data about what might be wrong with the patient. The remainder of the thinking process is restricted to finding supporting data for one of the hypotheses. The premature hypotheses are based on a small set of data, which are easy to obtain; data such as the age, the gender of the patient, the complaints and what started the complaints. In medical diagnosis physicians do not think regressively. At a very early stage in the construction of the problem space, this space is restricted from more than a thousand diseases to three to five diseases.

The same research results were found in research on decision making by judges (Duyne, 1983). Judges, like physicians, tend to look for supporting data and neglect data that are in contradiction with their first hypothesis. Confronted with a problem, often rather quickly, a mental idea of the problem is constructed. This mental idea often consists of some hypotheses of 'what the problem is' and 'what caused the problem', and from that point on one looks for data that support the early ideas. Other data are not collected or are neglected (Wortman, 1966). Crombag (1984) states that this is related to the limited working memory[12] of human beings[13].

Snoek (1989) examined the diagnoses of physicians in the field of neurology, also demonstrating that diagnoses were generally made at a very early phase of the diagnostic process and that new data were mostly interpreted on the basis of their compatibility with this diagnosis. In this way Snoek confirmed the results of the research of Elstein, Crombag and the other researchers mentioned above.

Even in scientific research the researcher often tries to find data and facts that support her or his model instead of seriously looking at data and theories that might refute the first hypothesis (Gleick, 1987).

The search for data is influenced by the mental idea, which in turn is influenced by (new) data. On the basis of the mental idea one can consciously or unconsciously, form hypotheses about what the problem looks like. Hypotheses can be seen as temporary answers to questions. In an iterative process of linking the mental idea to data, intuitive moments, some irrational moments and analogous moments (see also Selz, 1922) can all play a part.

A problem can be represented by a model. A model represents the phenomena involved in the problem and their relations. There are various possible ways of representing a problem by a model: expressing the model in words, in a formula or by a graphical representation or a combination of these. In the rather diverse literature from different disciplines[14], which reflects on model building there are various definitions of 'model'. Some of the definitions in the literature are:

"A possible realization in which all valid sentences of a theory T are satisfied is called a model of T."

".....a model does not represent a theory itself, but the function derived from it by replacing the primitives by variables."

".....every model of the set of axioms alsosatisfies all the theorems that can be derived from the axioms.."

"..a model of the set of axioms of our theory will also be indicated as a model of the theory itself." (Tarski, 1953, p. 133)

"Scientific models ... are representations of states, objects, and events. They are idealized in the sense that they are less complicated than reality and hence easier to use for research purposes. These models are easier to manipulate and 'carry about' than the real thing. The simplicity of models, compared with reality, lies in the fact that only the relevant properties of reality are represented." (Ackoff, 1962, p. 108)

A model is a symbolic representation of the problem. The definitions of Tarski and Ackoff make clear that the essence of a model is the representation of an issue. The definitions of Tarski and Ackoff might suggest that there is a more or less 'absolute' model of something that can be constructed, apart from the subject who constructs and apart from the subject who uses the model.

Apostel (1960) states that any subject using a system A that is neither directly nor indirectly interacting with a system B, in order to obtain information about the system B, is using A as a model for B.

Apostel thus includes in his definition a subject. The term 'subject' in this definition can represent one or more persons. The term refers to epistemology: how do persons acquire knowledge and about what.

"There is a reality of which one can acquire knowledge. In this reality one can make a distinction between subjects that acquire knowledge about objects and objects of which subjects acquire knowledge. This distinction is relative for social sciences, because subjects can be considered also as objects and objects as subjects." (Van Dijkum, 1992, p. 2)[15]

There is a relation between the objects that have been portrayed and the representation of the object itself. A phenomenon is, epistemologically seen, a coherent collection of objects.

What is relevant in a model of a problem and what is a relevant detail of the problem depends on the kind of problem, on the purpose of a model, and on the actual moment in the problem handling process. There are no objective criteria for deciding what is relevant to the issue one wants to express. This can only be based on a (inter-)subjective opinion of the person(s) constructing and/or uses the model[16]. In forming our own definition of a model of a problem we can start with:

A model is a representation of a problem when it expresses those phenomena and relations between the phenomena that the subject, who formulates the model, considers relevant

The above cited definition of models of Tarski, Ackoff and the description of Apostel, lack any mention of a goal. A model is related to a goal because one should selectively represent only those parts of the problem one is interested in, in the light of the goal for which the model should be used. That is, a model has a purpose. This is expressed in the (working) definition of Van Dijkum (1992, p. 3):

The goal of the model is to make it easier to reflect on (a part of) the problem.

A model is, by definition, not the same as reality, a model is an abstraction and a deduction from reality.

A model is a goal related image of a problem in reality, consisting of phenomena and relations between phenomena that the subject, who formulates the problem, considers relevant

A model is also subjective in the way that it can be subjectively evaluated. Whether a model is a correct or incorrect, a complete or incomplete representation of a problem, a matter of the inter- and/or intra-subjective opinion of the subject(s) that create(s) or use(s) the model.

A model of a problem can comprise known phenomena, relations that are certain and some possible relations between the phenomena. A model can be based on theory(ies), assumption(s), hypothese(s), experience(s), intuitions[18] or any combination of these. These theoretical ideas determine the way reality is viewed. A model can be used in different ways depending on the theory[19], and the theoretical ideas, the goal of the problem handling process and the ideas a person has about model building. It depends on the kind of problem to what extent the problem can be represented by a model.

A model can be used to form an idea of (a part of) the problem. A model is constructed in order to make it easier to find answers to questions about the problem, to gain insight into the phenomena and the relations between them that play a part in the problem. A model is not the solution of a problem, but can be used as a tool to discuss the phenomena and the relations between the phenomena. The model explored in the expectation that, in doing so, (a part of) the problem is also explored.

Note to a: A model can be a representation of the whole problem or a representation of a part of the problem. Models of the whole problem can be used, for instance to express the relation between phenomena, to discuss the (effect of) interventions, or to evaluate scenarios[20]. In the problem analyzing process two main models of the whole problem can be distinguished, the conceptual model and the empirical model. An empirical model is a (conceptual) model of a problem fleshed out with empirical data[21]. A model can also be used to reflect on the problem in order to find an approach to dealing with the problem. An empirical model can be used for suggesting interventions and developing scenarios.

Note to b: Sometimes a more detailed representation of a part of the problem is needed in order to look more closely at what is happening. There are many models that specify a special part of a problem, such as a flowchart, a hierarchical diagram or a color sample.

Note to c: A model can also be used as a research instrument to analyze the relations between variables. It can be used to formulate hypotheses as possible answers to questions, and to test some hypotheses about relations between the phenomena and to compare the structure of the model with reality (Van Dijkum, 1992). This is a way of theory testing. For instance, in a simulation model several hypotheses about relations between the phenomena can be tried out.

An empirical model can be used, for instance, to validate the model itself, in order to see what the relation is between the model and reality.

Note to d: An example of a model to support decisions is, for instance, a multi-criteria model. A multi-criteria model can be used to make it easier to choose several phenomena based on different criteria and different values.

Models can vary from simple schematic pictures to complicated mathematical models[22]. Examples of models are: an iconic model, an analog model, a symbolic model, an object model, a statistical model, a schematic model, a numerical model. What kind of model suits best depends on the kind of problem, the theoretical ideas, the moment in the problem handling process, on the language (the modeling language) a person understands, and the kind of detail what is needed.

The conceptual model can consist of a combination of various models, such as a semantic model, an influence diagram, a causal model, and a simulation model. Together these models give a clearer view of the problem.

An empirical model can be expressed for instance in a (system dynamic) simulation model[23], which will facilitate to make scenarios[24]. The empirical model can also consist of a number of (sub)models.

Different models can be expressed in different languages, verbally, in graphic representations (a graphic model) and in formulae (a mathematical model). By using different models it is possible to express the phenomena and the relations between the phenomena in different ways. The benefit of using different languages is that one can exploit the strong points of one language, while the weak points will be supplemented by another language. Models can be expressed in, for instance, a 'natural' language, a mathematical language, a simulation language, a computer programming language and a graphical language. Using several languages also allows one to admit the different ways that participants prefer to reflect on the problems.

Combining verbal descriptions with, for instance, graphic expressions can make it easier to conceptualize a problem. Some things can be better expressed in words, while others can be better expressed in graphic representations: one picture can sometimes replace many words. Graphic representation of a problem can be a good tool for analyzing the problem, helping to clarify the idea and serving as a communication tool. In a graphic model the relations between the phenomena can be visualized, making it easier to see how the phenomena are connected with each other. Some examples of graphic models are: a semantic model, an influence diagram, a causal model and a system dynamic simulation model[25].

A semantic and a causal model combine graphic representation with a natural language representation. A semantic model is used to describe the relation between the concepts and between the phenomena, in the form of a network. In a semantic model the names of the concepts and the phenomena of the problem are written within circles.

A semantic network is a network in which concepts and phenomena are described in nodes and functionally connected with each other by means of lines. This makes it easier to understand how concepts and phenomena are connected with each other. An influence diagram is a semantic network in which directions of the relations are indicated by arrows instead of lines. A causal model is a model in which the semantic network is rearranged and where arrows indicate cause and effect. In a causal network the directions can be indicated, in the sense of positive or negative influence of the phenomena on each other.

As we have noticed earlier, connections and phenomena can easily be overlooked in complex situations. Although this can never be completely ruled out, one should be alert. A way to conceptualize the possible white[26] and blind[27] spots, is to draw 'knowledge islands'. All the knowledge and data needed to handle the problem can be represented as a graphic space. When the whole graphical space is filled we know all there is about the problem. If we only have some parts of the knowledge that is connected with each other we can represent this knowledge as islands. Knowledge island constitute separate parts of the knowledge about the problem that are connected with each other and of which we know what phenomena are involved and what the relations between the phenomena are. One could presume that the whole computer screen is a representation of all the knowledge that is needed for analyzing the problem fruitfully and fill in some areas that represent some of the knowledge that one already has. These areas can be colored differently. When the knowledge in an area is sufficient to analyze the problem, one can fill in the area uniformly. Other areas could be shaded to indicate that one does know something about this area but not enough. The empty white spots symbolize white spots. The area between the islands represent the possible blind spots.

Many complex interdisciplinary societal problems, as we have stated in chapter two, undergo changes during their development and are imbedded in a changing environment. The above described semantic and causal networks are static models. In order to describe the changes over time a language should be employed that is able to express dynamic changes. Changes in time can be expressed mathematically by means of differential equations, which can also be used to describe how much phenomena influence each other. An example of a program based on a numerical language is a simulation model where time can be expressed by differential equations (Meadows, 1980; Forrester, 1987; Van Dijkum, 1992)[28].

