A Context-Based Representation of Knowledge

Download PDF
0 downloads 0 Views 542KB Size Report
Comment
to this human level. This level supports the dynamic building of upper level structures that help information flow more rapidly and easily among individuals.
A Context-Based Representation of Knowledge Flows in Dynamic Organizations Alexandre Gachet1,2, Patrick Brézillon3 2

ITM Department – University of Hawaii at Manoa 2404 Maile Way, A202 Honolulu, HI 96822, USA [email protected]

3

LIP6 – University Paris 6 9, rue du Capitaine Scott 75015 Paris, France [email protected]

ABSTRACT.

Traditional companies used to be authoritarian and built around hierarchical organizational charts. Facing more dynamic environments in recent history, many companies experimented with newer forms of organization, usually cutting down long decision paths and fostering lean structures able to react to rapid changes in the marketplace. One crucial difference between both types of organizations is the way information and knowledge flow inside the structure. Based on an existing multi-level model of organizational structures, this article introduces a context-based representation of knowledge flows in dynamic organizations. It is based on an innovative organization of context along two dimensions: one horizontal and one vertical. An example scenario is provided to illustrate the proposed model.

RÉSUMÉ.

Les entreprises furent longtemps organisées autour d'organigrammes hiérarchiques et autoritaires. Confrontés à des environnements de plus en plus dynamiques, de nombreuses entreprises se mirent à expérimenter de nouvelles formes organisationnelles caractérisées par des chemins de décision raccourcis et des structures allégées capables de réagir aux changements rapides du marché. La manière dont les informations et les connaissances circulent au sein de chaque structure représente une différence cruciale entre ces deux types d'organisations. Basé sur un modèle multi-niveaux positionnant les structures organisationnelles, cet article propose une représentation contextuelle des flux de connaissances au sein d'organisations dynamiques. Cette représentation repose sur une organisation innovative du contexte le long de deux axes: l'un horizontal et l'autre vertical. Un exemple est proposé afin d'illustrer ce modèle.

KEYWORDS: MOTS-CLÉS : 1

organizational structures, contexts, knowledge management. structures organisationnelles, contextes, gestion des connaissances.

Research partly supported by Swiss NSF grant Nr. PBFR2-104340.

“Like change, the idea of adhocracy is not new. But the idea that adhocracy can and should be managed in tandem with bureaucracy is new. We desperately need management theory and practice that codifies this proposition.” (Waterman, 1992, p. 115) 1. Introduction Most of today's companies are built around organizational structures ranging from bureaucracy to adhocracy. One crucial difference between both structures is the way information and knowledge flow inside the structure. In the bureaucracy, they flow bottom-up along a hierarchical path, before coming down along a different hierarchy. In the adhocracy, hierarchical ties are relaxed and information and knowledge mostly flow through lateral relations (Orlikowski, 1991). Gachet and Brézillon propose a four-level organizational model to show how an organization can dynamically move from bureaucratic to adhocractic structures, according to changing contexts and focuses of attention in its environment (Gachet and Brézillon, 2005). The proposed model helps managers and enterprise architects to better describe and understand the structures being built and dissolved inside the organization. Nevertheless, the model still needs to be completed with a formal identification of the different types of context existing at each level. As a matter of fact, the quality of the collaborative work in a group depends essentially on the cohesiveness of the context shared by all group members. However, the nature of this shared context – and thus the quality of the collaborative work – is different for each type of organizational structure. In other words, an unappropriate organizational structure used to deal with a given problem can miss the real objective of the problem-solving process. The main purpose of this article is to provide a context-based representation of knowledge flows within dynamic organizational structures. A secundary purpose consists in shedding a new light on the research on context, explicitly bringing a horizontal and a vertical dimension to the field. To reach these goals, this article focuses on the explicit consideration of context in the representation of knowledge and reasoning, an approach rooted in knowledge-based systems and artificial intelligence (AI) research. It proposes a new model analyzing how knowledge is exchanged and elicited between individuals and actors before, during, and after structural changes inside the organization. This article is organized as follows. Section 2 reproduces the main results associated with the four-level model presented in (Gachet and Brézillon, 2005). Based on this model, Section 3 presents the various forms of knowledge flows encountered at each level of the model. Section 4 presents an example scenario illustrating how this model can be used in practice. Finally, Section 5 wraps up the article and describes the future steps of the project.

2. A Four-Level Model of Organizations Gachet and Brézillon define a multi-level model categorizing various types of organizational structures (such as social networks, enterprises, communities of practice, task forces, and other related organizational structures2) into four distinct levels (Gachet and Brézillon, 2005). Figure 1 presents an integrative view of this model.

