Você está na página 1de 19

Generic Metamodelling

M.I.Yolles
Liverpool John Moores University Business School
Systemist (Dec. 1994)
June 1994

Abstract

It is important to provide structure and guidance to not only model building activities, but also
metamodelling building. Metamodels offer a structured guided approach to the modelling of
situations. Meta-metamodels can identify something about metamodels themselves. A generic
form of metamodel is proposed capable of providing a systematic approach to the selection of
metamodels. The ideas of structuring the domain of metamodels itself is a topic in its own
right.

1. Introduction

General Systems theory, or simply Systems, has been successfully applied to the modelling of
situations, in particular organisations and the identification and solution of their problems,
including the impact of change. It has integrated work pioneered by authors like Bertalanffi
[1968], Simon [1962], and Checkland [1981]. Systems is effective in the way it can describe
situations, formulate problems and define approaches towards their solution. This has enabled
it to permeate into a variety of human related subject domains.

The systems approach can be used very successfully in complex environments. It is, for
example, able to generate partitioned views of an organisation typically from a top down
approach. In this way the models it can generate are of reduced complexity. Systems also
provides relativistic modelling power to an investigator, enabling flexibility in the way a
situation is viewed. It is particularly powerful as a set of analytical and modelling tools.

It is often appropriate to offer guidance to build models for complex environments. This can
frequently be found as a set of rules or guidelines about what modelling or implementation tools
to use. A cohesive set of guidelines are themselves a model, and may be termed a metamodel.
A metamodel involves the modelling of the modelling processes themselves. It gives consistent
guidance and structured reasoning (call this homogeneous guidance) to situations that may be
messy [Ackoff, 1981].

The need for metamodels is made clear from simply noting that they frequently appear in
different domains of academic activity. More common examples of metamodels are Simon's
[1960] decision cycle from Management Science, the Systems Development Life Cycle [Davis,
Olson, 1984] for the development of computer information systems, the Organisational
Development approach used to solve small group problems [Pugh, 1984], Checkland's Soft
Systems Methodology use to solve organisational problems [Checkland, 1980; Checkland,
Scholes, 1990], the Conflict Modelling Cycle [Yolles, 1994, 1992, 1992a] to examine problems

1
of conflict in large social groups, and the Kolb cycle used to explain the process of learning
[Kolb, 1974]. There are a variety of other metamodelling approaches [Mayon-White, 1993,
p.133]. Some relate to similar domains of application, but vary in a variety of ways including
the philosophical approach taken, and involvement of people and their needs.

When students are faced with a multitude of metamodelling approaches from which they must
make a choice, they are easily confused by which one might be appropriate. The difficulty lies
in the fact that while metamodels enable investigators to apply modelling methods to situations
in a systematic way, there is nothing to help the student make a choice. That is, there is no
homogeneous guidance about what metamodel to select under different circumstances. Such
guidance represents a meta situation, and so we are forced to note the term meta-metamodel that
describes it. Meta-metamodels provide a systematic modelling view of the world of
metamodels itself.

The meta-metamodel is defined by identifying two elements: (a) a generic metamodel, and (b) a
set of constraints that can direct the application of the generic form. These operate by supposing
that different metamodels may be defined in terms of a set of constraints which bound the way
in which they can access tools. Thus, distinct metamodels may be differently bounded by the
constraints, and these constraints must necessarily be variable. Differences in a matamodel
boundary are determined by the following factors:

1. the philosophy that it adopts


2. the degree of softness
3. the level of structure
4. the degree of uncertainty that it assumes

In this paper a generic metamodel shall be defined, and the above concepts explored. In
addition, we shall mention some of the modelling tools that are typically used in different
metamodelling approaches. A generic metamodel must be flexible, and should therefore be
able to include new tools applied to old problems. As a demonstration that models not normally
considered part of a metamodel can be integrated into the generic form, we shall include two
modelling approaches from cybernetics: Double Loop Learning and the Viable Systems Model.

2. Constraining the Generic Metamodel

A number of constraints have been offered which can define the boundaries of particular
metamodels. Some of these constraints will be expressed as variables that can act on the space
of application of a generic metamodel. Together they enable the generic form to be applied to
different situations in different ways, either predefined according to well known metamodels, or
in a unique way. This enables a continuity of approach to develop when deciding upon how to
apply specified modelling ideas to particular situations. We now consider each of the factors 1-
4 in the above list.

2.1 Philosophy

Briefly stated, the philosophy of approach to modelling relates to its orientation. The modeller
will have a weltanschauung or "world view", a term used by Mannheim [1964] in his

2
examination of the sociology of knowledge and originating with Weber [1934]. It was later
used by Checkland [1980] as one of the cornerstones of his own systems methodology directed
at solving human activity problems. The use of the word by Checkland can be defined as "The
perspective of a situation that has been assumed...i.e. how it is regarded from a particular
(explicit) viewpoint; sometimes described as the assumptions made about the system."
[Patching, 1990, p282].

Consequently, we can see that weltanschauung is currently used in systems as the perspective
and set of assumptions associated with a situation. The perspective can change within a group
for a particular situation, and between groups for a given situation. The set of assumptions has
also traditionally been seen as the propositions that relate to a situation. These may consist of
statements that are commonly accepted and do not require proof or demonstration (axioms).
Propositions which are not axioms require proof or demonstration.