There exists continuous and discrete simulation models. In continuous models, time is simulated continually, whereas in discrete models time is simulated discretely. A computer simulation model may use different ways of expression: a pictorial representation of the phenomena, a graphicl representation, a more symbolic representation of phenomena, or a representation by means of formulae.

A simulation model consists of variables[29] that represent the phenomena constituting the problem. These variables can be manipulated by simulating progress in time or by simulating interventions. The model can be used to simulate changes over time, to try out interventions, to evaluate the effect of interventions, to construct and to compare scenarios[30]. All these models can be used as vehicles for communication. In a natural language there are often words which have an ambiguous meaning, or indicate vague things or are difficult to operationalize. The advantage of expressing a model in mathematical terms is that it can be expressed more exactly. Not everything can be expressed in mathematical language however. One must avoid attempting to describe something exactly which cannot be expressed exactly.

The problem handling process evolves step by step to a higher level of understanding and insight of the problem. The mental idea develops until it approximates a conceptual model of the problem, and at a certain point in this process becoming the more specific conceptual model. Each step in this process can be an on-going, circular and iterative process until the next higher conceptual level is reached and finally the conceptual model can be described by which the problem is defined[31]. In the conceptual model some of the relevant questions about the problem can be answered, and the relevant phenomena and the relations between the phenomena of the problem are described as clearly as possible. When the conceptual model is more or less satisfactorily described, the problem has been defined. The conceptual model may be compared with the initial state of Newell & Simon's problem space (Newell & Simon, 1972).

A conceptual model may contain both known aspects of the problem and unknown aspects which one wants to interrogate. Temporarily answers to these questions and suggestions can be formulated by means of hypotheses. Hypotheses are, then, possible answers to questions and in this respect a step towards a better view of the problem.

The conceptual model of the problem should be constructed as completely and clearly as possible. The conceptual model should be specific enough to consider the problem, at least temporarily, as defined. The conceptual model of the problem describes the contemporary situation of the problem. A full conceptual model of a complex interdisciplinary societal problem can consist of a combination of different models. It should contain a description of the problem in which all relevant aspects and all relevant phenomena and their relations are specified. It should also specify the domains where the required knowledge can be found.

However, in addition to describing the apparent problem, the description here should also cover the reason for this being a problem, to whom it is a problem, who is/are the problem owners and who the problem victim(s) and what the effect is of the problem on society.

This can be completed by a short historical review and perhaps some general remarks on future development. In the conceptual model the concepts and phenomena mentioned in the description, should be carefully defined, as well as the theoretical ideas on which the description is based and on which the connection between the phenomena are based. Finally what is known, what is not known and what has to be discovered should be described.

Knowing the structure of a problem is necessary in order to be able to suggest interventions, which will be the task in the second sub-cycle. Possessing a conceptual model of the problem however does not mean there are no further questions about the problem. The second sub-cycle can also be used to test hypotheses, for instance, in an empirical model (Van Dijkum, 1992).

Defining the problem and formulating the conceptual model is the last step in the first sub-cycle of problem handling.

Bots (1989) states that there may be uncertainty, or equivocality about the problem structure: its variables, relations, operations, evaluation criteria, or any combination of these. Equivocality refers to the existence of multiple and conflicting interpretations about an (organizational) situation. Daft (1986) states that uncertainty is a lack of knowledge with respect to the current state of the variables that are known to be important[32].

nor in our view is this is possible. As indicated above, there are often white spots in the knowledge and data, and blind spots. Given complex interdisciplinary societal problems, it is not always possible to determine exactly when a conceptual model is complete, when a problem is defined. This will be an inter-subjective or intra-subjective opinion. When a person or a group has the impression that the problem is complete, clear, and adequately specified, the problem is, at least temporarily, defined for this person or this group. Although the conceptual model has a direct relation with reality, not all the data have to be included in the model in detail. The conceptual model is primarily a model to articulate the concepts in a person's mind. The primary goal is to know what are the main phenomena and the relations between the phenomena. What we have said about the relation of the mental idea to 'reality' also applies to the conceptual model. The difference between a mental idea and a conceptual model is that a conceptual model is more differentiated. The time required for a fully diffentiated conceptual idea to develop from a mental idea can vary from seconds to years, depending on the problem, the time spent and the people who analyze the problem. It can be a continuous, intermittent or interrupted process.

Wierda (1991) emphasizes the importance of having a good conceptual model of the problem before initiating changes. Wierda analyzes several cases of the development of inter-organizational information systems[33]. Analyzing these cases, he observes that where group problem solving neglects a thorough analysis of the existing situation constant quarrels and misunderstanding later ensue. When there is no consensus about the situation of the problem, suggestions for change often lead to conflict and misunderstanding. Considerable differences between conceptual models of the problem situation make it very difficult to formulate empirical models and to formulate solutions together. He defines conceptualization here as determining the vocabulary in terms of entity-types[34], attributes, activities and relations. Converging the individual concepts to one shared concept can be achieved by dynamic modelling.

Three levels of aggregation can be distinguished. The macro aggregation level, the meso aggregation level and the micro aggregation level. The macro aggregation level is the highest and broadest level, including all aspects of the subject concerned. Regarding a problem on the macro level will normally, because of the wide scope, provide a more global view than the other levels. At the meso level the problem is regarded somewhat less generally and in more detail than on the macro level, though less detailed than on the micro level. The micro level is the most detailed level at which one needs to reflect on the problem. The details dominate the micro level while problems can be approached at any level, the aggregation level dictates the view on the problem. Most complex interdisciplinary societal problems exert effects at all three levels. However, for the sake of the discussion and the overview one level can be selected. The choice of aggregation level depends on what one wants to know about the problem and the aim of the problem handling process. Starting with the problem what is the desired end situation and what should be known in order to achieve this situation. Exactly how the aggregation level is called depends on one's point of view[35].

The aggregation level to be focused on must be made absolutely clear. Failure to explicate the aggregation level can give rise to much misconception in the discussion. Apart from the final aggregation level one wants to select in order to get a view of the influences the problem has on the phenomena and the influences the phenomena have on each other, one should view the entire problem first at the macro aggregation level. First, the conceptual model of the problem should be formulated according to the macro aggregation level. Then, in accordance with the desired situation, one can select a specific aggregation level of the problem (see figure 4).

Many complex interdisciplinary societal problems[36] influence or are influenced by very large numbers of phenomena. It is very difficult and sometimes impossible to formulate a realistic and correct representation of these kinds of problems in a model. One of the difficulties is to represent in a model the many phenomena known to be involved in a model and, at the same time, maintain an overview. This leads to the question of whether or not, and to what extent, the known phenomena and their relations can or should be represented in the model. For many complex interdisciplinary societal problems only a rough model on the macro aggregation level can cover the main phenomena involved in the problem. When more details are needed, the model has to be demarcated, or the model extended by means of several detailed models.

The scope of the mode can be defined in a restrictive manner. The scope should be selected in accordance with the aim of the analyzing process, in accordance with what one wants to know, in accordance with the desired situation, and with what one wants to accomplish. The sub-part must be carefully selected from the total problem. The problem should not be simplified too much, since this could imply that a part of the complexity is left out.

A time scope of the problem is a demarcation of the problem in time. Which time will be considered, a period in the past, in the future, for instance, next year, the next five years or the next decade?

The geographical scope is the area, countries, part of the world, selected for considering the problem.

These scopes limit the view of the problem but still keep all the phenomena and relations between the phenomena intact. This way of artificially limiting larger problems to a smaller scope can be based on efficiency, interests and necessity.

One can also limit the complexity of the problem by focusing only on a part of the whole problem. This is what system theoreticians call a sub-system. Within systems theory[37], a system is an abstraction of a part of reality that, given the problem, is relevant and can be considered as a unity. A system consists of connected entities that act as a unity and execute a kind of behavior. The system boundaries are relevant, but artificially imposed. The system is artificially separated from the rest of the world for the purpose of considering the problem (Morgan, 1971). This system can be represented in a model. An example of a subsystem is the economic system of a society.

Another kind of selection can be a domain scope[38]. In analyzing a complex interdisciplinary societal problem one can temporarily focus on a part of the complex interdisciplinary societal problem; on only one domain. For instance, focussing on the legal domain by focussing only on the legal consequences and possibilities of a complex interdisciplinary societal problem.

The problem should first be viewed in its entire scope at the macro aggregation level. First the whole scope of the problem should be put into the model. The question then arises at what moment in the problem handling process it is permissable to leave the overview of the whole problem at the macro aggregation level and to reduce focus to a scope of the problem, and at what moment should the whole problem be included again.

If the view of the problem, according to the scope, is narrowed too soon the danger exists that one will only get a particular view of the problem, whereas on the other hand, if the scope is narrowed too late the problem handling process will be very complicated.

Should the scope be narrowed in the process of making the conceptual model or at the moment the empirical model has to be constructed? It is difficult to say in advance at what moment a broad complex problem should be narrowed down to a selected scope. It is preferable to narrow the scope as late as possible in the process of problem handling, because it is then possible to keep an open view of all the influences exerted by the problem on the various phenomena. At present we hold the view that the scope may be narrowed after one has a reasonable idea of the impact the problem has on the different phenomena.

In constructing the conceptual model of a problem, one should first strive to include the 'complete' problem, all the relevant phenomena and concepts as far as, and as long as it is possible. First, a rough outline of the conceptual model of the problem should be drawn up, following which scope of the problem can be selected, based on necessity, interest and goal of the problem handling process. Then the conceptual model of the scope of the problem can be constructed. Based on this model, the empirical model can be formulated and, on the basis of the empirical model, interventions can be suggested[39].

One should be aware however, when suggesting interventions and with making scenarios that only a part of the problem has been demarcated and that there remain connections and phenomena not included in the model because they were not included in the scope of the model.

The scope can then be narrowed to the selected scope at the selected aggregation level. That is the reason why, after selecting interventions based on a scope of the problem, one must go back to the overall view of the problem to see whether the parts that have been omitted do not interfere with the suggested interventions[40]. When interventions are suggested on the basis of the scope of the problem, one must go back to the whole overview of the problem in order to see what the effect of the interventions is in view of the whole problem[41].

Keeping this in mind, it is permissible, for practical reasons, of time and money constraints, to view only a scope of the problem during a period in the problem handling process. In the evaluation of interventions, the process is reversed. Focusing again on the selected scope at the selected aggregation level, the scope has to be viewed again at the macro level of the whole problem.

The problem definition, the way the known and unknown but possible aspects and hypotheses in the conceptual model are formulated, is closely related to the subsequent interventions[42]. The definition of the problem is the basis for the second sub-cycle of problem analyzing. Of the many things that can go wrong during the process of problem handling, one of the major mistakes is to define the problem incorrectly.

Sometimes the scope of a problem is defined too narrowly, for instance as a domain problem where, in principle, it should be defined as an interdisciplinary problem. Defining an interdisciplinary problem as a single domain problem excludes many interventions in other domains. Or, as we indicated earlier, when the level of constraints is too restricted, a possible beneficial change toward the desired situation cannot be found.