Figure 1. The four levels of the organizational model (Gachet and Brézillon, 2005) Communities (such as communities of practice and virtual communities) are generated from social networks or from within organizations, when a focus of interest appears. They represent semi-permanent structures grouping actors with convergent, long-lasting goals. Short-lived, unpredictable events trigger missions that bring together actors of various communities, organizations, and social networks in a task force whose lifetime is relative to the time necessary to solve the problem. Once the problem is solved, the mission dies out and the members of the task force slip back into their respective structures, enriching them with chunks of knowledge acquired during the problem-solving process.

2

A complete presentation of these organizational structures can be found in (Gachet and Brézillon, 2005).

Figure 1 also outlines the boundaries of our four-level model. The purpose of this model is to show that a decision group can be seen as a task force built and assembled on top of communities (groupings of decision-makers or available actors able to fulfill required tasks), organizations (for example, in the case of an audit), and/or external individuals (for example, when external experts are called to help solving a very specific problem). The next sections review the four proposed levels of the model: human level, organization level, community level, and adhocracy level. 2.1. Human Level The human level is represented at the bottom of Figure 1. This level plays a double function in our model: it supports and encompasses the other levels at the same time. As shown in Figure 1, it supports the other levels because individuals are the elementary constituents of the entire model. In other words, the three upper levels are made up of individuals inevitably coming from social networks belonging to this human level. This level supports the dynamic building of upper level structures that help information flow more rapidly and easily among individuals. However, the human level also encompasses the other levels, precisely because actors in an organization, a community, or an adhocracy still belong to the grassroots human level. This second function is a direct recognition of the concept of “embeddedness” (Granovetter, 1985), that is to say “the argument that the behavior and institutions to be analyzed are so constrained by ongoing social relations that to construe them as independent is a grievous misunderstanding” (p. 482). In other words, an individual seldom belongs to only one level of the model; at any time, he belongs to several levels, but in varying degrees. Moreover, the time spent in an upper-level organizational structure influences the number of ties in the social networks of this human level, as well as the strength of existing ties. 2.2. Organization Level The meaning of the term organization adopted in this model remains closely related to social perspectives. For our purpose, we consider an organization as “a combination of human effort in a relatively stable network of social relations” (van Aken, 1982). An organization is structured around a concern related to the organization itself (i.e., the enterprise). The organization level is closely related to bureaucracy concepts. In sociological theories, bureaucracy is an organizational structure characterized by regularized procedures, division of responsibility, hierarchy, and impersonal relationships. According to (Weber et al., 1947), the attributes of modern bureaucracy include impersonality and the implementation of a system of authority that is practically

indestructible. Toffler (Toffler, 1970) sees it as a network of roles fulfilled by individuals (in opposition to a network of individuals, such as a social network). Bureaucratic organizations usually deal with routine operations. Instances of the organization level naturally include enterprises, but also other forms of organizations, such as nongovernmental organizations (NGO) or even political structures, such as congresses. 2.3. Community Level The term community has merely two opposing meanings. On the one hand, a community can be defined in a bottom-up way as a unified body of individuals emerging from an existing social network with some shared element. On the other hand, a community can be defined in a top-down way as a society at large, a local group of people that live in the same geographical area. Physical proximity leads the individuals to engage in social relationships. For the purpose of this model, only the first, bottom-up meaning of the term is retained. As shown in the middle section of Figure 1, communities are rooted in the organization and human levels. A community structure emerges when a focus on a specific domain arises among the individuals of an existing social network and/or organization. This shared concern, whose substance can vary widely from community to community, gives the community a collective context and individuals organize as actors with roles. Interaction inside a community is usually informal and spontaneous rather than procedurally formalized. Instances of the community level include communities of practice, virtual communities, communities of interest, and other forms of communities. 2.4. Adhocracy Level The term adhocracy was first coined by Warren Bennis (Bennis and Slater, 1968) and subsequently used by (Toffler, 1970) and Mintzberg to describe a structural configuration that “is able to fuse experts drawn from different disciplines into smoothly functioning ad hoc project teams” (Mintzberg, 1979). An adhocracy represents any form of organization capturing opportunities, solving problems, and getting results (Waterman, 1992). Beairsto (Beairsto, 1997) defines it as “the term used to describe the flexible structure of multidisciplinary teams which is best suited for complex tasks in a dynamic and unpredictable environment.” It can be characterized by shared values across various splinter groups, cultures, and individuals. It relaxes hierarchical ties and promotes lateral relations (Orlikowski, 1991). An adhocracy is not organized around formal rules or regulations, and it does not provide standardized procedures for dealing with routine problems. It is instead a response to environmental pressure (Mintzberg and Quinn,

