cybernetic paradigm of grissinger

21
The cybernetic paradigm Gyula MEZEY Zrínyi Miklós National Defence University, Budapest, Hungary E-mail: [email protected] Risk-based decision-making in reducing security risks is based on practical approach, usually a combination of rational and cognitive theories. In risk- management analytic, cybernetic, cognitive paradigms can be applied. The Analytic Paradigm has the potential to clarify a decision, but cannot guarantee an improved decision-making. Under the Analytic Paradigm it is asserted, that someone or a group will actively select from among distinct options in a stable decision environment so as to achieve a set of goals. Recently widely recognized, that this assumption is farfetched from reality. The Cybernetic Paradigm has been increasingly in use for group decision support systems. This paradigm focuses onto uncertainty control and not optimization. Qualitative modelling is becoming central instead of quantitative modelling and expert knowledge based systems are in the core. Shift from the Analytic approach Merkhofer (1986) [1] paraphrases a comment of a leader: ”the value of analysis [Decision analysis] lay not so much in its specific conclusions, (which he suggested were fairly intuitive), but in its clarification of the logic behind conclusions”. Analytic decision-making is based on abstract models, explicit formal theories, which create an information overload on decision-makers, and in many cases it is incompatible with both organizational structure and processes. Some applications may be motivated by the desire only to support a particular point of view. According to Merkhofer (1986) [2] decisions in organizations based on the Analytic Paradigm are typically motivated in certain situations when either: Procedural rules applied in organizations evolved in stable environment or Strategic decisions are to be made, or Responding to changes in the environment of the organization, or Organizational goals are implicit or fuzzy, or There is a group within the organization with interest in rational choice, or There is an inner rivalry between interest groups within the organization or

Upload: sushant-patil

Post on 07-Apr-2015

1.130 views

Category:

Documents


12 download

TRANSCRIPT

Page 1: Cybernetic Paradigm of Grissinger

The cybernetic paradigm

Gyula MEZEY Zrínyi Miklós National Defence University,

Budapest, Hungary E-mail: [email protected]

Risk-based decision-making in reducing security risks is based on practical approach, usually a combination of rational and cognitive theories. In risk-management analytic, cybernetic, cognitive paradigms can be applied.

The Analytic Paradigm has the potential to clarify a decision, but cannot guarantee an improved decision-making. Under the Analytic Paradigm it is asserted, that someone or a group will actively select from among distinct options in a stable decision environment so as to achieve a set of goals. Recently widely recognized, that this assumption is farfetched from reality.

The Cybernetic Paradigm has been increasingly in use for group decision support systems. This paradigm focuses onto uncertainty control and not optimization. Qualitative modelling is becoming central instead of quantitative modelling and expert knowledge based systems are in the core.

Shift from the Analytic approach Merkhofer (1986) [1] paraphrases a comment of a leader: ”the value of

analysis [Decision analysis] lay not so much in its specific conclusions, (which he suggested were fairly intuitive), but in its clarification of the logic behind conclusions”. Analytic decision-making is based on abstract models, explicit formal theories, which create an information overload on decision-makers, and in many cases it is incompatible with both organizational structure and processes. Some applications may be motivated by the desire only to support a particular point of view.

According to Merkhofer (1986) [2] decisions in organizations based on the Analytic Paradigm are typically motivated in certain situations when either:

− Procedural rules applied in organizations evolved in stable environment or

− Strategic decisions are to be made, or − Responding to changes in the environment of the organization, or − Organizational goals are implicit or fuzzy, or − There is a group within the organization with interest in rational

choice, or − There is an inner rivalry between interest groups within the

organization or

Page 2: Cybernetic Paradigm of Grissinger

− Conflict resolution is on the agenda. However, in search for resolution of conflicts the recent Cybernetic Paradigm seems to be a more efficient basis.

The Cybernetic Paradigm Conflict is a problem situation, characterized by misalignment of their

objectives is perceived by social entities. Misalignment is either different order of importance or mutual exclusivity of goals. A conflict is strategic, if some of these objectives are related to the survival of one of the social entities. Politics uses a combination of coerce, deterrence, and economics. In case of a commercial conflict the resolution is explicitly held in favour of the client company. However, in case of an IR (International Relations) conflict the above assumption is implicit. Explicit aim is to achieve a stable solution by sustaining the peace and international order. But even if violent solution (armed conflict) occurs, the aim in conflict resolution is achieving a stable solution [3]. This may be solved peacefully by finding super-ordinate goals to get only by co-operating together.

Results of decisions are hard to evaluate in security, thus in the absence of commonly agreed criteria, or methods, - which exist in business management - the judgement whether a decision was correct or not, depends on the later success (survival) of the respective organisation or individual politician. The Cybernetic Paradigm is grounded on the bounded rationality concept [4] and the cybernetic decision model [5] of Steinbruner (1976). In complex situations decision makers, inherently constrained by cognitive and information processing limitations, typically do not optimise expected value, – they satisfice, and minimize uncertainty.

In conflict management a natural goal is trying to predict the future, and the behaviour of adversaries in a conflict. As part of conflict management we deal with forecasting techniques within the context of Crisis Analysis. The quantitative models applied by game theory, decision theory had been proved un-predictive [6], and evaluation of the utility function of the opponent, or probabilities had rarely ever been attained in reality.

Decisions are made under risk in the possession of perfect information, when the outcomes of each option and their probability distribution is known, or made under uncertainty if not all the outcomes of each option and their probability distribution are fully known. Due to time-stress and cognitive computational limitations of leaders, rather than all alternatives and outcomes are to be considered, some of the alternatives are ignored in favour of a satisfying one. Minimization of uncertainty is

Page 3: Cybernetic Paradigm of Grissinger

facilitated keeping key security policy variables within tolerable ranges, through information feedback loops.