In the original literature, weltanschauung is the property of a group of people who take a
common theoretical or practical approach. It represents a structured community view of a
subject matter, a community which may itself be small. Relating to this concept, Kuhn [1970]
used the term paradigm. A paradigm may be seen as being contained within the
weltanschauung, and represents "the set of views that the members of a scientific community
share" [Kuhn, 1970, p.176]. Conversely, according to Kuhn, those who represent a scientific
community by necessity share a paradigm. The nature of the paradigm is that it "governs, in the
first instance, not a subject matter, but rather a group of practitioners" [Kuhn, 1970, p180].

The term paradigm involves four dimensions of common thought:

* common symbolic generalisations


* shared commitment to belief in particular models or views
* shared values, for instance in the nature of outcome evaluation
* shared commitments of exemplars (concrete problem solutions)

From this, it can be seen that different paradigms govern the way in which people build and
apply models. Different approaches thus occur because different paradigms operate within the
different groups. The question now arises as to how we can differentiate one paradigm from
another. One suggestion is that a set of variables is established which point the paradigm
directly at the metamodel. This shall be considered briefly in the following subsections.

In summary, the concept of the paradigm may be considered to be very appropriate in describing
the way in which groups of people see a situation and model it. It thus becomes clear that
paradigm becomes more appropriate to the contextual description of meta-metamodels than the
term originally suggested, philosophy.

2.2 Softness

Models or processes can be classified as having properties which are hard or soft. This is
defined on a hard to soft continuum representing a continuous qualitatively scaled axis. Thus,
softness operates as a variable. In entities with hard properties things tend to dominate, that is
properties can be objectively defined, measured, or assessed in some way that does not depend

3
on personal values.

Soft properties on the other hand, are relative to people. They cannot be measured objectively.
Personal values, opinions, tastes, ethical views, or weltanschauung are examples. It is people
and their psychological needs that dominate.

Some methodological approaches are harder than others. For example, in relatively hard
Systems Engineering it is the norm to model and test a solution before implementing it. In the
quite soft approach of Organisational Development the only way to test a model is to experience
it. The softness of a methodology therefore relates to the degree of people involvement and
their personal values. Thus, the methodology embedded in the Systems Development Life
Cycle (section 3) has a low level of softness, while Organisational Development has a high level
of softness.

2.3 Structure

Relationships between things (people, objects, or processes) may be highly (or well) structured,
unstructured (or illstructured), or semistructured. As with softness, the highly structured to
unstructured continuum may be considered to define an axis of variability. Semistructure rests
somewhere between the two poles.

In groups of things which are highly structured, the things and their mutual relationships are
known to exist. Unstructured situations involve things that do not have a relationship.
Structured situations may appear to be unstructured, if they involve things which have
unpredeclared or invisible mutual relationships. We can thus distinguish between visible and
invisible structure, a concept which is relative to the individual who is looking. This distinction
can be made objective by instead referring to deep and surface structure. Situations which
appear to be unstructured at the surface level may have a deep structure when examined more
closely or in a different way [Keen, Scott Morton, 1976, p94].

A semistructured situation exists when neither a highly structured nor unstructured situation is
found. Thus, a decision making process involving well known information about a
manufacturing process and unpredetermined ideas about where the process should be directed,
would be semistructured. It may be noted that the concept of semistructured processes is
important to the field of Decision Support Systems theory [Keen, Scott Morton, 1976; Alter,
1980].

2.4 Uncertainty

The purpose of organisation and control is to reduce uncertainty regarding the task to be
performed, and how and when it will be performed. There may for example be a measure of
uncertainty about what structural elements exist in a situation, or what their relationships are.
There may be uncertainty about what constitutes the paradigm of a problem.

According to Harry [1994], we can define each of these terms specifically as:-
Certainty: where each choice of action is linked with only one particular outcome.
Risk: where each choice of action may result in one of several identified possible outcomes.

4
Uncertainty: where the possible outcomes resulting from each choice of action are not
necessarily known, and in any case we cannot assign probabilities to them.

We can therefore conceive of a certainty-uncertainty continuum defining an axis of variability,


where risk lies somewhere on that axis. Risk changes with uncertainty, thus as something
becomes more risky it becomes more uncertain. Risk is often dealt with quantitatively, but in
the context here it is more appropriate to consider it only in a qualitative context.

It must be stressed that uncertainty is independent of softness, and structure. This does not
mean that there may be an indirect relationship between them, for instance in respect of the way
in which they are each measured. The nature of measurement will vary between qualitativeness
and quantitativeness. Measurement, like softness, has a degrees of human and thing
involvement. It is therefore dependent upon softness. Qualitative measurement is soft, but its
nature may also be unstructured and uncertain which qualifies the qualitative condition. For any
predefined degree of softness, there can also be varing amounts of structure and uncertainty
which qualifies the way in which measurement is perceived, undertaken, or used.

Measurement must, therefore, be seen as a variable dependent upon our three variables. A
measure of uncertainty may be easier to make in different circumstances. In a hard situation, the
degree of uncertainty is estimated more easily than in a soft situation since it is easier to take
objective measures about things as opposed to human values. Uncertainty will also be more
easily estimated in a well structured situation since we can then account for all the elements that
enter the situation and their known relationships.

2.5 Generating a Paradigm Space

Our interest lies in the modelling of iterative metamodels. Paradigms provide statements of
their nature, as defined by a group of practitioners. This statement requires constraints that are
provided by bounding variables. These we have identified as softness, structure, and
uncertainty. It is axiomatic that the bounding variables are necessary to define a paradigm in
respect of any iterative metamodel. Whether they are sufficient has not been proved. The
variables define a 3 dimensional space which we shall refer to as a paradigm space. It
establishes a bounded region which relates to the practioners' perceptions of the situations being
examined.