In defining a problem as a domain problem one restricts oneself to that domain in the search for interventions that could move the problem towards the desired situation. Defining a problem as a domain problem means, by definition, that one believes that the solution lies within that domain or in a certain part of a domain[43], and therefor one narrows the handling space to that domain. Often, too narrow a scope inhibits people from finding an adequate 'solution' to the problem. Defining the problem too narrowly can be one of the major causes for not being able to handle the problem fruitfully.

For complex interdisciplinary problems it can be a long and difficult process to define the problem, to suggest and implement interventions. Although not in all cases, one can say that the more complex a problem is, the more difficult it will be to define and change the problem.

Rosenthal (1984, p. 63-64), in analyzing the interventions that were suggested to handle crisis and conflicts, confirms that people, in this case policy makers, have a strong tendency to work from their definition of the situation whether this definition is wrong or right and tend to neglect information that is in contradiction with their strategy. This leads to one of the propositions of Rosenthal (proposition 10, 1984, p. 63):

"In a crisis situation decision makers are not able to re-define the situation. They are obsessed by a dominant goal-means scheme[44]."

Rosenthal finds confirmation of this proposition in the analysis of the Korean crisis and the Cuban crisis. Managers tend to interpret their information in critical or difficult situations in the way that best fits their way of thinking. However, they tend to emphasize the advantages of their approach and to neglect the disadvantages of their approach.

To avoid obsession with the dominant goal-means scheme, Rosenthal (1984, p. 64) gives some advice:

"Search for more and different kinds of information and try to avoid incorrect historical analogs. In recruiting participants for decision making choose pluriform participants and participants who are able to carry out critical analysis.[45]"

Sometimes a problem is so easy that it can be defined immediately or even be solved directly. In this case the problem handling process will be short and easy.

The second part of the problem handling cycle focuses on how the problem can be changed. In the second sub-cycle the following phases can be distinguished[46]:

The second sub-cycle in problem handling begins with construction of the empirical model. Here the desired situation, the goal of the problem handling process, if possible, will be described again. On the basis of the empirical model, itself based on the conceptual model of the problem, the desired situation can be formulated more sharply. The empirical model is constructed in accordance with the aim of the problem handling process, which means that the model should be constructed in such a way that interventions can be explored that might lead to the desired situation. It is mostly a model of a selected part of the problem, depending on the desired situation, the aim, the selected aggregation level and the selected scope(s). When the desired situation is not yet known, the empirical model can provide the opportunity to test certain possibilities of desired situations.

Although the empirical model and the conceptual model are both based on theoretical ideas in combination with real data and both function as a bridge between theoretical ideas and reality, the use and the construction of a conceptual model differs from that of an empirical model. In the conceptual model the emphasis is on the phenomena involved and on how the phenomena are related to each other. In the conceptual model the precise data are less important. In the empirical model, the emphasis is on the extent to which the phenomena are related to each other. In order to make an empirical model one must know precisely what the phenomena look like, what their value is and to what extent the phenomena influence each other, although a model is, by definition, always a derivation of reality.

In the empirical model, the conceptual model acquires empirical data. Codd (1985) states that the requirements for data (-bases) in computer science are that data have to be trustful, up-to-date, complete, correct and relevant, without homonyms[47], synonyms and redundancy[48]. This prescription is made for automatic (computerized) information systems, and although it would be desirable to have the same data reliability for complex interdisciplinary societal problems, in many situations this will not be possible.

For small problems within a domain, the data and information can be complete. Often the data of the problem on which the research of cognitive science of problem solving is based, is complete and reliable, as it was in the experiments of Newell & Simon (1972). This is one of the differences between handling complex interdisciplinary societal problems and the problems cognitive psychology focuses on.

With real problems, especially those that are more complex, the data are seldom complete. Often it is not even possible to know whether the model is complete let alone whether the data concerning this problem are reliable. Even if it were possible to make a complete model of a complex problem, it is not always possible to obtain complete data. Sometimes the circumstances of a problem change so fast that it is almost impossible to have updates[49].

Making an empirical model of a complex interdisciplinary societal problem is not easy. It often becomes painfully clear that it is not always possible to fill the model with relevant data. Data may be hard to obtain, unavailable, unavailable to the problem handlers[50], or there may be no time to gather the data[51]. The data may be incomplete, in contradiction with each other or may appear to be unreliable. In many cases it will only be possible to construct an empirical model that fits reality (DeTombe, 1993)[52] to a limited extent.

Data are required in both sub-cycles of the problem handling process. How specific the relevant data should be, however differs for each sub-cycle or phase. For constructing the conceptual model more general data will suffice, whereas for the empirical model more detailed data are necessary, while for suggesting and evaluating interventions quite different data could be required. What data are relevant depends on the problem and the moment in the problem handling process.

One may wonder whether complete data would enhance predictability. Whenthere are complete data of a problem and/or a complete model would it be possible then to predict how this problem is going to develop in the near future? Where the problem can be represented by a mathematical, linear model this is possible, but chaos theory[53] states that even in a static context and even when the model of the problem is completely expressed in a formula of a non-linear model, even then it is possible that at some moments the model cannot predict what is going to happen in the future[54].

After constructing the empirical model, the next step in the problem handling process is to define the handling space. Defining the handling space enables one to define how much and what can be changed in order to reach, or to approach the desired situation.

The handling space is a metaphor, a mental construct, for the space where interventions of the problem will be sought that might lead in the direction of the desired situation[55]. The handling space limits the space in which, and to what extent, the problem can be changed. The handling space is a different concept from the term 'problem space' of Newell & Simon (1972)[56]. The handling space as such is indifferent to whether the change will actually lead to the desired situation. One can only hope that it will be. The handling space can be described in terms of levels and kinds of constraints.

In changing complex interdisciplinary societal problems one has to take many constraints into account. These constraints narrow the handling space. To be able to indicate the different range of possibilities for changing the problem, we distinguish different levels and different kinds of constraints.

We distinguish four levels of constraints. The first level is the most restrictive, the fourth level is allows the most freedom. At the first level of constraints the interventions of the problem will be searched for within the current situation. At this the most restricted level of constraints, in principle the whole situation remains as it is, with only relatively small changes within the existing situation allowed. This idea comes close to what is colloquially called 'muddling through'[57]. At societal level this includes new laws, a better infrastructure and changes in pension for the elderly.

The second level of constraints allows some more changes in the contemporary situation, although not too many, but the changes can be greater. There is more space to handle the problem and there are more possibilities for change.

The third level of constraints broadens the possibilities as wide as can be, but still within the 'normal' possibilities of mankind and nature. On the societal level this involves fundamental changes in organizations, in politics and even in the way people think, hope and believe. This can constitute a totally new form of society[58].

The fourth level of constraints abandons the constraints of human possibilities and escapes into imagination. It is a level that can no longer be fruitfully implemented, but the most can be used to 'unfreeze'[59] people in the problem handling process. The distinction between level one, two and three is gradual. From changes within the existing situation (level one) to major changes of the situation (level two) to a whole new approach of living (level three). The distinction between the first three levels and the fourth level however is qualitative. Here, the levels of constraint pass from realistic (level one to three) to unrealistic (level four).

Note to level one: The situation of a firm which does not make enough profit can be changed to a (more) desirable situation by changing the management strategy, by buying new machines that do the work faster, or by firing some of the employees and hiring new ones. At a societal level, this means, for instance, limiting the speed of cars and limiting noise pollution.

Note to level two: The firm could decide to branch out in a wholly new direction and with new products. At a societal level, this level of constraints includes, for instance, a group of individuals who no longer wish to subscribe to the societal rules and wish to live by their own rules, like the Amish people in the USA, or the Walden group in Germany in this century.

Note to level three: The firm quits the capitalistic way of making money and changes to a completely new kind of organization. At a societal level, it includes the ideas of Marx (1890) about changing the position of people in the society in favor of the laborers or the ideas of Rousseau about education (Rousseau, 1762).

Note to level four: Living in an imaginary world or in fiction. This occurs, for instance, in the use of deus ex machina in theatre plays in the middle ages[60]. In this area one can find science fiction, and the fantasy of mentally disturbed persons (psychosis, schizophrenia). Here the solution of the problem is discovered in the mental realm instead of operating in the real world[61].

Research on how people like to live has shown that most of the interviewees remained at the first level of constraints. They stayed within the boundaries of their own, contemporary situation (Barre, 1982). When the constraints are too restricted it will not be possible to find a satisfactory change. In order to reach the desired situation the range of possibilities should be enlarged. Special effort has to be taken to unfreeze people in order to stimulate thinking about new ideas and different possibilities. Sometimes an intervention cannot be found at the first or second level, in which case one should raise the level of constraints.

In practice many problems are handled within the constraints of the present situation at only the first level of constraints, where only slight changes are allowed. The interventions will, in principle, not fundamentally change the situation.

Besides levels of constraints, there are different kinds of constraints: financial constraints, political constraints, psychological constraints, geographical constraints, physical constraints, time constraints etc. Each of these constraints can be located at the different levels. Examples are:

Financial constraints: some interventions are too expensive to be implemented. Organizational constraints: some interventions cannot be organized given the situation. Political constraints: some interventions cannot be carried out because it is politically not possible at that moment. Time constraints: a change of the problem has to be found within three months[62].

Our earlier account of suggesting, exploring and working out hypotheses[63] in constructing the conceptual model also applies to hypotheses in suggesting interventions.

Based on the empirical model one can work out hypotheses in order to see how certain phenomena are related with each other and test the outcome in reality.

The empirical model can also be used to suggest some interventions that will change the problem in the desired direction. This can be done on the basis of theoretical ideas about the effect of intervention.

An empirical simulation model can be used to see how a system behaves. It can be used to describe the behavior of a certain system. Playing with the system will give a better insight into the system, its bottlenecks and the consequences of alternative interventions.

Sometimes, in practice, suggesting interventions is more a matter of trial and error[64]. It can be an iterating process of suggesting interventions and implementing them, evaluating the effect, then redefining the problem, making an empirical model, suggesting (other) interventions etc[65]. In suggesting interventions one includes the constraints. As a kind of thought experiment, one could try changes that include changing the different levels of constraints in order to see where the desired situation can be found.

As said earlier, in evaluating the suggestions, one should return to the whole problem and include all the parts and the scopes one had previously excluded, in order to see whether the suggested interventions are the correct ones.

Making, selecting and evaluating scenarios should be carried out in accordance with the desired situation.

An empirical simulation model can be used as a model for making explorations of future developments of the problem or for making explorations of what the effect of the interventions on the problem will be. This is carried out in so-called scenario studies. The creation of scenarios is also based on theoretical ideas. Scenarios can be defined as explorations of future development (Jager, 1990).