The Cybernetic Paradigm is based on the assumption of uncertainty control since the psychological effects of uncertainty held to a minimum. The decision-maker does not need to engage in alternative outcome calculations or in updated probability assessments – much of that activity is performed by a mechanism: semi-automated.

It was hoped to have better results by not trying to predict “rational” results but to perform simulations in order to train one of the adversaries or give advice, prepare a proposition taking into consideration the behaviour of the opponent. The new concept of Decision Support System was introduced.

Cybernetic (or artificial intelligence/AI/, or business intelligence) decision-making is based on a working mechanism (for instance a type of organisms modelled by computer), but not dissent from rationality, or established procedures.

AI is defined as the application of knowledge, thought and learning to computer systems to aid humans [7] with its major overlapping sub-fields: robotics, information fusion, NLP (Natural Language Processing), CV (Computer Vision), KB (Knowledge-Based) or ES (Expert System), and learning systems. Systems ought to learn otherwise they run the risk of presenting obsolete knowledge.

In crisis, decisions must be quick to cope up with fast-changing situations typically in a context of uncertainty (lack of information and/or overwhelming flow of non-consistent data). Raw data collected must be analysed and structured to provide the decision-maker with usable information. Stored knowledge of experts readily available provides a more flexible tool for supporting quicker decision-making compared to traditional decision aiding systems. Restructuring, modification, prototyping of models is rapid using a tool-box and a flexible ‘qualitative model’ (for instance a knowledge base), than in a traditional information system.

Changes in the situation during a conflict can lead to the modifications of the objectives. Moving from a situation to another unforeseen in the beginning, threatening and counter-threatening, trial-and-error, feedback, simulation are typical functions in conflict-management. Whereas an axiomatic simulation model does not, a knowledge-based system with its ‘qualitative model’ allows dynamic modelling and ESD (Evolutionary System Design) [8]. If properly structured it is easy to change some of the RB (Rule Base) to observe what the new outcome would be.

Page 4: Cybernetic Paradigm of Grissinger

For decision support in conflict analysis knowledge based (KB) and extended game theory-based systems are frequently used. The former class will be tackled in the beginning, while the latter category will be dealt with in the end of this article.

Extended game-theory based systems in practice can have an important predicting capacity and a training effect. These models fundamentally rely on mainly quantitative [9] data, where freedom of the decision-maker results from choosing outcomes and trying different preference orders on the outcomes by means of simulation.

KB systems fundamentally rely on qualitative model, where freedom of decision maker is greater and comes from both dynamic modelling and flexibility by rapid prototyping.

An important factor is in crisis management, that for example EFAR can require 1-2 weeks [10] to get an output, whilst rapid prototyping in KB systems needs only 1-2 days.

Qualitative modelling Let us recall some notions in the beginning: − ES (Expert System) According to the BCS (British Computer Society), an ES “is

modelling, within a computer, of expert knowledge in a given domain, such, that the resulting system can offer intelligent advice or take intelligent decisions”. Rather than imitating the reasoning processes experts use to solve specific problems an ES concentrate only to achieve similar results. Many ES work only at an advisory level [11]. When a system has not got such an advanced level knowledge as an expert has, - then it is not branded as ES, just an advisory level KB system.

− KB (Knowledge Base) In an ES it is useful to separate functional entities (inference engine,

rules in the KB, facts in the domain DB (Data Base), DMBS (Data Base Management System), user interface, knowledge acquisition facility). The KB is a basic part of an ES, it is a specific software, which contains the facts about a narrow domain and relevant heuristic rules. Though KB is an essential part of an ES, not all ESs are RB (Rule Based) systems, – the representative form of knowledge can be for example frame-based, script-based or hybrid one as well, but these technicalities are not going to be discussed.

The rules (or the procedures) can be seen as a ‘qualitative model’ (RB). Though it is in combination with a normative component, this qualitative model fundamentally does not rely on a mathematical model.

Page 5: Cybernetic Paradigm of Grissinger

The KB contains either a person’s knowledge related to scenarios, or results of modelling and simulation. For the purpose of crisis management the contingency plan and basic response knowledge based on earlier decisions made in the planning, organizing and pre-crisis stages can be included in a KB.

Though no single expert is available to provide all the relevant rules from his intuition, it is a fundamental assumption of ES, that holistic expert decision-making is valid. Circularity (self-referencing), redundancy, incompleteness, conflicting rules, random error in human judgement can usually be found in the KB of an ES. These problems in the particular KB of crisis management are more risky, than in a simple ES. In ensuring completeness, the short of experts and some sort of empirical disaster data can be balanced by higher standards of transparent logic [12]. Neither a single KB, nor a uniform way to change knowledge, no goal-directed search with programming languages possessing inference capabilities using generalized search should be assumed.

− Distributed net of ESs To have a robust redundant ES a loosely coupled network of ESs

could be appropriate to use. Cooperating ESs in a distributed system could ‘talk’ to each other like experts do, - although, as is mentioned above not all KB systems are real ESs with deep procedural knowledge. The various RB agents would have to exhibit organizational behaviour and human-like thought processes. One difficulty is ‘hypothesis-conflict’ resolution: experts often come up with conflicting hypotheses.

When knowledge representation of a variety is distributed, then the control logic should be highly structured, procedural. When implementing a change in a rule, usually one merely can change a parameter (a data beforehand, or interactively, during simulation).

