Simulating Organizations: Computational Models of Institutions and Groups

Order this book

Reviewed by
Rosaria Conte
Division of Artificial Intelligence, Cognitive and Interaction Modelling, Institute of Cognitive Science and Technology, National Research Council, Rome, Italy.

What We Would Like To Know About Organisations?

Although the science of organisations has a prestigious tradition, a number of questions have received only partial (or not completely satisfactory) answers. For example, definitions of organisations abound, but nobody would find easy to answer the question of what organisations are and what constitutes their properties, activities, constraints and behaviours. Despite the evident link between the science and management of organisations, the problem of how to create, modify and make them more efficient is a puzzle for scientists, as it is for entrepreneurs and managers. Which factors are mainly responsible for the quality of organisational performance, its accuracy, rate of error and so on? Are these factors external (environment) or internal (structure, design, strategy) to the organisation? Aside from these problems, less evident questions are also waiting for answers. These problems revolve around issues concerning the nature of the links between cognition and organisations, an issue about which one of the editors of the volume under review (Kathleen Carley) is an acknowledged expert. Thanks to her efforts and achievements, we are now certain that this link must be modelled as involving mutual influence. On one hand, individual mental constructs affect organisational processes and efficiency while organisations and groups affect agents and their minds on the other. Still, this mutual influence needs further investigation and qualitative modelling. For example, which aspects of the mind are essential for organisational life? How and why do people enter, represent and evaluate organisations? How do agents mediate intra and inter-organisational relationships, recursive structures and processes? Moreover, which aspects of the natural mind can be extended to the level of organisations? What is meant when referring to organisations as "agents", what type of agents are they and what are the interrelationships between agency on these two levels, individual and organisational?

Traditionally, social scientists are sensible to the question of how to model organisational life and performance and the attempt to make these more efficient. The relevance of organisations in a variety of application domains draws attention from an ever-increasing number of other scientific disciplines and fields. In particular, the scientific interest in developing projects for specifying organisational tasks and subtasks and for the automated creation and execution of organisational activity is growing in some fields of Information Technology (IT), in particular in Artificial Intelligence (AI) and Multi Agent Systems (MAS).

The collection under review was published in 1998 but its editors pioneered the computational study of organisations in the early 90s. However, the volume is more than an assortment of contributions representative of the computational approach: it is intended to present and promote the interface and cross-fertilisation among research varying along many dimensions. Two of these dimensiond, the disciplinary and the methodological, deserve special attention. As to the former, the collection under review puts together studies from different fields of the social, cognitive and computational sciences: not surprisingly, the editors are each representative of different disciplines and fields (cognitive and social science on one hand, and social AI on the other). As to methodology, many contributions are inspired by the simulation approach, where computational models and techniques are employed for the purpose of visualising and experimenting with artificial systems. Let us consider these contributions in some detail.

The Contributions

One of the many good things about this volume is its final chapter, where the author (Richard Burton) presents a handful of themes common to most of the other contributions and envisages some challenging steps for the future. As to the common themes, he summarises them as follows:

• Computational and simulation approaches are essential in the study of organisations
• Allowing fundamental questions about organisations to be addressed, especially since
• Information creation, modification and transmission are what we mean by "organisation".

Let us see whether and to what extent these statements are in fact implied by the various contributions.

Efficiency in Group Work

To a large extent, evidence from natural experiments shows an interesting tendency of groups to work better than individuals. Which factors are responsible for this result? The answer to this question is sought in simulation-based experiments on group work.

In the second chapter, for example, TEAM-SOAR is presented as a computational model of multi-agent problem-solving for simulation-based experiments. After a summary of Carley's studies on the interaction between cognitive and situational aspects in organisational performance, two simulations are presented. In the first, the decision model is associated with accuracy and the rate of errors. In the second study, communicated meta-beliefs are correlated with efficiency variables such as waiting time. Findings always indicate that a host of factors - such as the decision pattern, the style of information distribution (whether hierarchical or statistic), and the "informity" of the group, i.e. the accessibility of information - have an interactive effect on organisational efficiency. The main positive aspect of this chapter lies in its cross-methodological approach. The computational methodology can be employed to provide better (more analytical) understanding of how a given effect (such as group performance, social facilitation or lofting) occurs in natural systems. However, the findings discussed in this chapter are sometimes not immediately perspicuous: why does the increase of information cause group performance to worsen? What is the impact of the task type on group performance? In the simulations presented, only distributed tasks are taken into account: what about iterated tasks?