Comparing the effect of interventions by exploring the future with each other implies that one is able to select an optimal strategy for changing the problem in the desired direction. Interventions should be chosen in accordance with the desired situation.

A system dynamic simulation model can be used as a scenario for exploring future developments, as carried out, among others, by Fleissner and his colleague Bruckmann (1989) who made a prediction of the Austrian economy based on a system dynamic model. Meadows, Meadows, Randers & Behrens III (1972) use system dynamic modeling to predict the use of future resources in the Club of Rome book 'Limits of Growth' and in their book 'Beyond Limits of Growth' (Meadows, Meadows & Randers, 1992). In the Netherlands these models are used, among other things, for future prediction at the National Institute of Public Health and Environmental Protection (RIVM)[66] and at the Research Center of Public Mental Health Care (NcGv)[67]. The outcomes of these studies do sometimes have a significant impact on society, as the report of the Club of Rome shows.

For a scenario for policy decisions, the model and the data in the model should be as closely related to reality as possible. Given the issues discussed earlier about the uncertainty of the data, about missing knowledge and data, about white and blind spots, and the incomplete model, it is questionable in how far system dynamic models of a complex interdisciplinary problem can be useful for future exploration and future prediction. With this consideration we arrive at expectation two:

models of complex interdisciplinary societal problems contain so much uncertainty that scenarios based on these models will also contain a large degree of uncertainty. This makes it hard to base reliable predictions on these scenarios for the future development of the problem

what is the relation between a scenario of a complex interdisciplinary societal problem and reality?

The possibility of using systems dynamic models as a scenario for future predictionhas been criticized from the point of view of the theory of complex interdisciplinary societal problems, from the point of view of some system theoreticians and from the point of view of chaos theory.

The criticism of the use of systems dynamic models as a scenario for future prediction is based on the theoretical ideas about complex interdisciplinary societal problems. Because the model of a complex interdisciplinary societal problem contains so much uncertainty, scenarios based on them are themselves uncertain and unreliable for policy decisions.

The use of systems dynamic models for future predictions is also susceptible to criticism from systems theory. Before we discuss what systems theory is, we shall discuss were systems theory comes from.

Before the ideas of systems theory were formulated there was a discussion between mechanist and vitalist points of view towards organizations and parts of the society.

"Mechanistic thinking adheres to analysis and reductionism claiming that all objects and events and their properties can be understood in terms of ultimate elements", (Flood & Jackson, 1991, p 3)

The mechanistic view was that everything that occurred was completely determined by something which preceded it. The mechanistic view leads to the view that the universe is constructed of building blocks ordered in a hierarchy, constituting a gigantic machine. This idea could easily be applied to organizations. Indeed the 'classical' or 'rational' view of organizations sees them as made up of parts, each of which can be optimized independently in pursuit of some goal. However, this has not worked well. Where the parts were all independently optimized, the organizations failed to perform as well as a whole, .

Systems thinking emerged as a response to the failure of mechanistic thinking to explain biological phenomena. In the organicist tradition on the other hand, organisms were now treated as whole entities, or systems whose identity and integrity had to be respected. They had 'emergent' properties peculiar to themselves which could not be derived from their parts. They were 'open' rather than 'closed' to their environments. This thinking was soon transferred to the study of other 'systems' such as organizations. The need for systems thinking was established (i.e. thinking at least about the interdependence of the parts). However, since the system view was originally born in biology it tended to rely on biological analogies, introducing ideas such as survival, adaptability, development, growth, flexibility and stability.

" 'System' is a term that is widely used in contemporary Western society. This is so much the case that it has effectively been rendered meaningless in everyday use (or should it be misuse?)." (Flood & Jackson, 1991, p. 2)

"There is a whole literature discussing the meaning of the term 'system' and the advantages of systemic over reductionist thinking" (Flood & Jackson, 1991, p. 1)

"our idea about the concept 'system' is in two complementary ways: first in the modern system approach the concept 'system' is used not to refer to things in the world, but to a particular way of organizing our thought about the world. Second we consider the notion of 'system' as an organizing concept, before going to look in detail at various systematic metaphors that may be used as a basis for structuring thinking about organizations and problem situations"

"The system perspective recognises multifarious interactions between all the elements making up a complex situation. A related change overtook the idea of 'system'. In mechanistic thinking a 'system' is an aggregate of parts in which the whole system is equal to the sum of the parts. In systems thinking, a 'system' is a complex and highly interlinked network of parts exhibiting synergistic properties-the whole is greater than the sum of its parts." (Flood & Jackson, 1991, p. 3)

"organisms are open systems with energy and material and material entering and leaving them; organisms are not 'at rest' within their immediate environments.

Thus the equilibrium idea from physics was abandoned in biology and replaced by homeostasis, a concept that refers to the maintenance of a steady state, a kind of continuity, in a changing environment." (Flood & Jackson, 1991, p. 4)

"A system consists of a number of elements and the relationships between the elements. A richly interactive group of elements can be separated from those in which few and/or weak interactions occur. This can be achieved by drawing a boundary around the richly

interactive group. The system identified by a boundary will have inputs and outputs, which may be physical or abstract. The system does the work of transforming the inputs into outputs. The processes in the system are characterised by feedback, whereby the behavior of one element may feed back, either directly from another element by way of their relationship, or indirectly via a series of connected elements, to influence the element that initiated the behavior. We give attributes to the elements and relationships according to how we measure them (e.g. for an element we might use size, weight, color, number, volume; and for relationships, measurements might be in terms of intensity, flow strength).

A system so described is separated by its designed boundary from its environment. It is termed an open system if the boundary is permeable and allows inputs from outputs to the environments." (Flood & Jackson, 1991, p. 5, 6)

We can apply the way an organization is viewed in systems thinking to represent problems and use a systems thinking approach for analyzing them. For instance the 'temporary' boundaries of an open system, including input and output and the communication between the elements. We also assume the concept of hierarchy in which a system can be a part of a wider system and in return can consist of smaller sub-systems (Flood, 1991, p. 6, 7). The macro, meso and micro aggregation levels can be understood in this way.

In his book Industrial Dynamics (1961) and Urban Dynamics, Forrester explains how the ideas of system theory can be applied to simulation models[68]. Gordon (1960, p. 83)

"The principle concern of System Dynamics study is to understand the forces operating in a system in order to determine their influence on the stability or growth of the system. The output of the study will, it is hoped, suggest some reorganization, or change in policy, that can solve an existing problem or guide developments away from potential dangerous directions. It is not usually expected that a System Dynamics study will produce specific numbers for redesigning a system, as occurs with engineering systems."

Using system dynamic modeling techniques has been criticized from two directions in systems theory, from the direction of hard systems scientists and the direction of soft systems scientists. Their respective critiques refer to theory, methodology, ideology and utility. We will reflect on some points described in Flood & Jackson (1991, p. 78-83).

Soft systems thinkers question the underlying assumptions of systemdynamic modelers that there is an external world made up of systems, whose structure can be grasped using models built upon feedback processes. To soft systems thinkers social systems are much more complex than this[69]. The soft system scientists state that the model cannot be quantified. We agree with this.

A more fundamental criticism is that subjective intentions of human beings cannot be captured in such 'objective' models. Models should be designed to increase mutual understanding, not to seek to represent external reality.

Another criticism is that systemdynamic modeling offers no point of view for comparing the model with other models.

"...... the creative construction of human beings whose intuitions, motivations and actions play a significant part in shaping 'system' behavior. ....System dynamics do not deal with the innate subjectivity of human beings and the consequences of this for the study of social systems."

We agree with this, because we recognize that uncertain factors can be interpreted as objective factors when they are captured in a model. This is one of the reasons why we emphasize the subjectivity of a model[70], and emphasis that the same group that has modeled should also suggest interventions and suggest the policy ideas.

Theoretical criticisms of hard system scientists (Flood & Jackson, 1991, p. 82) also emphasize that the group that has modeled should be the same group to suggest interventions and suggest the policy ideas.

"System dynamic jumps to conclusions about whole system behavior before the data have been collected and the laws verified which would make such conclusions justifiable."

This criticism echoes the critique coming from the theory of complex interdisciplinary societal problems. However, our response to this criticism is that this difficulty is inherent in the kind of problems referred to here.

"Correspondingly, many of the coefficients in the models of System Dynamics studies consist of estimates or best guesses, particularly since the models must sometimes reduce such qualitative factors as personal preferences or social tension, to quantitative form. Nevertheless, the lack of precision that may have to be tolerated does not destroy the value of the study. The model can establish the relative effectiveness of different policies under the same assumptions, or mark out ranges of values that can be expected to produce a given type of output."

Another critique of the hard system scientists is a critique on the methodology. There is no sufficient information to make the empirical model. This critique comes close to our critique of missing data and knowledge.

Another critique of the hard system scientists, is an ideology critique. The people who build the model have the idea, that they are elite technicians. The model is theirs. They do not allow any involvement of other 'stakeholders' (Flood & Jackson, 1991, p. 81).

This is also our critique to making models by (one to three) people of an institute such as often happens in reality[71]. The model is made by a small group of modelers, not by content experts[72].

Another critique of the hard system scientists concerns the utility of the model. The system dynamic model uses poor data. We agree to that, however for complex interdisciplinary societal problems there is nothing better.

The third critique of system dynamic modeling comes from the side of chaos theory. This criticism concerns two aspects, the unpredictability due to incomplete data and the unpredictability due to non-linear feedback loops.

In an attempt to find ways to describe the uncertainty in the prediction of the future development of complex interdisciplinary societal problems we turned to chaos theory. It is hard to find a suitable definition of the chaos theory. Tennekens (1990) describes chaos as a feature of:

"The theory of chaotic behavior of simple dynamic systems has been thoroughly examined the last 25 years." (Tennekens, 1990, p. 9)[74]

Gleick (1989, p. 14) connects chaos theory with such concepts as turbulence, strange attractor, fractals, periodicity. Toffler (1987, p. 9-24) connects chaos with concepts as uncertainty, dissipative structures, self-organizing, equilibrium systems, nearly-equilibrium systems and systems far from equilibrium, non-linear processes, fluctuations and the irreversibility of time. Prigogine & Stengers (1984) write about stability, fluctuations and entropy[75].

The paradigm of chaos theory seems able to provide a description and an explanation of certain issues which, until some decades ago, were unnoticed, were neglected or seemed to be inexplicable. An understanding of chaos theory enables us to notice things in a way we did not notice before. The language of chaos theory enables us to describe some of these views. Chaos theory as it is now developed, describes many and sometimes totally different matters. One of the general focuses of chaos theory is unpredictability. We will first discuss the unpredictability that is due to incomplete data and then we will discuss non-linearity.