3. The Generic Metamodel

The focus of this paper lies in interative metamodels concerned with human activity situations
that may be examined as systems. There are three primary elements in the examination of any
situation. The three phases depicted in figure 1 are Analysis, Synthesis, and Choice. Analysis is
the breaking down of something into its components, including its context, the identification of
its structures, and its nature. Synthesis is the building up of a set of components into a coherent
picture, from the integration of ideas derived from the analysis, to the construction of the
prerequisites for a model. Choice is selection and/or implementation of something. There is
nothing radical or new in this proposition. In fact validation of the proposition is explored in
this paper in the next subsection.

5
┌─────────────────────────┐
│ │
Analysis Choice
 
│ │
└──────── Synthesis ───────┘

Figure 1: Generic Metamodel Expressed as an Iterative Cycle

If a generic metamodel is to model a given metamodel, then a paradigm must first be


established. Values of the variables are estimated in some way, and the boundedness of the
metamodel thus defined. Since the generic metamodel is positioned in a continuous paradigm
space, the metamodel being modelled is theoretically capable of being precisely defined.
However, only discrete qualitative values are normally used for the variables defining the space.
The region of the paradigm in the paradigm space will therefore be inprecise. After the use of
the term fuzzy in set theory, we shall refer to this as a fuzzy region with fuzzy boundaries. The
fuzzy region has associated with it a region of modelling tools accessible by the metamodels
that operate from this space.

Some paradigms may insist on a restricted access to modelling tools. If one can place two
metamodel fuzzy regions coincidently in a paradigm space, then in principle both will have
potential access to the same space of tools. If one paradigm unaturally restricts the size of
modelling base available to it, then the restriction for access would have to be explicitly defined.

The fuzzy region can be shifted by re-adjusting the variable values. This enables iterations to
occur using tools appropriate to different paradigms. In doing this, it is possible to generate a
different view of the situation. As this occurs the objectives of each metamodel application
changes and will represent different aspects of the same situation to enrich its explanation. For
example, one paradigm might model a solution to a problem situation, and another might model
its implementation. In another example, a study of a situation might examine structures and
processes, and a shift in the paradigm space might enable examination of a subset of the same
situation from a totally different viewpoint, say in respect of control and the structural
relationship of the predefined processes themselves.

Applying the Generic Metamodel

Let us now briefly consider some examples of metamodels, and see how they might relate to our
generic metamodel.

Simon [1960] was a major contributor to management thought. His concern lay in the
development of a decision science. In order to do this he expressed the prevalent ideas on
modelling development under goal seeking behaviour. Its three phases are Intelligence, Design,
and Choice which can be iterated through to progress a problem. The cycle provides guidance
in modelling decision problems which tells us that problem domains must be properly
examined, options identified, and models generated and applied to the domains. Each of

6
Simon's phases maps directly into the generic metamodel's Analysis, Synthesis, and Choice.
The Simon cycle has provided guidance to the process of decision making. However, it does
not provide model builders with guidance on model selection or approaches, and it provides a
paradigm space which is perhaps too loosely bounded. This is because it does not restrict its
application to any particular ristriction of uncertainty, structure, or softness. Neither, however,
does it recommend particular sets of tools.

A development of this cycle was suggested by Rubenstein and Haberstroh [1965], and a
variation designed for computer software developers was later produced [Sprague, 1986] in
order to tackle well structured problem domains. The latter hard approach is related to the
Systems Development Life Cycle (SDLC), and this type of approach has been used in situations
in which the analyst does not envisage the involvement of people while addressing the problem
domain. It therefore provides for a highly structured environment since everything is assumed
to be known and no people are involved in consideration of the needs. It also assumes certainty
in the information it has access to. To tackle more unstructured processes involving people a
different need arises.

Systems Intervention Strategy (SIS) [Mabey, 1990] operates with a paradigm that is
semistructured, relatively hard, and relatively uncertain. SIS involves the three phases
Diagnosisis, Design, and Implementation. These carry through the following steps:

Diagnosis
(1) Systems description
(2) Identification of objectives and constraints
(3) Formulation of measures for objectives
Design
(4) Generate objectives
(5) Model options (selectively)
Implementation
(6) Evaluate against measures
(7) Design implementation strategies
(8) Carry through.

The difference between this approach and that of the generic metamodel lies in the phases 2 and
3. Here, transformation to the generic model occurs by renaming Design as Synthesis, and
shifting step (5) to to the next phase, the new name becoming Choice.

One well known metamodel [Checkland, 1981; Checkland, Scholes, 1990] called Soft Systems
Methodology (SSM) has been used to solve unstructured small group dynamic change
problems. Like many of the metamodels that can be found, the SSM paradigm is closely
bounded demanding tight conformity to a particular approach in modelling, and suggests that
consistent with soft perceptions, models of situations cannot be tested but rather must be
explored with the participants of a situation. It is therefore very soft. It is also concerned with
uncertain situations.

Considering SSM in terms of the generic metamodel, the first phase Analysis involves domain
examination by covering his first four steps in the seven step metamodel. That is Analysis, Root

7
Definition, Conceptualisation, and Comparison. The second phase Synthesis relates to the
Definition stage of SSM. Choice is the third generic phase, and relates to Checkland's Selection
and Design phases.