Many of the same problems are addressed in chapter four, where the relationship between performance and the structure of organisations and tasks is examined under different conditions of time pressure. The organisational structure or design may vary from flat systems (in which decisions are taken by means of voting mechanisms) to hierarchical ones (in which decisions are taken at the top level and transmitted downward). The task structure varies from non-decomposable to decomposable, and from the non-concentrated type, characteristic of an uncertain environment (where results are equiprobable and do not strictly depend on task achievement) to the concentrated type (where results are not equiprobable). The study shows that the most fundamental variable is the task structure. The highest level of correct decisions is made in concentrated environments with non-decomposable tasks and under little or no time pressure. Under these conditions, however, an asymmetry occurs in the types of errors: underestimation is more frequent than overestimation. With increasing time pressure, the difference between types of errors decreases and even disappears showing that time mitigates the effects of environment. It should also be noted that there is a trade-off between accuracy and the type of error: the risk of underestimation increases with more accurate decisions.

This type of study has the great merit of pointing out clearly the advantages of simulation, by leading to results that could not have been obtained using other methodologies or in natural conditions. What is less convincing however is the type of question posed in relation to existing conceptual models. In short, the work is too ambitious and poses too many questions at the same time. Too many factors are addressed in comparison with the limited conceptual equipment available. For example, the dimensions of task structure are not always clear (e.g. concentrated/non-concentrated environment). Furthermore, that the structure and design of the organisations is found to affect performance to a lesser degree than the task structure probably depends on how both notions are defined. Indeed, task decomposability is intertwined with the organisational structure, and it is quite difficult to disentangle these factors in practice.

Some of these questions are addressed in chapter six, where a domain-independent instrument (T�MS) is employed to test computational theories of co-ordination. T�MS helps model and simulate task environments at different levels of abstraction. It allows quantitative simulation - measuring variables assumed as parameters of the simulation model - of multi-agent system behaviour. As a consequence, centralised as opposed to parallel and distributed algorithms, different negotiation strategies and alternative organisational designs can be evaluated and compared. The domains of application for this system are shown to range from hospital patient scheduling and multi-physician consultation, to airport ground service management.

Although the agent model implemented in the system includes mental states (such as goals and beliefs), the perspective is a non-individualistic one. Social action is modelled as affected by the structure within which actions are framed. However, no attention is paid to the question about how the environment can affect social action without affecting the agents. Shouldn't this influence be modelled at the level of agent rather than being modelled at the level of action?

An example of analysis using the classic bottom-up direction of influence, from individual agents to social groups, is presented in chapter five, where Huberman and Glance address the fluctuations of individual contributions to global utility. Simulations results indicate that such fluctuations may have a dramatic impact on the average utility of a group. In other words, this utility decreases when individuals' contributions to the global benefit fluctuate strongly. The authors consider this a dynamic explanation for the fact that large groups are more likely than small groups to get involved in collective action, pointing to fluctuations as stronger in small groups.

Although interesting per se, this chapter is out of step with the rest of the volume. Whereas many contributors stress the constraining and influencing role of organisations, Huberman and Glance investigate how group properties like utility emerge from properties of individual action (fluctuation).

Organisations as Entities

The modelling of the organisation as a constraining framework is exemplified in the ontological approaches presented in two chapters. In chapter eight, Scacchi presents a taxonomy of resources within an organisational process meta-model (OPM). As the author says, organisations are "subclasses of agents performing tasks using tools and systems that consume or produce resources".

Analogously, an engineering approach to organisational ontology is presented in chapter seven, which is one of the most interesting contributions included in the volume. This approach consists of finding out which questions the ontology is constructed to answer. Indeed, ontology competence is the rationale for the ontology itself. Thanks to the influence of Weber's (1978) view of organisations as sets of constraints on agents' behaviours, Mintzberg's (1983) informal analysis of organisational features, and Winograd's (1987/1988) theory of communication, the authors provide a meta-model of organisations. This is used to specify the structure, behaviour, authority, empowerment and commitment (as well as the roles, hierarchy and communication limits) of any organisational entity.