"a certain intelligence would be able to include in one formula the movements of the largest celestial bodies and the tiniest atom. For her nothing would be uncertain and the future as well as the past would be like the present." (Gleick, 1987, p. 21)

Scientists once had the idea that everything can be known, predicted and possibly even controlled once all the data and the connections between the data are known.

In physics the common opinion is that unpredictability is a matter of incomplete data and limited capacity of computers. Tennekens (1990a) pointed out that according to meteorological models, even if one triples the data collecting points and ten times doubles the computer capacity, the weather cannot be predicted for more than five to seven days in advance with reasonable accuracy, and often not even that. Van Dijkum states in his article on unpredictability and chaos theory (1992, p. 5):

"A process that is unpredictable is undetermined. Time is in a prediction a variable that together with the values of other variables determines the state of the process. If there is no prediction possible then there is also no determination possible of all values." (Van Dijkum & De Tombe, 1992, p. 26)

The opposite case is that, when there is no definition of all the values of a function, there is no prediction possible[76].

Chaos theory shows that even in physics it is sometimes impossible to predict future developments (Broer & Verhulst, 1990). The theory of complex problems, as it is described in this study, shows that data of a complex interdisciplinary societal problem can never be complete. Even when there are no blind and white spots, which there always are, the data would never be complete because of the changing problem and the changing environment.

A part of chaos theory describes the unpredictability of non-linear processes (Gleick, 1987; Prigogine & Stengers, 1984; Lorenz, 1979). The models of complex interdisciplinary problems contain many non-linear feedback-loops. There can be moments in which these non-linear feedback loops are unpredictable[77].

Mankind has always been preoccupied with predicting the future. Often people who could predict the future were highly respected in the community, such as the priest in Greek culture -500 till 0 before Christ[78], the priests in the Bible, and now scientists are asked to predict the future. Knowing what is going to happen is very important for all kind of matters, such as preventing a problem from becoming a large problem, making investments and planning one's life.

It will be clear that empirical models that reasonably match complex interdisciplinary problems cannot be made and that scenarios that correspond to the future will be almost impossible. This does not mean that scenarios should not be made[79]. One can at least try, by making future explorations, to see what is going to happen. In a null-option[80] of a scenario one can try to see what is going to happen with the related phenomena that are involved in the problem in the future, provided there are no interventions. Other options (non-null-options) can suggest interventions and explore the effect of these interventions.

As we have already pointed out a system dynamic model used tp predict complex interdisciplinary societal problems should be handled with the greatest possible care. There is risk involved in making policy based on scenarios of complex interdisciplinary societal problems. The range of uncertainties can be so great that one can hardly expect to be able to make real policy decisions on these scenarios.

Nevertheless, future predictions are often based on uncertain models and uncertain scenarios in reality[81]. Such scenarios intended for political use already have considerable uncertainty built in and besides this, are also often misunderstood or misused.

The quality of interventions and scenarios will depend on the way the whole process of problem handling is performed. For complex interdisciplinary societal problems there will always be a large number of uncertainties in the model, in the effect of the interventions and in the scenarios. Even when the interventions are carefully selected and the comparison of scenarios is carefully carried out using as much knowledge, tools, methodological support and human effort as is available, one should be very carefully in using scenarios for policy making (DeTombe, 1992a).

Nevertheless, for the want of anything better many scenarios will be used for policy making[82]. The danger of doing this is that, although the designers of the model may be still aware of the problematic data and the uncertainties of the relations in the model, once taken out of the hands of its creators the model tends to live its own life. Other people may use the scenario as a model of reality rather than as a concept that can be used in discussion of the problem. A scenario can predict at most the direction(s) in which the problem can develop.

If interventions are carefully selected, one can begin implementing the interventions in reality. After selecting a scenario, a strategy should be planned and a decision should be made as to who should implement the changes. Change can cause a good deal of trouble and nuisance and can encounter resistance[83].

Knowing how to change a problem is not the same as to actually implementing the interventions[84]. Even knowing where to look for changes or to decide what kind of interventions are necessary, does not mean that the problem can be changed.

The actual process of implementation of the interventions in practice can be very difficult. Selz (1922) speaks of the difference between 'Mittelfindung' and 'Mittelanwendung'[85].

Even if a problem can be clearly defined, and both the major points and relations are known, it can still be very difficult to change the problem. Sometimes it is not necessary to change much, sometimes changing some elements leads to the desired situation.

In practice not all the interventions will be implemented even when they are clearly useful. Reasons for not implementing them may be that several changes have already been made that make it no longer worthwhile[86] or the policy for handling these kinds of problems has changed. Problem handling does not always mean actively changing the problem. Sometimes doing nothing can be a way of handling the problem.

The implementation of the interventions will result in a change of the problem. At that moment, one can reconsider whether or not the new situation is closer and close enough to the desired situation. If the situation has changed reasonably in the desired direction, one can leave it as it is. When there is still a large discrepancy between the desired situation and the new situation, one can consider the problem again, when there is the need, the time and the money to do this. The problem handling cycle will start all over again, but now with a changed problem.

Many problem handling researchers describe the different phases of the problem handling cycle. De Groot (1969) describes the phases in scientific research, for explanatory scientific thinking, which he calls the empirical cycle. The empirical cycle consists of the phases: observation, induction, deduction, testing and evaluation[87].

In 'group-problem-solving and decision-making' of organization consultancy, Schein (1969) distinguishes two sub-cycles each consisting of three phases.

The second sub-cycle identical, but now referring to action instead of merely to the thinking process:

- situation action planning which can be considered as a new problem definition (what are the possibilities and the obstacles of change);

Schein includes analyzing a problem and suggesting interventions in the first sub-cycle. The first sub-cycle contains some of the activities that we include in the first sub-cycle (defining the problem, the (temporary) diagnoses of the starting situation) and some of the activities we include in the second sub-cycle of our model (suggesting interventions; predicting the effect of the interventions and evaluating). Schein's second sub-cycle includes some activities we have included in our model in phase 2.2 to phase 2.5 (defining the handling space, suggesting interventions, implementing the interventions and evaluating the effect).

Van Strien (1975a; 1986, p. 19) distinguishes a think and handling cycle which he calls a regulative cycle for problem handling in practice. This is a special problem analyzing method developed for practice. The phases of this cycle are:

Comparing the phases of Van Strien with the phases in our model, we find that Van Strien distinguishes five phases where we distinguish nine. What Van Strien calls 'definition of the problem/formulation of the problem', is in fact an analysis of goals and wishes for constructing the desired situation (H. 't Hart, 1991a, p 55). This phase of Van Strien's model can be compared with defining the goals and the desired situation that is part of our phase 1.4 'forming the conceptual model' and phase 2.2 'the empirical model'. Van Strien's concept 'definition of the problem' is not the same as our 'defining the problem', which we use to refer to the conceptual model of the problem.

Van Strien uses the term 'diagnosing' for analyzing the situation. Although this is not quite the same as constructing a conceptual model, the diagnosis phase of Van Strien may be compared with a series of phases for defining the problem in our model: from extending the mental idea of the problem to defining the conceptual model of a problem (phase 1.2 - phase 1.4). Similarly Van Strien's 'plan' can be compared with 'suggesting interventions' combined with 'making scenarios' in our model, that are part of our phases 2.3 and 2.4.

The phase 'actualization of interventions' of Van Strien can be compared with the phase of 'implementation of interventions' in our model (part of phase 2.5), while

Van Strien's 'evaluation' phase can be compared with 'evaluating the effect (of interventions)' in our model (part of phase 2.5).

One may wonder how it is possible to analyze goals before first analyzing the problem, but sometimes it is possible. For instance, with a health-care problem the goal can be to become healthy. But with many problems and almost always with complex interdisciplinary problems, it can happen that after carefully defining the problem one realizes that the primary goals were not the right ones, and that the goals should be adapted. In practice the goals are also often adapted after trying out interventions (Van Strien, 1986). As said earlier, in practice one often jumps from forming a mental idea of a problem to suggesting interventions and back again to data gathering etc.. In our model one may jump from phase 1.2 to phase 2.3 and back to phase 1.2 respectively, or there may be a mixed iterating process of defining the problem and implementation of interventions. This way of problem handling is more or less a matter of trial and error, which is sometimes defended by stating that in order to see how phenomena are related, it is necessary to try out interventions[89]. But even then, one can define the problem as well as possible first, try out several hypotheses, suggest interventions and evaluate scenarios before implementing them in reality. Performing trial and error in practice can cause much unnecessary trouble for many people. Van Strien's phases refer to problem handling in practice. In practice a problem is often only weakly formulated, only labeled and not yet properly defined. Even a labeled problem should be defined first, also in order to see if the defined problem can be recognized. On the sequence of phases in the regulative cycle Van Strien (1986, p. 21) states:

"The regulative cycle is no more than a reconstruction of an, in practice, irregular process .......But making such a styled reconstruction is a condition for the methodological discipline and responsibility which the practice needs.[90]"

Simon (1969) has the same ideas about the regulative cycle when he refers to design. Simon defines design as developing artifacts with certain design characteristics. This implies gathering information about reality, analysis of alternatives and comparing demands, and choosing a satisfactory alternative. This differs from fundamental scientific thinking in which analysis is the central issue. Vlek (Vlek & Wagenaar (1979, p. 294) state that there is a clear similarity between the phases of the rational decision making analysis and the regulative cycle. Vlek's phases of problem handling contain nine steps which are basically the same as those of Schein (1969).

The phases we define are iterative. After defining a problem, it can happen that focusing on the interventions in the second sub-cycle leads one to return to the first sub-cycle in order to implement the new knowledge derived from the interventions into the model. H. 't Hart (1991) says about the phases of Van Strien that it does not have to be necessary that all phases will always have to be followed in the given sequence.

We can agree with this when rather small, domain related problems are concerned, such as medical problems or problems in the chemical domain. In such domains the problems are often handled this way. However, where complex interdisciplinary societal problems are concerned, this way of problem handling should be strongly discouraged. For these kinds of problems the phases should be traversed in the prescribed sequence. Although we can agree with H. 't Hart that at the moment of making plans for intervention it can become clear that the problem definition and the diagnosis should be adapted[91], in practice the goals of research are prescribed by the problem definition of the client and the available methodology (H. 't Hart, 1991). In practice, handling the problem is not so strictly divided into phases as is described here. For instance, in therapy the phases diagnosis, intervention and evaluation often alternate within minutes. In reality, the process will be more diffuse and mixed. Phases can be skipped or can be passed in different directions.

Doerbecker (1979) argues for combining the problem definition, solutions and means, and including the future solution and means in the definition. He argues for an iterating search process[92]. This is very practical. We believe that in real life this will often be the way one deals with problems. However, we strongly dispute this way of working because it already includes certain solutions and means at a moment when it is better they should not be included. There is a chance that the problem handling will be too restricted and that new changes, new 'creative' solutions will not get the chance they deserve.