Some authors are unconvinced by the tight boundedness of the SSM paradigm. Flood and
Jackson [1990] have defined their own paradigm offering what appears to be a less constraining
boundedness introducing the idea of a "metaphor" to represent an analogous concept to the
problem in hand. The metaphor thus helps to identify the context of a situation. The paradigm
provides a wide access to both hard and soft models. This approach, called Total Systems
Intervention (TSI), consists of a cycle of three phases: Creativity: use system metaphors as
organising structures; e.g. see an organisation as a machine (closed system), an organism (open
system), a brain (learning system), culture (norms, values), team (unitary political system),
coalition (pluralist political system), or prison (coercive political system). Outcome is the
dominant metaphor. Choice: select an intervention strategy or set of methodologies as
appropriate. Use any of the tools available from the hard-soft continuum of techniques. A
dominant technology may be found. Implementation: employs a particular system methodology
to translate the dominant vision of the organisation, its structure, and the general orientation
adopted to concerns and problems into specific proposals for change. In terms of the generic
metamodel it would seem that the following holds:- Analysis: part of TSI's Creativity (defining a
set of metaphors); Synthesis: part of TSI's Creativity (defining the dominant metaphor); Choice:
TSI's Choice (identifying the intervention strategy or set of methodologies) and Implementation.

Organisational Development (OD) is yet another approach that has been adopted to tackle
problems in organisations that are people related. Its paradigm represents a metamodel
approach which is very soft, unstructured and uncertain. OD has a cycle that includes the
following seven steps:

(1) Confrontation with environmental changes, problems opportunities;


(2) identification of implications for organisation;
(3) education to obtain understanding of implications for organisation;
(4) obtaining involvement in project (and relating to felt needs of participants);
(5) identification of targets for change;
(6) change and development activities;
(7) evaluation of project and programme in current environment and reinforcement.

In terms of the generic metamodel, the relationship must be a little more carefully represented.
It would seem that OD represents a two (generic) cycle event, and thus it is appropriate to
identify a unique representation that this is occurring. Suppose that for Analysis (A) the xth step
of the nth pass of the generic cycle is represented by the symbol CAnx, with equivalent
expressions for Synthesis (S) and Choice (C). The seven OD steps for the A,S,C generic phases
are:- CA11 CS12 CC13,4 CA25 CS25 CC26 CA37. OD is designed to structure an investigation of the
situation by using a set of tools designed for the purpose. These tools include the Pugh matrix,
Pugh rules and principles, Huse factors, and Kotter and Schlesinger theory [Mabey, 1990].

It may be inappropriate to apply the tools from one metamodel to another. Thus the application
of a feasibility study from the hard, certain, and structured SDLC would be inappropriate in
application to the soft, uncertain, unstructured OD. Since the weltanschauung of the two classes

8
are different, then necessarily the paradigms and thus the location of each in the paradigm space
must be different. If it were the case that the fuzzy region in the paradigm space of each
metamodel overlaps, then tools may be common. If both are independently bounded however,
then the tools are mutually exclusive.

4. Control and Evaluation

The power of a generic metamodel is that it can incorporate other modelling techniques which
are not normally considered in the context of metamodels. This can enhance the power of the
modelling technique itself, and it can more easily be compared with other metamodels. To do
this the first step must be to ensure that the models being examined relate to the paradigm space
defined for the situation.

In this paper we shall consider two modelling approaches. It has been recently said in private
communication by a colleague, Denis Adams, that they are related, but this has not been
demonstrated theoretically. The two approaches are Double Loop Learning, and the Viable
Systems Model. Both deal with control and evaluation, and this might well represent a second
level activity which could be attached to a generic metamodel. Our intention is now to consider
control and evaluation slightly more carefully in this section.

A system view of a situation conceives it as being bounded and having a set of participants and
defined entities which can operate together. This suggests that there must be flows between
each of them which will include information and resources. There is also a chain of operations
and communications between the defined entities within the boundary.

Many of the participants in the situation do not fully comprehend the nature of the exchanges
that occur, or the needs of each of the entities. The ability for the participants of a situation to
learn as a result of the processes that occur in it is essential.

4.1 Double Loop Learning

Figure 2 represents a conceptual learning loop that results from experiencing the interrelation
between the action an enterprise takes as the result of a decision, and the environment. The
internal loop is a representation of what is called Single Loop Learning (SLL), while the
incorporation of knowledge support for the purpose of problem diagnosis is referred to a Double
Loop Learning (DLL) [Argris, 1992; Hampden-Turner, 1990; Garfield, 1992]. Monitoring and
evaluation involves a data base, an index generator for evaluating the processes numerically,
comparison, and in the event of problems within the process, a mechanism for alerting the
participants of the situation. This alerting activity can occur through the generation of an
exception report which then passes on to the knowledge support component of the system.
Knowledge support involves research, a knowledge base which may be an explicit expert
system or implicit knowledge to the individuals involved in the process, and a decision event.