Organisations as Agents

The last two chapters are of particular interest not only from the information technology perspective, but also from a classic social scientific one. In these contributions, the attempt is made to model organisations as "agents" accomplishing tasks and doing cognitive activities like "remembering" or, for that matter, "forgetting". In chapter nine, Kaplan and Carley present COMIT as a computational framework where three components of organisational settings (individual decision-making, organisations and information technology) are integrated. In the following chapter, Sandoe gives more substance to the idea of an organisational entity. The role of organisations in the processes of remembering and forgetting is analysed and three types of memory - structural, mutual and technological - are distinguished. From this perspective, one should not forget the classic anthropological work by Douglas (1986) which is not cited in this chapter.

This approach is anticipated in the introduction, where a new science of organisations as webbots is proposed. These are computational entities that include autonomous agents across which tasks and knowledge are distributed. Hence, the importance of computational and simulation techniques for an innovative study of organisations.

Along these lines, the first chapter by Carley and Prietula is intended to show the impact of a webbot architecture (essentially SOAR) for simulating organisational behaviours and their effects on organisational performance. In line with experimental evidence on co-operation (Ostrom 2000), the paper emphasises the necessity for taking social and cognitive capacities into account. Webbots are systems endowed with a number of specific skills, along with cognitive and social capacities allowing them to explore communication networks in an intelligent way and provide relevant and reliable information to their users. For this reason, the main question addressed is the extent to which trustworthiness is crucial for efficient performance. Indeed, social cognition issues such as trust or reputation are receiving growing attention in many domains where automated intelligent systems are applied, first and foremost in agent-mediated interaction. Not surprisingly, the simulation studies discussed in this chapter indicate that unreliable providers do increase the cognitive effort required by the organisational system. However, this effect is milder than expected. On one hand, untrustworthy webbots cause recipients to waste resources and activity - as happens when users go to a given site but do not find the expected information. On the other hand, however, perceived unreliability reduces the number of communicative actions since agents are discouraged from acquiring information via others. Communication causes the duration of organisational activity to increase in time and time is a precious resource to be invested with moderation. Hence, the negative effects of untrustworthiness are less dramatic than expected. Despite the intuitive truth of these results, we probably expect something else from simulation of social cognition, which is more than a benevolent or malevolent attitude towards others. Social cognition is a rather complex configuration of the mind that integrates social beliefs and goals, emotions and motivations, reasoning, deciding and acting capacities. It consists not only of reasoning about others' actions but also about others' minds. A limited view of social cognition might be good enough even for some important objectives, such as measuring the effect of unreliability on organisational performance, but a more complex view is needed for qualitative analysis. For example, what about the socially undesirable exploitation of trust: acquiring a good reputation in order to negotiate with trusting agents to one's own advantage and then possibly cheating them?

Why the Volume is Still Fresh and What Can Be Found In It

Approaches based on computational modelling and experimental simulation are perhaps the most promising in the study of social phenomena. As pointed out in the conclusions, this approach is common to all the contributions included in the volume. Organisations are defined as computational entities, distributed algorithms that can be run on computers, and therefore are approached with computational instruments. In addition, organisations are not only modelled by means of computational instruments; they are also experimented upon by means of computer simulation. Simulation adds value to computational modelling, allowing different organisational models to be implemented, visualised and compared.

As recalled in the conclusions, this allows a number of important questions to be explored: which factors affect organisational and group performance? How to avoid errors and what types of errors are more likely to occur in which circumstances. To what extent is trustworthiness essential for co-operation? Which factor is more influential with regard to the efficiency of performance: the structure of the organisation or the structure of the task? What is the effect of time pressure on organisational performance and the accuracy of decisions?

The idea that one can not only model but also make experiments on entities so complex and yet so volatile as organisations is refreshing. As Burton emphasises, the importance of this type of modelling is to be found in its capacity to reflect the real world, but not in that of mimicking it. The aim is not to reproduce one or more real organisations but to explore and model ideal-type organisations; formulate, and possibly control, hypotheses about general organisational mechanisms; determine factors (or set of interactive factors) responsible for relevant effects.

It should be noted that the volume has not aged much in the five years since it came out. One reason could be that organisations are shown to revolve around the notion of agency. Firstly, organisations are sets of agents (both natural and artificial) achieving co-ordination. Secondly, and more interestingly, organisations per se are often assigned the properties of agents, including the capacity to accomplish mental actions (as in chapters nine and ten). The notion, theory and field of agents has proved fairly promising in the last few years (consider the success of AgentLink). Developments continue particularly in the social sub-fields, for example, the Agent-Based Social Simulation Special Interest Group within the AgentLink Network of Excellence and the high-quality workshops on Multi-Agent Based Simulation (MABS) co-located with ICMAS '98, ICMAS '00 and AAMAS '02. Furthermore, the MAS community (represented by the ICMAS conference) has now been unified with the more formal agent community (represented by the ATAL workshop) and the application-oriented one (represented by the Autonomous Agents conferences) in one and the same event: AAMAS started in Bologna in 2002). This shows that one of the main domains for application of agent systems is necessarily social organisational.