The evaluation of a rule base (RB) depends only on the adequacy of its resulting outputs fired by the facts. For example, in case of a negotiation support system, the coherence of the qualitative model is self-contained, and completeness and lack of contradiction are relevant. A meta-level evaluation related to the theoretical coherence of the representation is irrelevant.

Next two particular types of ES (useful in defence systems) are identified:

− Bootstrapping ‘Bootstrapping’ is the full replacement of the human decision-maker

by a linear statistical model. A fundamental assumption of ‘bootstrapping’ is, that decision-makers are able to identify the key predictor variables of

Page 6: Cybernetic Paradigm of Grissinger

that model [13]. Bootstrapping models are best in routine simple decision-making while ESs are in advisory roles.

− Linear modelling A particular type of KB, linear modelling uses human judgements,

predictions to build a statistical model [14]. If the person is an expert, the assumed quality of his judgmental input is high. Know-how of employing the appropriate models, interpreting the results is also should be contained, since beside deep procedural knowledge, meta-knowledge also should be handled.

– A KB for contingency planning There are six major functional elements of a framework [15] to

support scenario generation for contingency planning. KB is a central element of this framework. The KB should consist of at least three major functional components of a problem: a video-archive can be instructive as to the dimensions, dilemmas, multiple goals; a model-base refers to the decision aids available; and a group-process base can be used to handle the group-influencing options.

The collection of historical cases and their resolutions constitutes a KB and CBR [16] or ANN [17] (Artificial Neural Network) can be applied to identify patterns, relationships, which subsequently lead to formulating rules for ES and infer potential solutions for solving future problems. Even if it is so, building an ES is highly labour intensive. Reducing labour, for structure discovery interactive induction and knowledge acquisition can be combined and automated (for instance: Auto-Intelligence [18]).

Nevertheless, there are difficulties using ES in dealing with unanticipated events.

MKB (Meta Knowledge Base) Meta-knowledge has a narrow meaning: the system’s knowledge

about ‘how’ it reasons is called Meta-knowledge [19]. In a broader meaning all is known about the structure and content of the KB (‘rules about rules’ provided the KB is a RB system) can be called Meta-Knowledge as well.

In contingency planning likely the most important is to represent: • Characterization of alternative command level models • Strategy of decision-making: Categorized either according to certain paradigms: Analytic, cybernetic,

cognitive (bureaucratic political), or according to Flin [20]: RPD (Recognition-primed Decision-making), Creative, Analytical,

SOP-based,

Page 7: Cybernetic Paradigm of Grissinger

• Framework for organizing information: 1. Crisis states; 2. Crisis levels: the rules are modularised by the current situation,

ordered by an escalation ladder, and within the modules hierarchically structured;

3. Hierarchical situation assessment: − National command level: few key variables, soft data; − Operational command level: intermediate variables, soft data; − Tactical command level [21]. DSS (Decision Support System) DSS can support all phases of the decision process, includes a model

base handled by a model base management subsystem, a communication (dialog) subsystem, a DB (Data Base) handled by a DBMS (Data Base Management Software), and optionally a KB with a knowledge management subsystem, or/and a gaming system. In case of group decisions, a GDSS can be used.

By definition a DSS (in case of group decision-making a GDSS, and often an ES as well) normally in every phases of the decision process can be useful in providing some sort of support. According to the model of Sprague (1993) [22] the following phases of the decision process can be supported by DSS, or ES, or EIS (EIS=Executive Information System):

− Support for the intelligence phase: Ability to scan the environment, external and internal databases for

opportunities and problems, interpretation of what the scanning discovers. An EIS helps in continuously monitoring by accessing databases

rapidly and efficiently. A DSS can model and analyse data fast. An ES can diagnose problems and interpret information. − Support for the design phase: A DSS usually has the capability to generate alternative courses of

action, forecasting the future consequences, criteria for choice and their relative importance.

− Support for the choice phase: By means of a DSS different scenarios can be tested before the final

decision. An ES can assess the desirability of the solutions and helps in

recommending one. − Support for the decision implementation phase: A DSS can assist in decision communication, explanation, and

justification.

Page 8: Cybernetic Paradigm of Grissinger

An ES can provide training. We will deal with versions of DSS and technicalities applicable for

crisis management later in an other article. Next we are going to discuss conflict analysis decision support only for the design phase when forecasting future consequences.

Forecasting techniques • A forecast can be based on some normative (Analytic Paradigm)

techniques, for instance: 1. Decision matrices 2. Relevance trees [23] 3. OR (Operation Research) techniques [24]: for instance linear

programming, dynamic programming • Exploratory forecasting techniques could be applied, for example: 1. Trend extrapolation: extrapolation of time series, time-independent

plots, envelope curves 2. Morphological analysis: discovers the totality of options [25] 3. Intuitive techniques: for example Delphi method 4. Economic analysis: CBA, CEA [26], Discounted Cash Flow 5. Games 6. Modelling 7. Scenario writing 8. Assumption-based multi-scenarios Shift from trend-based planning The steps of trend-extrapolation in an outside-in perspective [27]: 1. Identifying trends in the environment (‘environment scanning’); 2. Selection of important trends likely to drive the future (expert

judgement); 3. Generating a ‘future world’ from important trends (‘extrapolation’); 4. Planning in dealing with the ‘future world’. Uncertainty can often be handled by probability theory, Monte-Carlo

analysis. Causal chain can be used to relate each individual decision to an

overall objective [28]. A causal chain path between cause and possible effect (benefit/cost) can often be represented by a relevance tree [29], or an influence diagram. In practice both normative (decision matrices, relevance trees, etc.) and exploratory forecasting can be combined.