We are of the opinion that for handling complex interdisciplinary societal problems the phases should be followed in order to guarantee a fruitful result, and that these phases should not only be sequentially but also be repetitively gone through[93].

There is an interaction between the problem and the problem handling process. Human beings are subject to cognitive constraints with regard to problem solving in general and decision making in particular. One can call this bounded rationality. March (1978) distinguishes several kinds of bounded rationality:

"Limited rationality: emphasizes the extent to which individuals and groups simplify a decision problem because of the difficulties of anticipating or considering alternatives and all information."

"Contextual rationality: emphasizes the extent to which choice behavior is embedded in a complex of other claims on the attention of actors and other structures of social and cognitive relations.

Game rationality emphasizes the extent to which organizations and other social institutions consist of individuals who act in relation to each other intelligently to pursue individual objectives by means of individual calculations of self-interest.

Process rationality emphasizes the extent to which decisions become meaningful in attribute of the decision process, rather than in attributes of the decision outcomes.

Adaptive rationality emphasizes the experimental learning by individuals or collectives

Selected rationality emphasizes the process of selection among individuals or organizations through survival or growth

Posterior rationality emphasizes the discovery of intentions as an interpretation of actions rather than as a prior position"

"In order to understand the process of problem solving one has to understand both the task environment in which the problem is located, and the limits of rational adaptation to task environments. .....when we study a properly motivated subject confronted with an intelligent task, we are observing intendedly rational behavior or behavior of limited rationality (Simon 1947)."

In each phase of the problem handling process there are moments where a decision about what to do, how to continue, what to choose, what to select, has to be made. Decision making is a part of all the phases in the problem handling process. What aspects play a role in decision making? Rosenthal (1984) distinguishes six theories of decision making:

The ideal type of decision-making is: the decision maker knows all the alternatives, all the consequences of the following alternatives. He or she is able to organize them in order sequence of preferences and chooses that alternative which leads to the most preferable consequences. According to Rosenthal, this theoretically leads to the best decision. Although in agreement, we nonetheless doubt that it is possible. This approach to decision-making costs much time, human and other (material) resources and money. Moreover, we think that the rational-synoptic theory is only an ideal-type, and that it is not realistic to think one is able to analyze all the alternatives. When the range of the problem and the handling space are too large, and this is soon the case, the handling space cannot be searched exhaustively by humans. In practice, a person will narrow the search for intervention and selecting hypotheses by using heuristic search techniques[94].

Braybrook & Lindblom (1963) also criticize this type of decision-making. One of the reasons that claim that the rational-synoptic theory cannot be applied is the inadequate information about the problem, and the limited problem handling capacity of humans. One cannot find all the alternatives and calculate all the consequences. Etzioni (1968) finds the rational-synoptic theory too demanding, unrealistic and utopian. Neither does Simon (1957) find this theory very realistic.

The optimum theory of decision-making is propagated by Dror (1964). Besides rational aspects irrational aspects like intuition, experience[95] and insight are also taken into account in decision-making. In decision-making, meta-policy-making plays an important role. In the optimum-theory the emphasis is not on the search for accurately determined or even quantifiable decisions, but rather for an optimal decision, a qualified positive determined and even quantifiable difference between marginal output and marginal input. The optimum-theory is different from the rational-synoptic theory in that it also includes irrational aspects of decision making. Dror (1964) wants to give a realistic theory that is oriented towards an optimal result. In this theory, personal, material and time costs will be realistic factors.

In the mixed scanning theory formulated by Etzioni (1968), decision-makers list all the relevant alternatives they can think of, including the less familiar ones. From this they select useful alternatives to be analyzed in detail. This procedure continues until only one alternative remains. The decision-makers carry out interventions in such a way that there will be room to adjust to new information. The mixed scanning strategy is, in particular, meant for situations in which strategic decisions will be made. This way of decision-making is used for military decisions and for meteorological information collecting, a combination of global and detailed information. For Etzioni, the mixed scanning theory is a workable alternative to the rational-synoptic theory.

In the theory of the satisfying alternative of Simon (1957), the decision makers postulate some logical alternatives, based on experience, and determine the aspiration level and the goals. In relation to the goal, he or she decides to take the first alternative by which the goal can be reached. This does not lead to the theoretically best solution. Costs in terms of human effort, material and time are less than in the other procedures. The starting point of this decision-making theory is 'bounded rationality', meaning that the intellectual capacities, habits, preferences and knowledge of things are bounded, and that the policy making itself has its boundaries. A negative aspect of this procedure is that the solution that is found may only be a temporary solution which can sometimes create even worse problems for the future. Simon (1945[96]) states that organisms, including human problem solving, adapt well enough to 'satisfy', they do not, in general 'optimize'.

"They seek improvements with respect to the current situation, but will not continue their search until they have found the optimum solution (even when they would be able to determine whether a solution was optimal or not)." (Newell & Simon, 1972, p. 791)

The incremental theory is associated with Lindblom (1964), who calls this way of decision-making 'muddling through'. The decision maker concentrates on well-known alternatives, aims at marginal improvement of the existing situation and gives much attention to the means of realizing this improvement. He or she chooses a change that only slightly deviates from the existing situation. Decisions are often compromises of vested interests. However, a range of incremental decisions can lead to a quite different situation.

The theory of non-decisions with which the names of Bachrach & Baratz (1963) are associated, emphasizes why policy-makers do not take a decision eventhough there are problems enough demanding a decision. Rosenthal (1984) says that a general conclusion about the quality of non-decisions and about the costs of the decision making cannot be drawn. In some cases, problems disappear on their own and in this case interference by the politico-governmental institutions will only harm the case, and in some cases make the problems worse.

The fifth and the sixth ways of decision making match the first level of constraints[97]. We believe that people make when handling complex interdisciplinary societal problems based, make decisions on the basis of mixed approaches of rationality and irrationality. At some moments in the problem handling process, and these moments can change, rational decisions can be made and at some moments irrational decisions are made. Many preliminary choices are often based on irrational motives (Habermas, 1972) or on semi-rational motives. Decisions based on experience, hypotheses and theories tend to be more rational than decisions based on intuition, assumptions, tradition and personal style. At many moments in the decision making process there is a combination of rational and irrational decision making. The range can vary from total rationality to total irrationality.

Knowledge, methods and problem handling techniques are needed for handling a problem. Problem handling techniques and knowledge will be discussed here, beginning with three rational problem handling techniques: trial and error, algorithm and heuristic.

Skinner's (1953) name is associated with research on problem handling by trial and error combined with reinforcement[98]. Problem handling by trial and error can be completely at random but can also be conducted within a certain handling space and with the exclusion of certain fields. At random, trial and error search will often be limited by commonsense knowledge and knowledge based on theory and experience. Commonsense knowledge and knowledge based on experience will often narrow the handling space[99]. As we said earlier, using this kind of method for handling complex interdisciplinary societal problems would not be advisable. Since it can lead to rather negative effects concerning people, loss of money and time. This way of problem handling is mostly applied when there is no other way of predicting the effect of an intervention. Among the six theories of decision making distinguished by Rosenthal (1984)[100], there are certain aspects of trial and error in the mixed scanning theory.

An algorithm is a strictly defined schema of steps that always and unambiguously leads to a solution[101]. Often a specific algorithm is directly connected with a problem. As in algebra, each kind of problem has its own algorithm. When we compare this with the six theories on decision making cited by Rosenthal[102], the rational synoptic theory and the satisfying alternative theory can be based on algorithms. According to Brookshear (1991), to understand how algorithms are discovered is to understand the problem-solving process.

Finding an algorithm to implement in a computer program can also be considered as problem solving. In Brookshear (1991, p. 140) this is described as[103]:

These ideas are based on the ideas of Polya (1957), a mathematician who defined steps of problem solving. His ideas are, among others, based on the ideas of Helmholz (1867) and Dewey (1933). The phases of problem-solving defined by Polya (1957) are:

This is ideally the way in which a software program for a problem can be developed. In practice these stages can be scrambled. In computer programming people sometimes begin by formulating strategies for problem solving before the problem itself is completely understood. If the strategies fail, the problem solver gets a better idea of what the problem looks like and may be able to use new strategies, a kind of trial and error process. To handle large programs like this, however, could easily lead to disaster.

Another way of discovering an algorithm is to work the problem backwards, just as a complete paper bird can be unfolded to see how it is constructed. Or a problem can be found that is analogous to the problem under consideration and has already been solved. Often large computer programs are programmed by dividing the program into sub-programs.

The idea behind viewing the problem in terms of sub-problems is that the different sub-problems are easier to solve than the overall problem. Dividing the program and from there on carrying out a stepwise refinement into smaller steps, is a top-down methodology from the general to the specific. This is in contrast to the bottom-up strategies from the specific to the general. In practice these two methods mostly complement each other.

According to Brookshear (1991), the procedure of decomposition into sub-problem is compatible with the concept of team programming. Now each team member can have a special task. Stepwise refinement is a major design methodology in data processing. We have, however already formulated our objections to this way of problem handling with regard to complex interdisciplinary societal problems[104].

A heuristic is a problem handling method that does not guarantee a solution but will enhance the possibility of a correct solution compared with the random use of trial and error techniques. Heuristics are also called 'rules-of-thumb'. Heuristics can be based on knowledge, on experience or theory. In a heuristic, the problem handling steps cannot be defined strictly in advance. In a heuristic there is not an exhaustive search into a problem space as Newell & Simon (1972) defined it, but there are small parts of solving steps that might lead to a solution.

There are general heuristics and domain related heuristics. To divide a large problem into sub-problems, solving the sub-problems and then putting the solved parts back together again is often called a general heuristic. A general heuristic can be applied to many kinds of problems. Domain related heuristics are heuristics that can only be used within a certain domain.

Snoek (1989) established, in his research on medical diagnosis in neurology, that both experienced and inexperienced physicians use many 'rules-of-thumb' in diagnosis, though experienced clinicians use more experienced-based heuristics based on intuition. Snoek argued for looking at clinical intuition not as an irrational and mystical concept but as a highly effective and efficient strategy based on specific knowledge in combination with experience.

In everyday life many decisions are made under time pressure, and because of this time pressure rules-of-thumb must be used. However, many rules-of-thumb are too simple and sometimes in contradiction with each other. Furthermore, it is not always easy to decide when rules-of thumb will be adequate or when more specific problem handling methods or knowledge is required. There is a continuous tension between knowledge and method. With using rules-of-thumb people tend to regard highly probable solutions as definitive and neglect highly uncertain solutions (Bree, 1989).

In comparison with the six theories on decision making described by Rosenthal (1984), the optimum theory is quite close to a heuristic.

There are no algorithms capable of solving complex interdisciplinary societal problems. However, within the long problem handling process there can be moments or exist areas where algorithms can be used. One can nevertheless develop certain guidelines for the process of problem handling, for instance, the distinction of phases. These guidelines can be considered as heuristics. They do not guarantee success, but phases serve as reflection points in the problem handling process.