1996), meant to cope with exceptional situations and adapt quickly to changes within its environment. In the context of the enterprise, an adhocracy allows teams to make decisions without approval from higher-level members of the organizational chart. Adhocracies are traditionally at work in high risk organizations or in emerging industries. The lifetime of an adhocracy is usually limited. As shown in the upper part of Figure 1, an adhocracy structure emerges from one or several communities, organizations, or human structures when a mission arises in the context of the enterprise. This mission gives the adhocracy a shared context and all the actors become organized toward a unique goal: accomplish this mission. Instances of the adhocracy level include task forces, working groups, and virtual teams. 2.5. Discussion One of the primary purposes of this multi-level model is to provide managers and enterprise architects with a tool to better describe and understand the back-andforth movement of internal organizational structures from bureaucracy to adhocracy, according to changing contexts and focuses of attention. Whereas the levels of the model describe different forms of organizational structures, the focal point of the dynamic movement is set on the individuals and actors interacting inside these structures. Individuals and actors (depending on the level where they are) can be seen as information processing nodes acting like the vertices of a graph whose edges would support flows of knowledge. The critical aspect, then, is to understand how these flows of knowledge are organized and to find a relevant representation to describe them in the context of the four-level model. The next section offers answers to both questions. 3. A Context-Based Representation of Knowledge Flows The origins of the explicit consideration of context as a research field of its own are traditionally associated with McCarthy's notes on formalizing context (McCarthy, 1993). Since then, context has been adopted by the artificial intelligence (AI) community in any domain where understanding, reasoning, problem-solving, and learning are needed. For the purpose of this article, the concept of context is closely associated with the concept of knowledge. Brézillon and Pomerol consider three parts for knowledge, namely the external knowledge, the contextual knowledge, and the proceduralized context (Brézillon and Pomerol, 1999). The contextual knowledge (CK) represents the part of the context that is relevant for the current step of a decision process, whereas the external knowledge (EK) represents the part of the context that is not relevant at that point. The proceduralized context (CK) is the part

of the contextual knowledge that invoked, assembled, structured and situated according to the focus of the current step. By analogy, one can consider the proceduralized context at a given step of the process as the explicit instantiation of the underlying contextual knowledge, which remains implicit. A given focus and its associated context are interdependent. The context is dynamic (Brézillon, 2003a; Brézillon, 2003b). It follows a pendulum-like movement between contextual knowledge and proceduralized context during the evolution of the process. Pieces of contextual knowledge are proceduralized when explicitly needed during the process, then go back to the contextual knowledge when the process moves to the next step. A piece of proceduralized context going back in the contextual knowledge becomes a “chunk of contextual knowledge” a la Schank (Schank, 1982) that can be recalled later as a whole in a new proceduralized context. It naturally follows that the more experienced an operator is, the larger his available structured knowledge becomes. Recent research tends to show that this bi-directional movement between contextual knowledge and proceduralized context can follow horizontal or vertical paths. The next two sections briefly present the horizontal and vertical views of knowledge exchanges. 3.1. A Horizontal View of the Knowledge Flow Figure 2 illustrates the horizontal flow of knowledge in the context of a single actor, at a given step of a problem-solving process. A complete discussion of the dynamics of the horizontal view is presented in (Brézillon and Araujo, 2005). In accordance with the definitions of external and contextual knowledge presented in the previous section, the actor polarizes his knowledge by grouping in his contextual knowledge (CK) all the knowledge that is relevant for the current step of the process and by leaving the rest of his knowledge in the external knowledge (EK). This organization of knowledge should be regarded as a snapshot of the process in time. The evolution of this organization is actually continuous. Pieces of contextual knowledge that were deemed relevant eventually prove irrelevant and move back to the external knowledge. Conversely, pieces of external knowledge that were deemed irrelevant suddenly become necessary and relevant and are imported in the contextual knowledge (arrow 1 in Figure 2). This typically happens when the PC cannot be built with the available contextual knowledge; new pieces of knowledge need to be imported from the external knowledge. Although contextual knowledge exists in theory, it is actually implicit and latent, and is not usable unless a goal (or intention) emerges as a focus. It is only when the actor's attention is focused on a specific event influencing the problem-solving process that part of the contextual knowledge is proceduralized (PC), that is to say assembled, structured, and instantiated (arrow 2 in Figure 2). When the current focus disappears, this piece of PC moves back in the contextual knowledge and

becomes a chunk of contextual knowledge (arrow 3) that can be recalled later as a whole in a new PC. Finally, when the problem-solving process itself is terminated and/or replaced by a new one, the chunk of contextual knowledge goes back into the external knowledge and the entire flow starts again from the start.