In identifying a single trend it is difficult to reach a common understanding of a group

Page 9: Cybernetic Paradigm of Grissinger

Assumption-based planning This is failsafe planning, since it is not necessary to reach a common

view of the group. It is easier contemplating changes to the current world, than to predict one future world. The steps of assumption-based approach in an inside-out perspective start from a common understanding of a group, looking for changes, which are likely to violate assumptions:

1. Identification of assumptions underlying current operations (reading documents and checking interviews);

2. Gathering plausible elements of change (Delphi-method by a group of experts);

3. Identification of elements of change that could upset current assumptions (no systemic commonly agreed method);

4. Generating a world for each violated assumption (multiple world generation);

5. Developing ‘signposts’, ‘hedging and shaping’ actions (‘hedging and shaping’ plans) [30].

Assumption-based planning leads to multiple scenarios and can be seen as if collective decisions were fragmented into small segments treatable sequentially. This fits to the “labelled sequential attention to goals” process at the top of a hierarchy of units of a large organisation [31]. Goal integration is “refused” by top management in practice, its high-level decision preserves the fragmentation in the set of the lower-level decisions [32], [33], [34] provide scenario-construction methods.

A fundamental assumption: Value integration rejected Value integration is purposefully rejected and replaced by only

preservative values under the Cybernetic Paradigm. Also it is not yielded a coherent preference ordering for alternative states of the world under trade-off conditions. Co-ordination of executives does not mean analytical integration in practice, but it does mean that established routines in the organisation must be rendered consistent. These clusters of routine SOP’s (standard operating procedures), once established, are not readily changed.

A decision-making mechanism produces a variety in range of outcomes as a consequence of completing a process, without any clear picture of the actual product, but monitors a few (critical) feedback variables. The dominant assumption is to control the variety inherent in the decision problem, rather then to find directly an optimal solution. It is assumed, that subjective probability assessment of rare events is of poor quality. In practice it is really difficult:

Calculating probability of hazard, because of lack of data (events remain unreported), estimating dose-response relationships (most populations are exposed to multiple hazards, and isolating effects is

Page 10: Cybernetic Paradigm of Grissinger

difficult), predicting exposure, extrapolating the effects, estimating consequences, measure benefits and costs, because of uncertain models – better saying: our limited knowledge.

That is why a “learning mechanism” selects solutions not on a theoretical, but rather on an instrumental basis. Rather than identifying an optimal, it tries to find a robust option that satisfies across a range of possible futures. Lack of appropriate sensitivity analysis tools for identifying robust solutions led to the recent development of some new software tools, for example Dynarank, @RISK (an add-in for Excel), VISA [35].

Robust solutions to changes in inputs may provide an easier solution of disagreement about weights, when members of the decision-making group see, that differences in their individual weighting often do not matter and so there is no point in debating weights, priorities [36]. This can contribute to a much quicker decision-making process, - however, on the expense of explicitly expressed opposite personal opinions. But as is well-known, it is impossible to derive a truly democratic decision for resolving differences of opinion, according to Arrow’s Impossibility Theorem, so a search for a group value or utility function is futile.

The adaptive behaviour of a cybernetic decision-mechanisms occurs when the decision-making happens in a stable environmental subsystem [37]. Defence planning is risk averse, normally the pessimism of the ‘maximin’ decision criterion is applied: choose the best of the worst possible outcomes [38]. However, the aim of the decision is not to select the option with the highest expected value or utility, but to find the ‘most robust’, or at least a ‘robust enough’ solution. This is psychologically likely most important for those preferring the certainty to choice between risky options (‘maximin’ criterion). By generating scenarios of favourable futures, windows of opportunities can be anticipated, so making possible (at least in theory) immediate actions. Rehearsing the future of adverse scenarios sensitises for specific triggers for these unfolding scenarios as early warnings in order to prompt rapid deployment of intervention towards an unfavourable not impossible future. Potential threats otherwise may be underestimated or ignored.

Rehearsing the future by predictive and exploratory models Predictive models Computer modelling has been successful in those domains where the

predictions of the models can be verified by experiment and troubled when there is no possibility of experimentally validating model correctness and utility. Verification, validation, opaqueness of model internals and outputs, sensitivity analysis in our case are problems, because predicted behaviour

Page 11: Cybernetic Paradigm of Grissinger

of social models cannot be experimentally validated, so these are to be seen as ‘exploratory models’. Where prediction, and/or experimental validation is not possible, no single ‘true’ model can be agreed upon. It does not matter, how detailed is the model, no amount of detail can provide validation, only the illusion of realism.

Validation of results of simulations in scenarios with a model based on historical experiences is still an open question as to the relevance of such efforts. Validation of models in the social sciences is inherently difficult. Given the absence of sufficient data, peer review of models, redundancy of analysis and comparison of results, partial validation of the model, sub-model validation where possible, can contribute to maintaining the integrity of planning – but the model remains still non-predictive. Non-predictive models must be treated differently from those experimentally validated.