3.7.4 General problem handling techniques and domain related problem handling techniques

Two kinds of handling techniques can be distinguished: domain related and general problem handling techniques. Problem handling techniques that are domain related include domain related algorithms and domain related heuristics. General problem handling techniques include, for instance, general heuristics and trial and error.

A general heuristic is a heuristic that can be used for many problems, and can therefor be very useful. For handling complex interdisciplinary societal problems we suggest at least two general heuristics: following the problem handling phases as described in this chapter and defining the problem before attempting interventions.

Breaking down a large problem into sub problems can also be regarded as a general heuristic, although it is probably not best to regard it so eventhough it is a heuristic that is often used. This kind of heuristic can only be applied to static problems. Static problems are not too complex problems in a static context. There is time to solve all the sub-problems and subsequently to put the sub-problems together again provided there are not too many interactions between the sub-problems.

For handling a problem, knowledge, methods and problem handling techniques are needed. Knowledge of the domain(s) involved in the problem will be discussed in this section.

There are different kinds and different levels of knowledge. Different kinds of knowledge include domain related knowledge and commonsense knowledge, context dependent and context independent knowledge. Then there are different levels of knowledge, the first and the second level of knowledge.

A domain is a part of a discipline, as developed in science[105]. Domain knowledge is knowledge a person acquires during education and professional work in that particular domain. Together with domain knowledge, a person learns domain related problem solving techniques. Domain knowledge may be divided into general domain knowledge and expert knowledge. By general domain knowledge we understand a lower level of knowledge than expert knowledge. Expert knowledge is highly specialized knowledge acquired over five to ten years of professional work in a particular field (Steels, 1987). General knowledge tends to be broad but not deep whereas expert knowledge tends to be more specific, deeper but restricted to a smaller part of the domain.

In addition to domain knowledge, there is commonsense knowledge. Commonsense knowledge can be defined as that knowledge of the world possessed by normally educated[106] persons. For example, fish swim in water; one can travel by different forms of public transport; the law of gravity works everywhere on this planet. Commonsense knowledge is necessary to handle domain related problems as well as interdisciplinary problems.

There is, of course, a large overlap between commonsense knowledge and domain related knowledge, the question depending on what view and at which level one considers the knowledge.

Many facts and rules in the world are time dependent and context dependent knowledge. Questions of how people live, can only be fruitfully answered if one includes a certain period, a certain kind of people and a certain place.

There is also universal time-invariant knowledge. However, this amount of knowledge is smaller than many researchers would like to believe, since even physical situations are context dependent. For instance an object moves differently under 'normal' situations, than in a vacuum or on the moon.

Although in some areas of science one assumes that the laws and rules have the same predictability everywhere, we think that many scientific laws and theories only work under exactly the same circumstances, background and other variables that influence the situation. However, the circumstances are seldom the same[107].

The first level of knowledge is knowledge as such. The second level is meta-cognitive knowledge. Meta-cognitive knowledge is knowledge about knowledge (Batenson, 1979; Klabbers, 1989). This level involves learning skills, content domain and commonsense knowledge. We can take learning as an example, to explain what we mean by these two levels of knowledge: Learning as such is learning how to row a boat or how to apply the rules of the law of gravity, or learning about how bees live. Meta-cognitive knowledge about learning, is learning about learning, reflecting the learning activity, how to learn a certain thing or a certain subject.

For problem handling, different kinds and different levels of knowledge are needed. Knowledge about facts and rules is first order or first level knowledge, whereas reflecting knowledge as such, reflecting knowledge by knowledge is called second order or second level knowledge.

In this chapter, human problem handling process has been discussed. The problem handling process can be divided into two sub-cycles. In the first sub-cycle the question to be answered is: "what does the problem look like?". In the second sub-cycle the questions can be answered: "which interventions can be implemented and how can these interventions be implemented?". The phases of the first sub-cycle of problem handling are becoming aware of the problem and forming a (vague) mental idea of the problem (1.1), extending the mental idea by hearing, thinking, reading, talking and asking questions about the problem (1.2), gathering data and forming hypotheses about the problem (1.3), and forming the conceptual model (1.4). In the first sub-cycle the emphasis is on thinking, discussing and asking questions. In describing the phases of problem handling many researchers skip the phase of awareness of the problem, or mention it only briefly. Being aware of a problem is very important, especially for complex societal interdisciplinary problems. One hopes that the sooner one is aware of the problem the greater the opportunity to tackle the problem in order to prevent the problem from becoming too severe.

In order to describe the mental process of thinking about a problem we use two concepts: the mental idea and the conceptual model. A mental idea is an often rather vague idea one has in mind about something. By thinking, discussing, data gathering and asking questions, this idea develops into the conceptual model of the problem. Data gathering is dictated by the mental idea of the problem. Research by Crombag (1984) on undefined problems indicates that people tend to analyze only a few hypotheses and from that point on only look for data that will confirm or falsify their hypotheses. The problem can be described by different models, embodying different languages.

In the conceptual model, the phenomena and the relations between them have to be carefully formulated. The conceptual model is formulated on the basis of theoretical ideas. The conceptual model can consist of a combination of models that together express the problem. Because complex interdisciplinary societal problems are often large problems in which many phenomena are involved, it is possible to select a scope of the problem, although this must be done only after a broad orientation on the whole problem at the macro aggregation level. The scope can be a demarcation in time or distance, a subsystem or even a domain. However, when suggesting interventions on the basis of the scope of the problem, these suggestions should be considered in the light of the whole problem. In the first sub-cycle the problem will be defined. Defining a problem carefully before handling a problem is very important. There is a close connection between changing a problem and the definition of a problem. Complex societal interdisciplinary problems, however, cannot be defined completely, since missing data and unknown connections between the phenomena, white spots and blind spots are bound to occur.

The second sub-cycle is a combined process of thinking and acting. Here the focus is on interventions. The phases in the second sub-cycle are: constructing the empirical model (2.1), defining the handling space (2.2), developing hypotheses and suggesting interventions (2.3), constructing and evaluating scenarios (2.4), implementing interventions and evaluating the effect (2.5). The second sub-cycle begins with constructing the empirical model followed by defining the handling space. The empirical model can be constructed on the basis of the conceptual model of the problem. The handling space is the space where one hopes to find some possible answers to (some parts of ) the problem. The handling space limits the range of possible changes. The space available to operate in can be described by four levels of constraints. Levels one to three range from rather restricted, kept within the boundaries of the contemporary situation to a situation where everything is different, yet still within human possibilities. The fourth level of constraints is an imaginary world. In addition to levels of constraints there are also different kinds of constraints.

In the change from conceptual model to empirical model we discuss the extent to which a complex interdisciplinary societal problem can be modeled. This discussion is based on the second research question. We concluded that a scenario of a complex interdisciplinary societal problem includes reliable knowledge and data, but also considerable uncertainty. Scenarios of complex interdisciplinary societal problems, based on system dynamic models, should be used with great caution.

The hypotheses and interventions are then discussed. The phases of the first sub-cycle of the problem handling process of the problems dealt with in cognitive psychology are similar to those of complex interdisciplinary societal problems. Although there are some similarities in the second sub-cycle, defining the handling space, suggesting interventions and implementing and evaluating the interventions, in most problems dealt with in cognitive psychology the interventions do not have to be implemented in real life.

The problem handling process is not as rational as one often thinks or hopes. Rational behavior is only a part of the handling process. The problem handling process is often a mixed rational and emotional process (Frijda, 1988).

In problem handling, different problem handling techniques, different kinds of knowledge and different levels of knowledge can be employed. With regards to problem handling techniques, a distinction can be made between an algorithm, which is a problem solving method that will, in most cases, guarantee a solution and a heuristic which is based on theoretical ideas that will increase the chance of achieving a satisfactory solution to the problem, although this cannot be guaranteed.

There are different kinds and different levels of knowledge. Different kinds of knowledge include, for instance, domain knowledge and commonsense knowledge; while the first and second level of knowledge comprise knowledge about facts and rules, and knowledge about knowledge respectively.

Given that the kinds of problems we focus on are very complex and that it is often not possible to know all the phenomena or all the knowledge about the phenomena and their relations and the data, it is not possible to define the problem completely. The fact thathumans have limited cognitive skills with regard to problem handling, combined with a changing and mixed use of rational and irrational decision making, makes it clear that an optimal handling of complex interdisciplinary societal problems is very difficult. Nevertheless we will try to find some ways to support the process of problem handling. The special character of complex interdisciplinary societal problems make it necessary to develop special methods that can guide the problem handling process. In chapters seven and eight we will describe a method that can support the problem handling process.

In answering the question of which tools can support the problem handling process we focus mainly on the computer, which in the last half century has become an important tool in problem handling. In chapters four, five and six, we discuss some aspects of computer support for problem handling. In chapter four we focus on the computer as a problem solver. This is a research object of Artificial Intelligence. In doing this we reflect on some aspects of knowledge based systems and general problem solvers. In chapter five we discuss some knowledge based systems that have been developed for handling real life problems. In chapter six we discuss some aspects of the computer as a tool to assist the humans in problem handling.

[1] The way the phases are described here is more or less an ideal theoretical problem handling. In practice, the phases are often not so strictly divided. In practice, the order of several phases may be inverted or one may jump from one phase to another and back again, even from the first sub-cycle to the second without completing the first sub-cycle. However, as described in chapter seven, in order to handle the problem optimally one should not reflect on the activities belonging to a higher level phase before the lower level has been performed.

[2] As described in chapter eight, the phases are iteratively traversed several times at different conceptual levels.

[3] See section 3.5, for instance, Van Strien (1986).

[4] An example of problem finding where not knowing may be preferable to knowing is having an incurable cancer for which there is no intervention that will diminish the disease. A discussion on this issue is described in 'De Aansprekers' by Maarten 't Hart (M. 't Hart, 1979).

[5] See the definition of problem solving given by Newell & Simon (1972) in section 2.1.

[6] See section 2.9.2.

[7] The literature discussed is mainly taken from the seventies and the eighties. More recent literature shows that these ideas and terms are still being used. Not much concerning the topic has changed since.

[8] He uses the Dutch word: informatie-versmalling.

[9] He uses the Dutch word: informatie-uitbreiding.

[10]Since intuition occurs, according to Snoek, more with experienced experts than with students, we believe that what is called intuition is mainly a combination of expert knowledge and experience. However, it is often called intuition because it is hard to reconstruct the thinking process and because the thinking process goes very fast. Snoek considers the use of intuition a very effective and efficient method of thinking. Selz (1922) also refers to intuition. See section 2.9.1.

[11] Reflecting on a problem does not always mean that the problem becomes more clear. A problem that looked rather simple at first can become more complicated the more a person reflects on it.