In decision science, it has been shown that it is often the case that the theoretical ideas about a
situation which a decision maker claims to use in action applied to a decision are different from
the theory actually applied. This is the difference between claimed and effected theory. The
term sometimes applied to effected theory is 'Theory-in-use'. In practice a researcher of decision

9
processes should always therefore check whether actions of the decision maker are consistent
with the espoused or claimed theory being applied to a decision.
┌───────────┐
┌──┤Environment├──┐
│ └───────────┘ │
│ ┌──────┐ │
└─>──┤ ├────>┘
┌─────────>──┤ ├────>─────────┐
│ ┌─────>──┤Action├────>────┐ │
│ │ └──────┘ │ │
│ │ ┌──────────┐ │ │
│ │ │Real time │ │ │
│ └───<──┤monitoring├─<─────┘ │
│ │ and │ │
│ │evaluation│ │
│ └──────────┘ │
│ ┌──────────┐ │
└────────<──┤ Knowledge├──<────────┘
│ support │
└──────────┘
Figure 2: Double Loop Learning

Decisions require access to a Knowledge Base. This is the existing knowledge contained in a
system. It may simply exist in peoples heads and/or may be fully recorded. The knowledge
may or may not be transferable.

SLL/DLL assumes that a decision maker uses some 'Theory-in-use' as a basis for decisions. The
explanation of each component in the learning loop system can be expressed in terms of the
generic metamodel. It starts at the choice part of the cycle just prior to entry to a second pass
around the cycle. Thus, the first phase of Analysis has been accomplished, that the Synthesis
phase has been achieved, and that a decision must now be made on the basis of choice. There
are consequences to this which need to be controlled, and these are identified in analysis and
synthesis. Thus:

Choice:
Decision: A selection between the alternative courses of action. In practice this could be the
result of a Means-End analysis where a model has been used concerning the relationships
between the Ends (the objectives to be satisfied) and the Means (the way by which the ends
have been achieved).
Instruction: Having made a decision, instructions are usually required to ensure that appropriate
actions are carried out to bring about the objectives of the decision.
Action: These are changes in the situation brought about to achieve the desired state of the
situation, including its outputs.
Analysis:
Measurement: These are the measures of the system state(s) which is being controlled to some
desired level or value.
Data: These are the values of the variable produced by the measurement.
Actual Values: These are the actual values of the variable that are to be compared with the
expected value.
Synthesis:
Expected Values: These are the expected values of the variable; they may be expressed as a
forecast, a prior estimate, or some desired value.
Comparison: This compares the actual and expected values and reports any significant

10
deviation. As such this converts raw data into information.
Forecast: This represents a process for providing the expected value. In practice it may not be
a forecast but an estimate, a desired value, etc.
Model: This is a representation of the situation including the control system components. It
will reflect the content of the Knowledge Base of the system
Knowledge Support leading to Diagnosis: The process of trying to explain and understand
unexpected deviation produced by Comparison. It will involve the current model and
Knowledge Base, but should be approached from a multi-perspective point of view which
generally provides the greatest chance of a fundamental improvement in the Theory-in-Use and
the accompanying model and Knowledge Base.

The ideas are totally in keeping with cybernetic principles, and relate to evaluation and control
aspects of the enterprise. It assumes a paradigm that is semistructured, relatively hard, and
uncertain.

They could also be connected with the ideas of Kolb [1974] who developed a learning cycle
offering perceptions about learning to learn. It suggests that the learning process passes from
observation (Analysis), to conceptualisation (Synthesis), to experimentation, and finally to
experience (Choice). Yet another learning related cycle is due to Crampes [Yolles, Pirani, 1991;
Yolles, Pirani, 1992]. The Crampes cycle is particularly appropriate for complex learning
material, and describes a learner in a learning process who first passing through a voyage of
discovery (Analysis) identifying the nature of the learning material and determining its form and
structure. The second stage is that of exploration (Synthesis), when the material is examined,
and its content, its meaning, its message, is explored. Evaluation (Choice) may be self
determined or objectively done by the system as well as subjectively by the learner, and could
occur for example through some form of testing, or by participation in simulations. The result
of evaluation will determine the success of the exploration. If evaluation demonstrates an
adequate result, then the cycle may end. Evaluation may lead to

a) a poor assessment which itself can encourage or demand a re-evaluation of the


learning material and continuation of the cycle, or
b) extended discovery, particularly in more complex learning domains which have
more than one level of perception or understanding.

4.2 The Viable Systems Model

One of the conceptual constructs of cybernetics is the Viable Systems Model (VSM) which can
usefully be used to enhance the understanding of process evaluation and control
implementation. This feature enables the improvement of the quality of entity processes to
occur through its ability to address the communication and control system of any viable
enterprise operating coherently as a system. A system is viable if it is capable of operating
successfully in achieving its objectives. A viable system is a system able to support adaptability
and change while maintaining the stability of its key operations.

VSM is normally placed in the context of control and evaluation of enterprises. It can be
embedded into DLL in order to enhance its diagnostic power [Yolles, Adams, Pirani, 1994].
The purposes of DLL is as a mechanism by which enterprises are able to learn about problems

11
involving processes under uncertainty. That is, not enough of the process itself can be evaluated
to be certain about what is happening. VSM, however, can be used as a framework for
modelling and diagnosing weaknesses in the communication and control system of the
enterprise.

VSM has its origins in Beer [1959; 1966], its structure and operational use in Beer [1979;
1981], and a methodology for use in Beer [1985]. The basic foundations in terms of feedback
thought have been clearly presented by Richardson [1991]. The basic model structure of VSM
is shown in figure 3. It conceives there to be 5 interacting sub-systems that are management
processes, together with the communication channels between them. It assumes a paradigm
boundary that is semistructured, relatively soft, and uncertain.

The enhancement of DLL by VSM can be more immediately seen by relating them through the
generic metamodel. Evaluation the organisational aspects of processes are undertaken in the
Synthesis phase. Analysis of the situation occurs in the Analysis phase in order to generate a
future orientated examination of the enterprise. Policy is determined within the Choice phase.