Exploratory models Exploratory use involves the ‘guessing’ of details of a system for

which there are no data. The implications of these guesses are computed, which might lead to exploration (for instance Chaos Theory was explored as a result of simulating anomalous behaviour). Sensitivity analysis (by which uncertainty in inputs is related to the uncertainty of outputs) is critical for a model, which is believed – because of confidence in the correctness of the conceptual model – to be predictive. For an exploratory model assigning error ranges to the outputs is essential. So we not only can opt for a multitude of sketchy, though easy to understand, verify, and analyse sensitivity models, but we cannot help doing so. Even when a model is not validated, it can serve as an ‘inference engine’ in search for conclusions to facts or relationships that are invariant across all the set of plausible models to reveal the desired invariance in a context by some external analytic strategy. A complete sensitivity analysis might assess the behaviour of all plausible models, but in practice we rely on a sampling search for critical aspects of flexible allocation of resources even when the exact range of outcomes is unknown.

Page 12: Cybernetic Paradigm of Grissinger

In the absence or short of data the model contains, ‘exploratory modelling’ focuses the question to be answered. The exploratory model should be revised by iteratively redesigned and implemented, during the course of learning through its use. The complexities of the system and the uncertainties in the environment are represented by a set of models [39]. The scope of any individual model can be designed to maximize its utility for answering a particular question.

Forecasting, conflict and negotiation-control by use of extended game theory

Game theory assumes conflict situations (‘a game’) controlled by ‘depsychologised’ rational behaviour of the decision-makers. Classical game theory deals with non-cooperative behaviour and zero-sum games. The approaches of Nash and Harsanyi [40] in modelling bargaining and negotiation, first between two opponents, then covering their co-operative game, later embracing more then two players, and cases where the total payoff to all players was not constant. Co-operative game players have the opportunity to communicate and to form binding and enforceable agreements.

Equilibrium concepts in games are usually: Pareto optimum, non-myopic, Stackelberg, Nash equilibriums. A myopic attitude is to put the player and the other player at personal risk in order to maximise his own gain. Pareto optimum exists if no player gets into a worse situation. Stackelberg equilibrium is when the second player moves in full knowledge of what the first has done. Nash equilibrium exists when no player has motivation to make a move even if he is capable to do so [41].

According to Downs (1967) [42], individual actors’ game strategies and collective bargaining manoeuvres [43] are often understood on game theory basis.

There are empirical evidences [44], that – at least in some circumstances – human mind works in accord with assumptions of the rational decision theory model. Decisions were successfully modelled by game theory, which was also applied to explain animal, and VUG (variable universe games) [45] for explaining human behaviour [46].

Bargaining theory tries to explain how wars can occur within rationalist framework of international relations (IR). Rather than only defending the own position, a creative attitude based on multi-attribute value analysis to negotiating problems led to a better deal by trading off interests and Pareto optimum was applied [47]. However, in game theory normally it is assumed that the players may have different pay-offs from the other, for any particular choice of options, but they each know those pay-offs. But in practice this is difficult to assume. A meeting, or decision

Page 13: Cybernetic Paradigm of Grissinger

conference may be used supported by a decision analyst in case of group decisions examining the impact of losses to be compensated by the increased benefits of other members of the group.

It is a practical disadvantage that expressing the mutually exclusive options of the opponents leads to a large number of course of actions even in simple cases. Negotiation problems can be characterized by a small number of objectives and a large number of potential courses of actions. That is why such a software, like EQUITY [48] is used by the decision analyst in order to perform the reallocation or trade-off calculations within a cooperative group. This is facilitated applying a common scale to measure benefits and a combined multi-attribute utility function [49] or in some cases values and not utilities are applied.

Both, a consensus developed by means of trade-offs within a group and a negotiation between two opponent groups rely on multi-attribute value analysis, with the additive value model being appropriate in many cases [50].

Strategic conflict decision-support requirements Strategic conflict is a conflict of vital interests. Organisational

decision-making, and defence decision making particularly requires analytical support in managing conflict. A strategic conflict decision-support system would accommodate the following assumptions and require the following characteristics [51]:

− An assumption, that some equilibrium existing in game theory is used;

− The DSS can model the potential, power, motivation of participants in control and direct the conflict. That game is unlikely to be zero-sum;

− DSS can accommodate differing belief spaces of participants in the rationality, objective, structure, and intent dimensions. The connections between system states, the nature of transitions themselves between the states, what constitutes success, the nature of solution, may not commonly viewed by the players, and likely to change as the game evolves. The transitions usually are unrepresentable simply by cost, or probability of transition. Multiple valid viewpoints may exist at all stages of the struggle. Due to the difficulties over transferability of utility functions [52] on which based a common transaction variable between players the utility outcomes could be ordered by preference. The moves will be ruled by authority reputation, personal influence position of senior

Page 14: Cybernetic Paradigm of Grissinger

players, under the urgent need for action, the cost of delay, and sanctions;

− Adaptive to changing circumstances – the original aim can change as well;

− The DSS accommodates limited, defective and deceitful data, but; − New evidence or perception should lead to data improvement; − The DSS accepts human and group irrationalities; − Depth-changing ability between strategic and tactical level

negotiations and decision-making in practice means high and low resolution representation (zooming). Within a local subset of the wider network a variety of particular micro-economics and game theory methods should be invoked to reach a local (tactical level) optimum for local negotiations and decisions;

− System-centred and not viewing the conflict/struggle from the viewpoint of one party;

− Both hard and soft data should be accepted; − Auditability/backtraceability; − Be itself acceptable to the social environment in which it is going to

operate; − Non-conflicting with the already existing systems. Extensions of game theory approaches to conflict Conventional game theory fulfils only few of the criteria established

above, although at a potential lower level for specific purposes and situations it must be taken into account.

IR (International Relations) conflict resolution and behavioural conflict analysis approaches do not seem to be easily usable for computation.

Non-game theoretic conflict analysis is usually underdeveloped to provide a procedural basis, except for Knowledge Based systems.