Figure 2. Generic, horizontal view of context

3.2. A Vertical View of the Knowledge Flow The horizontal view of the knowledge flow represents a strong formalism to describe how a PC is built within the context of a single actor. The process changes radically, though, when several actors are involved in a collaborative way. We then need to introduce another dimension – the vertical view – for representing this second form of knowledge flow. Figure 3 illustrates this view in a generic form studying the interaction context of two actors (Actor A and Actor B). It is an extension of the contextualization concept proposed by (Brézillon, 2005). The transformation of contextual knowledge into a proceduralized context corresponds to a transfer of contextual clues from a context at one level to the context of the upper level (the interaction context in Figure 3). The top part of Figure 3 represents how the PC is built from the contextual knowledge of two actors during their interaction at a given step of the problem-solving process (the process can easily be generalized for n actors but has been scaled down to two actors in Figure 3 for readability purposes). The interaction context contains the pieces of contextual knowledge that each actor brings from his individual context to publicly share them with other actors. These pieces of contextual knowledge are then assembled and structured by the two actors during their interaction to constitute the PC needed at the given step of the problem solving process. Each actor keeps his own PC (at the actors context level), which can be seen as his own mental representation of the collaborative work. This PC is built according to the horizontal view presented in the previous section.

Figure 3. Generic, vertical view of context Once the PC of the interaction context has been used, it then becomes a piece of the actors' shared context. A longer discussion on this topic can be found in (Brézillon, 2003b). As a consequence, a strong tie is established between the actors through the construction of this PC and its subsequent move into their shared context. This tie will persist between them, even after the interaction context disappears and the actors go back to a lower level. The bottom part of Figure 3 is needed to express the recursive relation that can exist at each level of the model of organizational structures presented in Section 2. According to our definition of context, the contextual knowledge at one level is transformed by interpretation into PC at the next level. Thus, an element of context at one level is at the same time contextual knowledge at the more generic level (down) and proceduralized context at the more specific level (up). Finally, the fact that the PC is stored into each actor’s context implies that there is a shared part of the individual contexts which is progressively developed. The vertical view of context unleashes its full potential when used in conjunction with the multi-level model of organizational structures introduced in Section 2. Figure 3 provides a generic representation of the vertical view. This view can be recursively applied to the four levels of the organizational model. For example, if the actors context of Figure 3 corresponds to the organization level in Figure 1, then the foundation context would be part of the human level and the interaction context part of the community level. If the actors context corresponds to

the community level, then the foundation context is part of the organization level and the interaction context part of the adhocracy level. In a metaphorical way, it would be possible to vertically slide a transparent view of Figure 3 on top of Figure 1. Note however that the human level is the lowest level of the model and does not have a foundation context. At the other end of the model, members of an adhocracy do interact in an interaction context, but this interaction context never materializes as an independent organizational structure. This explains why there isn't a one-toone relation between the organizational model and the vertical view. It naturally follows from Figure 3 that structures at each level are associated with a shared context. From a static perspective, the shared context between two actors is the intersection between their contextual knowledge (as in Figure 3). From a dynamic perspective, however, two parts must be distinguished in any shared context, namely the external shared context and the internal shared context. From an internal viewpoint, the foundation context contains generic contextual knowledge, including general policies, roles and tasks to accomplish, rules, constraints, and objectives. This contextual knowledge is proceduralized for the specific problemsolving and strategies applicable by the actors (within their own contexts). In other words, the internal context is intertwined with the generating factor of the corresponding level (discriminating factor, vision, focus, or mission). From an external viewpoint, the organizational structure of any given level interacts with other organizational entities and has an interaction context in a group of larger scope (e.g., a market, an transnational project, etc.), that is a PC corresponding to the contextual knowledge in the group. In other words, the external context contains relevant information about the environment of the organizational structure itself. 4. Example Scenario The purpose of the following example scenario is to illustrate the vertical view of knowledge flow in a typical organizational context. This example also relies on the four-level model described in Section 2. Another example illustrating the horizontal view is provided in (Brézillon and Araujo, 2005). 4.1. Presentation The case study is based on a fictitious firm called Robots Manufacturers, Inc. Robots Manufacturers produces ten different types of robots (named Robot1 to Robot10). Three production steps must be carried out3: 1.