[12] Also called internal memory or short term memory.

[13] Crombag not only uses the computer as a metaphor for human thinking, but actually compares the human memory with the memory of the computer with regard to direct access of hypotheses and data. In comparison with the computer memory, the human memory is far more limited in its recall of facts.

[14] Models are used in many disciplines to represent issues.

[15] Translation by the author.

[16] This is important, as we will see later, for making a model of a complex interdisciplinary societal problem. The content of the models of the same complex interdisciplinary societal problem can vary a great deal. It is difficult to prove whether the model is a correct or incorrect model.

[17] Translation by the author.

[18] When we refer to theories, hypotheses, assumptions, experiences, and intuition in general, we will use, for the convenience of the reader, the term theoretical ideas.

[19] Selecting a theory is more or less arbitrary, depending on a person's scientific and/or political view, and on a person's belief, depending on the methodology and the discipline (Kuhn, 1970).

[20] See section 3.4.4.

[21] See section 3.4.1.2.

[22] Often called formulae.

[23] See chapter seven.

[24] See section 3.4.4.

[25] See section 3.4.4.

[26] White spots indicate that some areas that one is aware of, are not explored yet. This indicates the gaps in the knowledge.

[27] Blind spots indicate something that is basically known, but has been unintentionally omitted from the idea.

[28] See for further discussion on simulation models section 3.4.4 and section 7.4.

[29] In a model we can call the phenomena variables. A variable gives information about (a part of) a phenomenon involved in the problem.

[30] See section 3.4.4.

[31] See phases of problem handling in section 3.2.

[32] Chaos theory shows that uncertainty is more than a lack of knowledge.

[33] Cases in the field of a banking system, a library system and a transport system.

[34] An entity type is a data category. A person can be an entity type in a database of an organization with the attributes of name, salary, age etc.

[35] For instance, in the case of pollution, the macro aggregation level might be the earth, the meso aggregation level Europe, the micro aggregation level a country. Viewed from the perspective of a country, the macro level would be the country itself, the meso level a town and the micro level a person.

[36] Examples are the food distribution problem in the world, the quality of water, the refugee problem and international trade.

[37] See section 3.4.4.2.

[38] This should not be confused with defining a complex interdisciplinary societal problem as a domain problem.

[39] See section 3.4.3.1.

[40] We believe that one of the reasons for treating a problem sub-optimally is that only a part of the problem is handled instead of the whole problem. Interventions based only on a partial view of the problem can easily lead to pseudo solutions. An example of this is that the Ministry of Education in the Netherlands advised school career advisers to advise girls to study mathematics and technology instead of, for instance, languages. The advice did not have the positive effect the Ministry had hoped for. One of the reasons for this is that only a small part of the problem was handled while the main reason, the position of women in society, was left untouched. In order to handle this problem the cause and not one of the symptoms has to be changed (Witte, 1994).

[41] See figure 5

[42] Interventions are not always needed in order to reach the goal of the problem handling process. Sometimes conclusions can be directly drawn based on the empirical model. In psychology and in psychiatry the goal of the problem handling is sometimes only to perform a diagnosis.

[43]For instance: once a problem is defined as a legal problem, it is implied that a legal practitioner will 'solve' the problem. Only occasionally when new problems arise is there a consultation of experts from other domains. In most of the cases experts from other fields seldom reflect on this problem again.

[44] Translation by the author.

[45] Translation by the author.

[46] See also section 3.2. The main focus of this study is on defining the problem, the first sub-cycle of the problem handling process. Therefore only a few issues of the phases of the second cycle will be discussed.

[47] A homonym is the same word refering to different entities.

[48] Redundancy is the same information at more than one place in the information system.

[49] An example of this are the fast changing circumstances in the East-European countries in the period of 1989-1991.

[50] Military data, data about firms and plants, politically sensitive data etc.

[51] This is often the case with urgent problems as with sudden disasters such as typhoons, aeroplane crashes or political riots or coups (Rosenthal, 1984).

[52] See section 3.4.4.

[53] The paradigm of chaos theory focuses on uncertainty in dynamic systems, on unpredictability and uncertainty in quite different objects. Chaos theory is a generic term for theoretical ideas and models used by researchers from different fields focusing on the changes in systems where periods of predictable change are alternated by periods of unpredictable change. Order alternates with periods of disorder and chaotic changes. Chaos theory is applied in many disciplines, and a paradigm sometimes used in an interdisciplinary manner. See for further discussion on this subject section 3.4.4.3.

[54] We will consider also this problem in chapter nine.

[55] The desired situation can, for instance, be reorganization of the institute or diminishing the discharge of chemical plants.

[56] We do not use the term problem space to avoid the idea that the goal, the operations and the intermittent steps are already known and that the solution of the problem can be found within the problem space. It is possible, however, that the desired situation is not clear, or the goals are in conflict with each other, or that the kind of operations and tools are not clear. This is the reason we avoid the term problem space.

[57] See section 3.6.

[58] See for instance the situation in France of the revolution of 1789.

[59] 'Unfreezing' means inviting people to include, as a thought experiment, a higher level of constraints. This can be done in order to stimulate people to think about quite new situations, to realize that the present situation is also constructed by people and as a consequence is not rigidly determined. The hope is that people will come up with quite new and creative ideas for changing the problem.

[60] A contemporary example of this is the Woody Allen film "The Purple Rose of Cairo" in which a movie star walks out of the screen and becomes an artificially 'real' person (Allen, 1985) another example is the television serie Superman.

[61] Although this level can have some effect in real life. In Colombia a kind of 'Superman' calling himself Superbario supports poor people in their struggle to get better lives. Commercials also take advantage of the fantasy that using a certain article leads to a desired situation. Eating certain slimming products will make you look like the beautiful, slim, young lady on the picture, and smoking cigarettes makes you a real 'he-man'.

[62] Time is an important constraint in urgent problems, like riots and disasters (see Rosenthal, 1984).

[63] See section 3.3.4.

[64] See section 3.7.1.

[65] This is an often followed research method in (bio) chemical research, see section 3.7.1. However, for complex interdisciplinary societal problems implementing interventions, we strongly recommend to evaluate them first before trying them in practice.

[66] Rijksinstituut voor Volksgezondheid en Milieuhygiene (a Dutch semi-governmental institute).

[67] Nederlands Centrum voor Geestelijke Volksgezondheid (an independent Dutch institute dealing with information, research and advice in the field of public mental health and public mental health care).

[68] See also discussion section 7.4.

[69] This criticism comes close to our critique as it is defined in the theory of complex interdisciplinary societal problems, see section 3.4.4.1.

[70] See section 3.3.5.

[71] For instance, institutes such as the National Institute of Public Health and Environmental Protection (RIVM) and the Research Center of Public Mental Health Care (NcGv).

[72] See chapter eight.

[73] Chaos refers here to a certain kind of unpredictability. This differs from the commonsense use of the word, which means disordered.

[74] Translation by the author.

[75] See for a historic overview on the development of chaos theory Gleick (1989, p. 11 - 54) and Verhulst (1990, p. 15 - 33).

[76] The mathematician Penrose (1989) wonders what the reason is for this indefiniteness of chaotic functions. According to Penrose it can be found in the incompletion theorem of Gödel, which proves that some things are undecidable.

The question whether a certain point belongs to the values of a chaotic function or not is undecidable. One can never say with certainty that certain elements do not belong to the chaotic set.

[77] This description will be continued in chapter nine.

[78] For example the Delphi oracle.

[79] However, making scenarios one should take some precautions as far as this is possible based on the criticisms formulated against of using system dynamic modeling for future exploration.

[80] A null-option is that no explicit elaborate interventions are made. The situations will remain the same as they always have been (Bots, Sol & Thissen, 1992). Scenarios based on the null-option can also be used to explore the relation between the variables in the model (Van Dijkum, 1992).

[81] In western science, predictions are often based on extrapolation or regression analysis. In extrapolation the future development of the phenomenon will be based on the extrapolation of the development of the phenomenon. The values range within certain limits (minimum and maximum). In extrapolation, the future is predicted on the basis of an analysis of the data of the previous periods including predicted changes. With regression analysis future developments are forecasted based on the assumption that present causes have future consequences in the same way that past causes have been followed by present consequences.

[82] In this we believe that predictions based on system dynamic modeling, if done according to the method we suggest in chapter eight, increase the chance of a better prediction model of future explorations than models based on regression analysis. However, this statement should be supported by scientific research.

[83] Even small changes can encounter much resistance. New changes in society or in organizations should be carefully prepared and guided. See also some examples of implementing knowledge based systems in real life in chapter five (section 5.4, the Insurance Company).

[84] An example of this is curing leprosy throughout the world. Knowing how to cure the disease is different from actually solving the leprosy problem.

[85] Especially with productive ways of thinking, there is a distinction between discovering the problem handling method and the application - application in the way of arranging, testing and control.

[86] This was often the case with automation, starting with problem analyzing, working on building a computer system for two years and realizing that at the moment of implementation, the system no longer fitted the contemporary situation.

[87] See chapter one.

[88] Translation by the author.

[89] One tries something, sees the effect, and when the effect does not meet the desired situation, one goes back and reconsiders the problem, this time with the new knowledge.

[90] Translation by the author.

[91] Translation by the author.

[92] See our comment earlier in this chapter. However not all the phases have to be worked through, one can also stop and leave the problem as it is.

[93] At least the phases 1.3 to 2.5. For instance, in discussing the empirical model it can happen that one realizes that there are other relations between the variables than has been supposed. In that case it is sensible to go back to the former phase of formulating hypotheses about the relations between the phenomena, going back to phase 1.3.

[94] See also section 3.7.3. See also the discussion in section 3.4 on hypotheses.

[95] Intuition and experience also play a role in the description of the problem handling activities of neurology experts (see section 2.9.1, Snoek, 1989).

[96] Later also published in Simon (1979).

[97] See section 3.4.2.1.

[98] The coincidental connection between action and positive result can be strengthened by reinforcement.

[99] Trial and error is a commonly used research technique in medical research and in chemical research in order to find an answer to a problem.

[100] See section 3.6.1.

[101]This was/is the leading opinion. Although chaos theory shows that even when an algorithm is defined, in some cases there cannot be an unambiguous solution.

[102] See section 3.6.1.

[103] See also the phases of Van Strien in section 3.5.

[104] See section 2.9.5.

[105] See section 1.2.

[106] We are aware that the concept 'normally educated' is vague and tendentious. By 'normally educated' we mean what can be expected in relation to function, level of education and age (child or adult) according to his or her culture.

[107] An example of the misconception that a situation remains constant can be seen in the early days of automation. A frequent mistake in building a large database for an organization was to assume that the organizations would not change during the one to two years of development. At the time the database was finished the organization had changed so much that the information in the database did not meet contemporary needs.

Chair Operational Research Euro Working Group

Complex Societal Problems