Extensions of game theory (deterministic graphical games [VUG, DG], hypergames, metagames, foveal games, and confrontation analysis and drama theory, multiple futures planning (ie. FAR) have shown the potential to be developed for DSS (Decision Support System), because these models are brought down from the pedestal of the infallible ‘black box’ to occupy a more modest position as a complement to the thinking and deducing powers of crisis management.

Variable universe games (i.e. VUG, VUG1, VUG2, VUG3) Players have different knowledge of characteristics of the game space,

and different pay-off for given terminal states. The key is the different

Page 15: Cybernetic Paradigm of Grissinger

‘belief sets’ of the players [53], [54]. Players may disagree about the very structure of the game.

Deterministic graphical games (i.e. DG, DGT, DGA) Brams (1994) established a complete taxonomy of 2x2 discrete DG’s

[55]. DGT (DG terminal) games are formalised as cyclic graphs, might

be unending (with 0 payoff). DGA (DG with time-averaged pay-offs) games extend DGT and

stationary equilibrium have been described [56] and is shown, that pure strategy history-remembering Nash equilibriums always exist. But it is difficult to establish a cost function associated with a player’s changing his tactical choices.

For practice it is a problem, that both DGT and DGA assume perfect information.

Super-games It concerns decisions of parties with a choice between continuing co-

operation (alliance) or betrayal [57]. Hyper-games Complex interactions among protagonists can be flexibly modelled

[58]. It has a depth-changing, data-improving, and ‘forgiving’ ability. Its problem is the sparse literature.

Meta-games Representing the conflict situation a meta-game offers a list of options

asking the user for identifying which options are excluded, then rank ordering the rest on a diagram [59]. Scenarios are seen and their related possible ‘improvement paths’ with associated benefits and costs. Human attributes like hate, anger, can also be represented, and recommendations are presented in a graphical way. Literature is available on case studies of incompatible goals of players. Both the CAP (Conflict Analysis Program) and DECISIONMAKER were designed by Fraser and Hipel (1984) [60].

But one cannot control the outcomes simply by choosing an attractive one – this is just ‘management by hope’. You may control only your moves, usually these are fragments of the ‘improvement paths’. In practice an earlier analysis (‘sense-making’) and planning is required [61] before the list of outcomes put out. Planning here means action planning – the aim of the meta-game is to produce coherent action plans for the player. ‘Sense making’ is a specific communication using Habermas (1981) communicative rationality concept [62]. Next, large number of tables must be generated; in meta-game analysis each player needs a separate table for each outcome. It is a problem when information in a table is incomplete or

Page 16: Cybernetic Paradigm of Grissinger

misleading, and quantitative input data are difficult to absorb, the method is difficult to computerise.

Most of the outcomes form meta-equilibrium, an outcome, which is rational for all players. Here 3 levels of rationality are specified: rationality, symmetric meta-rationality, general meta-rationality, embodying traditional and new concepts on the verge of rational and ‘irrational’ decision-making [63].

Confrontation analysis and drama theory Drama theory is an extension of the meta-game [64] to understand the

effect of changing motivations and utilities by players during the evolution of the game. A drama evolves through scenes with tactical choices (represented by trees) can be made by each actor. Each player in each scene adopts a position – a scenario of the wished outcome. It is the mathematical treatment within confrontation analysis, which provides suggestions, how the drama is likely to develop [65]. It seems as if the establishment of the positions of the players could only achieved with their co-operation. It has some applications and has a DSS available [66].

A ‘crystal ball’ of the UK A number of tools: FAR, Foveal games (EFAR), Powergraph (UK)

and KORA (US-Russia) have been in use for decision support purposes in political, military, business long-range planning.

Multiple futures planning (i.e. FAR) FAR (Field Anomaly Relaxation) is a future scenarios development

tool [67] used in political and military planning [68] This is a discrete state transition based network analysis. In its first stage FAR applies the AHP (Analytic Hierarchy Process) to be filled in by experts [69].

Getting from the situation, where we are (a present state in an N-space of situations) it is desired to transit us to a wished situation (the object state). The problem space can be considered as a N-space of combinations of N field descriptors with each combination describing a conceivable future outcome (scenario), some of them are desirable, others are adverse. The output of FAR is that the states are finally arranged into a ‘future tree’ form, whereby no audit is possible [70]. The tree can also be considered as a N-space of combinations of N field descriptors with each combination describing a conceivable future outcome (scenario).

After clustering the values of the output, the discrete state transition based FAR tree can forecast for instance the development of the possible futures of a region.

Foveal games (i.e. EFAR) The ‘foveal game’ structure consists of three games: the strategic

scenario game is based on heuristically derived future scenarios, local

Page 17: Cybernetic Paradigm of Grissinger

games are using conventional game theory, and there is a transition game in order to ensure consistency between strategic and local levels.

EFAR (Extended Field Anomaly Relaxation) extends FAR to business strategic planning purposes as well [71]. An output of the EFAR analysis is a network of scenarios, where the environment can recover to previous states[72] so backtracking/audit is possible.

EFAR can represent the more dynamic nature and future movement of organisations within the business context.

By an iterative solution process, intertwined action planning and sense-making of a complex scene at any stage is possible. It allows the specific analysis of the transitions between states. It has depth-changing and focusing abilities between strategic and tactical level decision-making. Actions, decisions, negotiations at tactical level may affect the position of the strategic level player so maintenance of validity of the local/tactical decision is necessary.

An output of EFAR is a set of strategic directed graphs, but these are not detailed enough for use to play a local game.