Synthesis:
Operations/implementation is a representation of what the system does and produces; it is
usually broken down into functional units.
Co-ordination tries to harmonise the culture and structure of the enterprise whilst also trying to
reduce chaos and introduce order. It amplifies the control capability to try to induce self
regulation into the implementation of operations.
Integration/control is in charge of the functional units of the system. It controls and monitors
what is going on. It is responsible for the implementation of policies, resource allocation, and
the control and monitoring of the implementation activities.
Analysis:
Futures/development acts as an internal and external eye of the enterprise gathering information
from the environment and the system itself.
Choice:
Policy provides the systematic capability to choose from the different problem situations or
opportunities thrown up by the environment.

┌────────┐
│ Policy │
└───┬────┘

▌▌▌▌▌▌▌▌▌▌▌▌▌ ┌───┴─────┐
▌ Future ▌──────────────────┤ Futures │
▌Environment▌ └───┬─────┘
▌▌▌▌▌▌▌▌▌▌▌▌▌ │
│ ┌──────┴───────┐
│ ┌───────────────────┤ Integration ├──────────────┐
│ │ └──────┬───────┘ │
│ │ │ │
│ │ │ │
┌─────┴───┴──┐ ╔═════╧══════╗ ┌──────┴───────┐
│ Operations │ ║ Management ╟────────┤ Coordination │
└─────┬──────┘ ╚════════════╝ └──────┬───────┘
└─────────────────────────────────────────────────────┘

Figure 3: Viable Systems Model of Management Processes and their


Communications Channels
It is clear, therefore, that a preliminary first pass around the generic metamodel could provide a
very useful understanding of the full structure and processes at work in the organisation, prior to
a deep study of the relationships between the processes themselves. Like DLL, VSM also

12
represents a second level of processes evaluation within a system, enhancing DLL through
compounding its constructs, and providing the ability to monitor, evaluate, and control dynamic
processes which are prone to unwarranted or inappropriate interference. DLL and VSM can be
compounded, as shown in figure 4. However, since their paradigm spaces vary (DLL is
ralatively hard, while VSM is relatively soft), then it would seem that the two approaches must
be applied independently, VSM following on from DLL or vise versa, depending upon where
you are in the examination of the situation.

In terms of VSM, it is feasible to examine the relative nature of the different processes at work
in a problem space coincident with the paradigm space. This can be done in the following way.
Under synthesis, consider that Operations, Coordination, Integration represent primary control
factors of an organisation. These can be evaluated as a problem space defined by softness,
structure, and certainty. This provides the evaluator of the control and evaluation system an
immediate view of the relative nature of the totality of process in operation, and this can
contribute to the way in which they are dealt with or related. The nature of the paradigm space
that is defined in this application of the generic metamodel would be examinable through the
table given in figure 5. It is feasible that graduations of the three variables could be chosen,
with hardness, highly structured and highly certain being represented by a 1, while softness,
unstructure, and uncertainty by a zero. The level of importance of particular processes or
activities may also be a factor to be taken account of, and this might be shown by arranging
them in order of their appearence in the cells of figure 5.

Model DLL VSM

Phase
Analysis Measurement Futures/development
Data
Actual Values

Synthesis Expected Values Operations/implementation


Comparison Co-ordination
Forecast Integration/control
Model
Knowledge Support & Diagnosis

Choice Decision Policy


Instruction
Action

Figure 4: Enhancing DLL with VSM

Clearly, there is more to the control and evaluation process using VSM than has been explored
in this paper. However, its operations through the generic metamodel here provides it with the
possibility of operating as a metamodel in its own right.

13
Operations Coordination Integration
Nature of cell entries Specify nature of Specify the Specify elements being
processes or activities coordination activities controlled and
that occur in a that occur in a monitored in a
situation situation situation
Softness
Hard

Soft
Structure
Highly structured

Unstructured
Certainty
Highly certain

Uncertain

0 ... Measurement 0 1 0 1 0 1
scale... 1

Figure 5: Problem space of the situation relating to the paradigm space

5. Conclusion

The development of the idea of a generic metamodel with the phases Analysis, Synthesis, and
Choice, has led to the introduction of a paradigm space that defines the boundaries in a fuzzy
way for different metamodels. Taken together, the generic metamodel and the paradigm space
define a meta-metamodel, a higher level of model process modelling.

Appropriate independent models can be incorporated into the generic metamodel, therefore
providing more modelling power. As a demonstration of this, both Double Loop Learning and
the Viable Systems Model have been examined. Defining VSM as part of the generic
metamodel offers a more integrated way of looking at the control and evaluation of a situation
already analysed. Typically, the generic form might be applied using a traditional systems
approach examining structure and process. However, a second iteration enables control and
evaluation to occur. In this case, the paradigm could feasibly be adjusted to enable the
metamodel to access an alternative set of tools able to address the nature of the existing primary
control elements. Alternatively, the control elements might themselves be adjusted in order to
improve control and evaluation itself. Here, improve probably relates to the systems being more
certain that it is operating more stably, so that the objective of control is closely adhered to.

14
No attempt has been made to specifically define boundaries within the paradigm space except in
general terms as suggested is possible. This is because this paper is initially exploratory. The
location of a metamodel in the paradigm space has been a side consequence, however, of
different intentions rather than an objective in itself. This paper has primarily attempted to
provide a broad structural view of metamodelling approaches which until now have offered
little non adhoc homogeneity.