What is Not Found in the Volume and Why

While reading the book here and there, one is impressed by two contrasting facts: on one hand, organisations are treated as entities; on the other, they are often equated with aggregates of agents in co-ordination, or even more weakly, with aggregates of agents variously constrained.

This variability in the notion of organisation is baffling. Of course, the population of car drivers who necessarily achieve a variable (though sub-optimal) degree of co-ordination can be seen as the result of some organisational process. Still, it is something completely different from the existence of entities like General Motors or Al Qaeda. Undoubtedly, the variety of social and collective actions is a puzzle that only computational and simulation techniques can help us to solve, not only by providing frameworks for task or structure specification, but also for theorising about organisational entities.

However, the potential of agent-based research is not fully exploited in the work presented here. Often, it is essentially used to model classes and subclasses of agents that organisational tasks are assigned to, rather than to explore what type of agent an organisation is, what its properties are and so on. This kind of investigation aims to do something quite different from working out meta-models for task specification. It aims to develop an explicit theory about what organisations are, why they are formed and how they work. This of course includes (but does not reduce to) modelling the process in which a task is identified, formulated and assigned. Interestingly, by adopting the agent theory perspective, one can try to theorise about organisations as agents at a further level of complexity. An ontology or meta-model is a domain-independent framework for specifying task and sub-tasks for classes and subclasses of agents. A theory of organisational agency would probably lead us to apply an integrated agent architecture at the level of organisations.

A similar (but not identical) way to formulate this problem is in terms of theory-model alternative. Are descriptive models, of the type exemplified by T�MS or by the ontology presented in the chapters, good enough to understand and possibly manage organisations, their performance, and the factors affecting it? Don't we need a further level of explanation, i.e. a theory of:

• why and how organisations are formed,
• how they affect the agents' minds and not only their behaviours;
• whether and how they are represented in the agents' beliefs;
• whether, how and why agents accept at least part of the organisations goals as their own;
• which modifications in the structure of individual agents' social reasoning, intentions and actions are owed to the organisational framework;
• what structure does commitment display within an organisation, which levels and inter-connections among various levels of commitment are implied
• what about responsibility and its delegation
• what about the mechanisms of social control, reputation, gossip and so on

The definition in terms of sets of constraints on agent's behaviours is good enough to produce meta-models of the type presented in the volume, but is insufficient to answer the question as to what type of entity an organisation is.

Is this further level of investigation needed for theoretical-scientific purposes? It is, but the role of such a theory is also pragmatic for at least two reasons.

First, it is needed to highlight the interrelationships between the properties that individuals have on their own account and those that they derive from their organisations, for example the goals they adopt, the powers they inherit, the roles they play and the tasks they accomplish. These are not simply added to the list of properties characterising individuals, but have a special status in their minds according to which they are manipulated. What are the consequences of positive or negative interrelationships among individuals' properties with different status and from different sources?

Secondly, and more importantly, properties recur at the entity level. Like individuals, organisations have commitments, powers, goals, responsibilities, reputation and obligations. How do these properties interact with those of the affiliated agents? What are their effects on the individual agents? If these belong to an organisation, they derive not only specific tasks and powers, obligations and responsibilities, but inherit further properties. For example, they will inevitably share (at least to some extent) the reputation of their organisations and will have to put up with it.

References

DOUGLAS M. 1986. How Institutions Think. Syracuse University Press, Syracuse, NY.

MINTZBERG H. 1983. Power In and Around Organizations. Prentice-Hall, Englewood Cliffs, NJ.

OSTROM E. 2000. Collective action and the evolution of social norms. Journal of Economic Perspectives, 14:137-158.

WEBER M. 1987. Economy and Society: An Outline of Interpretive Sociology. University of California Press, Berkeley, CA.

WINOGRAD T. A. 1987/1988. A language/action perspective on co-operative work. Human Computer Interaction, 3:3-30.

Return to Contents of this issue

© Copyright Journal of Artificial Societies and Social Simulation, 2003