Powergraph ‘Powergraph’ is a technique [73], which allows – in the frame of the

EFAR a ‘foveal game’ – the representation of the specific options open to either an operational, or a tactical level player. It is appropriate for the decision support of the management in a specific conflict, campaign, battle, business competition struggle for a bid, etc. By means of forecasting, helps in the formulation of a strategy taking into consideration capability and intent, and that decisions can depend both on the beliefs and desires of a decision maker.

‘Powergraph’ identifies the players, states of the future (as a network), the players preferences to those states, considers who controls the transitions between the states specified above (Boolean expressions in a transition power matrix expressing the ability of each player to move the system between the states), rank-orders the players preferences to the states, algorithmically performs both activities: it determines motivated power and examines feasible developments; then identifies responses based on: power plays, motivational changes, contingency plans; in order to convert them into an action plan. ‘Powergraph’ is most effective when used in an iterative fashion [74].

References, notes [1] Merkhofer, M.W., Decision science and social risk management;

technology, risk and society, (D. Reidel Publ.Co., Netherlands 1987) p.295.

Page 18: Cybernetic Paradigm of Grissinger

[2] Ibidem, p.175. [3] Azar, E.E., Burton, J.W., International Conflict Resolution Theory

and Practice, (Brighton: Wheatsheaf 1986) p.89. [4] Simon, H.A., The new science of management decision, (Harper and

Row Publishers New York 1960). [5] Steinbruner, J.D., The Cybernetic Theory of Decision, New

Dimensions of Political Analysis, (Princeton Univ. Press 1974). [6] Levine,P., Pomerol, J-Ch., Negotiation Support systems: an overview

and some knowledge-based examples, (in: Avenhaus, R., Karkar, Rudnianski, H.,M., eds.: Proceedings of the 1st ARESAD International Conference on Decision Making and Defence, Paris, Nov 22-23. 1989. Springer Verlag 1991) p.241.

[7] Daniels, J.D., Artificial Intelligence: A Brief Tutorial, (in: S.J.Andriole: AI and national defence: Applications to C3I and beyond, AFCEA Washington D.C. 1987) pp.3-12, p.3.

[8] Shakun, M.F., Group Decision and Negotiation Support in Evolving, Nonshared Information Contexts, (Cahier du LAMSADE No 89, 1986)

[9] Levine, P., Pomerol, J-Ch., Op. cit., pp.241-256. [10] Yong, Y.C., Creating coherent appreciation with Field Anomaly

Relaxation, (MSc. dissertation RMCS, University of Cranfield 1994). [11] Goodwin, P., Wright, G., Decision analysis for management

judgement, (Wiley&Sons 1998) pp.410-11. [12] Dubois, O., Carlier, J., Checking knowledge bases by methods solving

satisfiability problem, (in: Avenhaus, R., Karkar, H., Rudnianski M., eds.: Defence Decision Making, Proceedings of the 1st ARESAD International Conference on Decision Making and Defence, Paris, Nov 22-23, 1989. Springer Verlag 1991) pp.29-39.

[13] Goodwin, P., Wright, G., Op. cit., p.25. [14] Ibidem, p.401. [15] Belardo, S., Harrald, J., A framework for the application of group

decision support systems to the problem of planning for catastrophic events, (IEEE Transactions on Engineering Management, November 1992).

[16] Turban, E., Decision support and expert systems, management support systems, (Prentice Hall Intl. Inc. 1995).

[17] Artificial neural network. [18] Turban, E., Decision support and expert systems, management support

systems, (Prentice Hall Intl. Inc. 1995), p. [19] Ibidem, p.

Page 19: Cybernetic Paradigm of Grissinger

[20] Chrichton, M., Flinn, R., Command decision making, (in: Flin, R., Arbuthnot, K., eds.: Incident command: tales from the hot seat, Ashgate Publishing Company, England 2002), pp.201-238.

[21] Bryson, K.M., Millar, H., Joseph, A., Mobolurin, A., Using formal MS/OR modeling to support disaster recovery planning, (European Journal of Operational Research 141 2002), pp.679-88.

[22] Sprague, R.H., Watson, H.J., Decision Support Systems, (Englewood Cliffs, NJ: Prentice Hall 1993), p.13.

[23] Defence Intelligence Agency Methodology Catalog, An aid to intelligence analysts and forecasters, (Chapter II, DDE-2200-227-86 November 1986).

[24] Beattie, C.J., Allocating resources to research in practice, (NATO conf. On Applications of Mathematical Programming, Cambridge, UK (EUP) 1970).

[25] Zwichy, F., Morfologische Forschung, Winterthur, Winterthur A.,G., Switzerland, 1942.

[26] Quade, E.S., Cost-effectiveness: an introduction and overview RAND No.P-3134, 1965.

[27] Dewar, J.A., Levin, M.H., Assumption-based planning for Army 21, RAND 1992, p.7.

[28] Esch, M.E., Planning assistance through technical evaluation of relevance numbers, Proc. of the 17th National Aerospace Electronics Conf., Dayton, Ohio, 10-12 May 1965 IEEE, New York.

[29] Jones, P.M.S., Technological forecasting as a management tool, (P.A.U.M 10, HMSO London 1969).

[30] Schwartz, P., The art of the long view, (New York: Doubleday, 1991), pp.7-10.

[31] Cyert, R.M., March, J.,G., A Behavioral Theory of the Firm, (Englewood Cliffs, N.J., Prentice-Hall, Inc., 1963).

[32] Crecine, J.P., Govenmental Problem Solving, (Chicago: Rand McNally & Co. 1969).