7. References

Argyris, C, 1992, "The Executive Mind and Double Loop Learning", Org. Dynamics, Autumn,
pp5-22.

Ackoff, R.L., 1981, Creating the Corporate Future, Wiley, New York.

Alter, S.L., 1980, Decision Support Systems: Current Practice and Continuing Challenges.
Addison-Wesley.

Beer,S., 1959. Cybernetics and Management. English Universities Press. Also, 1966, Decision
and Control. Wiley. Also 1967, Cybernetics and Management. Wiley. Also, 1979, The Heart
of Enterprise. Wiley. Also, 1981, Brain of the Firm (2nd edition). Wiley. Also 1985,
Diagnosing The System for Organisations. Wiley.

Bertalanffi, L von, 1968, General Systems Theory. Braziller, N.Y.

Checkland, P., 1981, Systems Thinking Systems Practice. John Wiley & Son, Chichester.

Checkland, P., Scholes, 1990, Soft Systems Methodology in Action. John Wiley & Son,
Chichester.

Davis, Olsen, 1984. Management Information Systems, McGraw-Hill.

Garfield, C, 1992, Second to None: how our smartest companies put people first, Business One
Irwin.

Hampden-Turner, C., 1990, Charting the Corporate Mind: from dilema to strategy, Blackwell.

Keen, P.G.W., Scott Morton, M.S., 1976, Decision Support Systems: An organisational
Perspective. Addison-Wesley.

Kolb, D.A., et al, 1974. Organisational Psychology: An Experiential Approach. Englewood


Cliffs, New Jersey. Prentice-Hall.

Kuhn, T.S., 1970, The Structure of Scientific Revolutions. University of Chicago Press.

Mabey, C., 1990, Management of Change. Open Business School, course course text. Open
University, Milton Keynes.

15
Mannheim, K., 1964, Wissenssoziologie. Nenwied/Rhein, Luchterhand.

Mayon-White, W., 1994, Problem Solving in Small Groups. Managing Change, Ed. Mayon-
White, W., Mabey, C., Open University.

Pedler,M., Burgoyne,J., Boydell,T., 1991. The Learning Company. McGraw-Hill.

Patching, D., 1990, Practical Soft Systems Analysis. Pitman.

Pugh, D.S., 1984, Organisational Theory: Selected Readings. Penguin.

Porter, M.E., 1985. Competitive Advantage. Collier McMillan.

Richardson, G.P., 1991. Feedback Thought in Social Science and Systems Theory. University of
Pennsylvania Press.

Simon, H.,1960, The New Science of Management Decisions, Harper Bros., New York, 1960.

Spague Jr., R.H, Watson, H.J.,1986, Decision Support Systems. Prentice Hall.

Weber, M., 1934, Wissenschaftslehre. Tübingen.

Yolles, M., 1985, "Simulating Conflict Using Weibull Games." Modelling and Simulation, Ed.
A.Javor,IMACS; Elsevier Science Publishers (North Holland).

Yolles, M., 1987, "Modelling Conflict with Weibull Games.", Mathematical Modelling of
Conflict, Ed. J.Bennet, IMA, Heinamann Press.

Yolles, M., Pirani, M., 1991, Computer Aided Learning and the Pedagogic Paradigm. Contained
within: COMETT/VOLET Cb: Final report strand Cb-Contract year 1990/1991. Reference
number 90/2/5081/Cb, Brussels.

Yolles, M.I., Pirani, M., 1992, September. "Design for a Computer Based Remote Learning
System". 14th International Conference on Information Technology Interfaces. Pula, Croatia.

Yolles, M., 1992, The Conflict Modelling Cycle, J. Conflict Processes, 1,1,39-56

Yolles, M.I.,1992a, October. "Towards Simulation of the Conflict Modelling Cycle." Presented
at the Chinese Association of Modelling and Simulation conference in Beijing, October 20-23.

Yolles, M., 1994, The Generic Metamodel, the Conflict Modelling Cycle and Decision Support.
Presented at the Seatle conference of the American Association of Artificial Intelligence,
August.

Yolles, M., Adams, D. Pirani, M., 1994, Systems, Cybernetics, and the Management of
Conflict. J. Conflict Processes. Vol.2, No.1

16
3.2 Analysis

Analysis of a situation requires that it is defined, examined, and understood. It is concerned with
examining the situation for participants and conditions which relate to the resultant states that must be
defined and selected; data inputs are obtained, processed, and examined for clues that may identify
problems or concerns with gathering data, identifying objectives, diagnosing problems, validating data,
and structuring problems and environments. Once a situation has been defined as a system, we can
examine various aspects. For example, systems may be dynamically and structurally stable. Dynamic
stability relates to the movement of a system in its achievement of desired goals. Here, a desired goal
will be achieved as time passes if the system is dynamically stable. Necessarily, therefore, dynamic
stability is time related. If a system has reached a goal and adheres to it, then the system has achieved a
condition of equilibrium. Structural stability is more properly referred to as structural criticality. A
system which is close to criticality may react significantly to a small change in one of its parameters.
Under change, a system may pass through a critical condition, when its structural relationships alter. A
big change caused by a small event may show the system not to be stable.

The nature and context of the situation should be examined in general terms, and the environment in
which it operates. The type of problem space being examined which it contains should be identified.
Thus the problem space may be hard (quantifiable or single optimal solution) or soft (unclear solution),
and bounded or unbounded. Tools like mind maps, spay diagrams, force field analysis, relationship
diagrams may be appropriate.

The situation should be defined at various levels; system map techniques could for example be used, and
the nature and boundaries of all the entities defined. Systems methodology may be useful in identifying
domain structure and its description (e.g. boundary parameters, degree of boundary fuzziness, system
entropy); participant problem perspectives should be identified. General goals of the participants should
also be identified.

Problem definition and the movements that are made in solving it should be identified. This defines the
trajectory of the problem. The problem domain is the problem and its set of participants, parameters,
variables, and constraints. Clear definition of this can be difficult when there is sufficient complexity.
In reducing complexity one might: (a) examine the changes that may have invoked the problem, (b)
identify the problem boundaries and associated parameters, (c) examine how many problems exist and
whether they can be reduced to discrete sub-problems, (d) examine possible problem solving schedules.
Each participant in a situation is a player with a framework of perception, perspectives to the problem,
and decisions and actions taken which constitute a pathway through the domain. A player may have a
stake in the situation (as a stakeholder) or may not. The pathway will have a direction which, if
intended, represents the aim of the process, and identifies a set of vectors of movement. The trajectory
is the set of vectors taken with the resultant goals that may be achieved. The difference between an
intended and an actual trajectory is an indicator of how stable the situation is. Thus, if measures of
achievable and intended goals, aims and objectives can be made, a measure of stability can be found.
The rationale and feasibility of player trajectories should be evaluated in relation to earlier phase
elements. Techniques like objective trees, and multiple cause diagrams can be used.

Relationships should be established between entities within the system and its environment. Techniques
like influence diagrams can be used.

To undertake analysis, it is essential that participants and their influences are adequately understood.
Players are participants with goals, objectives, strategies, and an external environment with which they

17
interact. They have internal constraints as well as external ones, variables which include general
cultural attributes. This applies to all classes of player, whether they are loose collections of individuals,
enterprises, cultural groups, or nation states.

3.3 Synthesis

By synthesis is meant selecting, inventing, or developing possible options or decision scenarios. The
development of decision scenario alternatives requires an adequate knowledge of the problem area and
an ability to generate feasible alternatives.

The assumptions or propositions associated with a particular situation should be stated. This can be
important because participants in a situation operate within their own distinct frameworks; whether one
participant acts rationally may depend very much upon the nature of the framework, and it should be
possible to map from one framework to another coherently. This in essence defines a principle of
interactive relatively for the participants in the situation. Without the statement of suppositions or
propositions about the situation, it may be inappropriate to seek modelling tools.

This phase is often referred to as "design", since it addresses a plan of action that relates to the situation.
This is a result from analysis and is a construction of a view or set of views. In very hard situations,
this might mean the construction of a single solution to a prespecified problem. It is, however,
infrequent for situations of this type to occur. More typically, a range of options might be more
appropriately generated: these may include participation, and possible choices for the participants. This
may involve modelling interactive relationships between participants. Decision table techniques may be
appropriate to show possible options in a compact way, as might be other decision related approaches
like Pugh's matrix [1984]. However, the appropriateness of such tools will be dependent upon the
paradigm selected for the generic metamodel.

In general, this phase is concerned with the manipulation of data, quantifying objectives, generating
reports, generating alternative scenarios, establishing structural relationships and assigning uncertainties
or values to alternatives. If a set of alternatives are identified, they should be reduced to a useful set.
Those eliminated will typically be unuseful or unstable scenarios, so that the alternatives will be reduced
to a core set of option. In very soft applications, options may be reduced to the tactical ones available as
a core approach.

3.4 Choices

Modes of implementation define the way in which implementations are to occur. They are thus
themselves models.

Modelling choices may involve identifying/selecting models or modes of implementation capable of


representing feasible solution scenarios from those options available. It may include the evaluation of
model options and their ability to represent environments and decision scenarios, or to evaluate modes
of implementations. Examination of the consequences of modelling option in respect of a changing
environment might also be appropriate. It would be necessary to activate these models as a solution to
the problem. This will generate outcomes which may be possible problem outcomes. The models could
be validated and the outcomes evaluated in terms of the problem domain.

Modelling options and approaches should be identified; so too should methods for modelling perceived
and hypothetical situations which might also model the future; provide the choice of selecting model
alternatives; identification of model demands, constraints and perspectives should be made explicitly;

18
model propositions should be defined, as should stability mechanisms, and a propositional base
including any normative assumptions. Models should be evaluated, comparing selected models to the
real world situation, and identifying convergence between modelling options and real world situation.

Choices involves setting up a modelling technique or mode of implementation. Any additional


parameter estimation or variable estimation requirements based on historical evidence should be
Identified, data assembled, and process modelling components prepared.

Modelling results should be generated, or an activity implemented. In the case of hard modelling, for
instance stochastic processes, approaches like Monte Carlo simulation, Markov processes, or Weibull
games [Yolles, 1985; Yolles 1987] can be adopted, and perhaps compared. In soft problems this step
relates to starting the experience. Dynamic and structural stability of the synthesised system should be
investigated.

Models used should be validated, examining the selected output or implementation and this related to
observed events. In relatively hard situations this may well involve a quantitative approach, and in the
case of there being numerical outputs these must be interpreted qualitatively. A soft approach requires
checking that the progress of the experience is appropriate. A match between model outputs and
commonly agreed real world events will indicate the level of ability of the modelling approach.
This phase distinguishes the ability of each model to represent the situation and the constraints under
which it operates. Validation of a model only occurs if the modelling option evaluation has been
successful.

19

Você também pode gostar