[33] Jackson, J., Statistical Models of Senate Roll Call Voting, (in: American Political Science Review, Vol.65. June 1971).

[34] Goodwin, P., Wright, G., Op. cit, pp.360-383. [35] Ibidem, pp.441-448. [36] Ibidem, p.330. [37] Ashby, W., A Design for a Brain, (New York: John Wiley and Sons

Inc. 1952) p.76 ( “joined systems”). [38] Mintz, A., Geva, N., Redd, S.B., Carnes, A., The effect of dynamic

and static choice sets on political decision-making: an analysis using

Page 20: Cybernetic Paradigm of Grissinger

the decision board platform, (American Political Science Review 91 (3) 1997, pp.553-566, p.554.

[39] Richardson, G.P., Rohrbaugh, J., Decision making in dynamic environments: exploring judgements in a system dynamics model based game, (in: Borcherding, K., Larichev, O., I., Messick D., M., eds.: Contemporary Issues in Decision Making, Elsevier, Amsterdam 1989).

[40] Harsanyi, J., Normative validity and meaning of Neumann and Morgenstern utilities, (in: Skyrms ed: Studies in logic and the foundations of game theory, Dordrecht: Reidel 1992).

[41] Shubik, M., Mathematics of Conflict, (Amsterdam: Elsevier 1983). [42] Downs, A., Inside Bureaucracy, (Boston: Little, Brown & Co. 1967). [43] Olson, M., The Logic of Collective Action, (Cambridge, Mass.:

Harvard Univ. Press 1965). [44] Sweets, J.A., Tanner, W.P., Birdsall, T.,G., Decision Process in

Perception, (New York: Holt, Rinehart & Winston, 1968) pp.78-101. [45] Fudenberg, D., Tirole, J., Game Theory, (Cambridge MA: MIT Press

1993). [46] Erickson, G.M., Empirical analysis of closed loop duopoly advertising

strategies, (Management Science, Vol 38 No 12 December 1992). [47] Phillips, L.D., People centred GDS, (in: G.Doudikis, F.Land, G.Miller

eds., Knowledge-based MSS Ellis Horwood, Chicester 1989). [48] Barclay, S., A user manual to EQUITY, (DA Unit, London School of

Economics 1988). [49] Goodwin, P., Wright, G., Op. cit., p.134. [50] Ibidem, p.135. [51] Powell, J.H., A network-based framework for strategic conflict

resolution, (Ph.D. thesis Cranfield Univ. 1997) pp.96-98. [52] Hirschleifer, J., Riley, J.G., The analytics of uncertainty and

information, (Cambridge:Cambridge UP 1992). [53] Bacharach, M., Variable Universe Games, (in: Binmore ed.: Frontiers

of game theory, Cambridge MA: MIT Press 1993). [54] Fudenberg, D., Tirole, J., Game Theory, (Cambridge MA: MIT Press

1993). [55] Brams, S.J., Theory of moves, (Cambridge: CUP 1994), pp.215-219. [56] Alpern, S., Stationary equilibria for deterministic graphical games,

(in: Binmore ed.: Frontiers of game theory, Cambridge MA: MIT Press 1993).

Page 21: Cybernetic Paradigm of Grissinger

[57] Liu, X., A spatial supergame model of bilateral interactions: the case of US-China relations, (Ph.D. thesis Texas A&M University, College Station, Texas 1999).

[58] Bennett, P.G., Modeling interactive decisions: the hypergame focus, (in: Rosenhead ed: Rational Analysis for a problematical world, Chicester: Wiley, J., 1989).

[59] Karkar, H., Mirror effects, metagames and crisis, (in: Rudnianski, M., Avenhaus, R., Karkar, H., Defence decision making 1991), pp 257-271.

[60] Fraser, N.M., Hipel, K.W., Conflict analysis: models and resolutions, (Amsterdam: NorthHolland, 1984).

[61] Whittington, R., What is strategy and does it matter? (London: Routledge 1995).

[62] Habermas, J., Moral consciousness and communicative action, (Cambridge: Polity Press 1990).

[63] Howard, N., Drama theory and its relationships to game theory, (Group Decision and Negotiation 1994, 3), pp.187-206, 207-253.

[64] Howard, N., Drama theory: fundamental theorems, (Proc. Conf. IMA at Waldham College April 1997).

[65] Bennett, P.G., Op. cit. [66] Bennett,P.G., van Heeswijk, S., Using software for confrontation

analysis: a case study in food safety policy, (Working paper at IMA conference at Waldham College April 1997).

[67] Rhyne, R., Whole pattern futures projection using field anomaly relaxation, (Technological Forecasting and Social Change 1981, 19) pp.331-360.

[68] Coyle, R.G., McGlone, R., Projecting scenarios for SE Asia and the SW Pacific, (Futures 1997, 27 (1)), pp.65-79.

[69] Saaty, T.L., The analytic hierarchy process, (RWS Publications, Pittsburgh 1990).

[70] Coyle, R., G., Crawshay, R., Sutton, L., Futures assessment by field anomaly relaxation, (Futures, 1994, Vol 26 No1), pp.25-43.

[71] Powell, J.H., An application of a network-based futures method to strategic business planning, (JORS 1997 Vol 48 No7), pp.857-872.

[72] Coyle, R.G., Powell, J.H., A network based approach to strategic business planning, (JORS 1997 Vol.48 No8), pp.793-803.

[73] Powell, J.H., A network-based framework for strategic conflict resolution, (Ph. D. thesis Cranfield Univ. 1997), p.239.

[74] Ibidem, p.258.