3

Production of the components, according to various sequences of processes.

This example scenario is inspired from (Gachet, 2004), p. 20-52.

2.

Mounting, which takes several hours for each type of robot, with a limited work capacity per week.

3.

Calibrating and testing, which takes several hours for each type of robot, with a limited work capacity per week.

Each type of robot has a different selling price and is ordered in various quantities. Individual components can be produced in-house (by the components manufacturing team of the company) or bought (by the purchases team). Stored robots are managed by the warehouses unit of the production department. The firm pursues two main strategies: maximizing its selling profit and minimizing its variable costs. Production planning is established on a monthly basis (based on the order book managed by the sales department). 4.2. Normal State It is the start of a new month. Bob, a specialized worker belonging to the components manufacturing team, starts his afternoon shift on a casting machine tool. The shop foreman just briefed Bob and his colleagues on the monthly planning. The order book is full and, due to high market prices for the purchasable components, the top management decided to produce as many components as possible in house. The coming weeks promise to be busy. Bob has been working with the casting machine for about one hour when he detects worrying vibrations during the casting process. He immediately monitors the temperature dial and sees that the machine tool is dangerously overheating. Bob knows the machine very well and reacts quickly and appropriately to avoid irreversible damage. He launches the emergency shutdown process. 4.3. Alert State Bob alerts the shop foreman at once. After a first diagnosis, both men are afraid that the machine tool suffered a major breakdown and will be out of order for a period depending on the time needed by the maintenance company to fix it. The shop foreman alerts the production manager, who in turn informs the director and immediately contacts the company with which Robots Manufacturing, Inc. has a maintenance contract. An expert is on site about one hour later. He checks the machine tool and confirms the severity of the breakdown. Spare parts have to be ordered abroad. The repair will need at least one day, and the machine tool will need a complete check up and recalibration before being used in the production line again. The expert forecasts that the entire process will take about five working days.

4.4. Design and Implementation States Hearing the bad news, the director meets with his production and sales manager to see how to adapt the monthly planning and to analyze which units in his company would be affected by the new plan. He wants to be sure that the proposed plan can be actually implemented. Finally, the planning unit works in close cooperation with the shop foreman to prepare a reorganization plan for the production line. The shop foreman knows the skills and affinities of his workers very well. He is able to say who works well with whom and, ultimately, who should be affected to which machine to carry out the plan. 4.5. Organizational Structures and Knowledge Flow The above example scenario, even though simplified, remains in essence typical of organizational decision processes. A problem is detected, people that will be involved in the decision gather together, information is collected, alternatives are identified and evaluated, and a decision is made, implemented, and assessed to be sure that it does not lead to new problems. In this section, we revisit the informal presentation of the scenario using the vertical view of the knowledge flow introduced in Section 3, as well as the four-level model introduced in Section 2. When operations are normal, the enterprise adopts a bureaucratic attitude: information circulates among individuals along the hierarchical network of the organization, each individual using or not the information for its personal goal without correlation with the goals of other individuals (the “normal state” in Section 4.2). This describes well the normal flow of information in the company: customers order robots from the company, the purchases team orders components from the suppliers, the shop foreman briefs the workers on the planning, etc. All this information flows within the organization without problems being detected. Whereas the global organizational chart of Robots Manufacturers, Inc. determines the structure of the enterprise, its breaking up in departments and units gives a good idea of the communities that emerge within the company and its environment (for example, the executive committee of the company, horizontally including the senior executives of all the departments, or the manufacturing community of practice (CoP), including specialized workers on the production line and representatives of the machine tool manufacturer). In that case, an organizational chart is not prescriptive, but rather descriptive. Actors are naturally allowed to form communities spanning several units and/or departments. However, when the unpredicted machine breakdown occurs, a mission (“problem to fix”) appears in the context of the company (“normal state”). Both the mission and its context have to be considered jointly. Both mobilize a group of

individuals in the enteprise and its communities that have a same concern for the mission and are sensitive to its context. Together with the mission and its context appears an organization of tasks to accomplish and of roles that individuals of the group must play. Individuals with roles become actors and the actors group becomes assembled, organized and structured as a task force. At that stage, the functional ties between the actors of the enterprise (i.e., the organizational chart) tend to fade in the background during the assembling of the task force and are overlaid with the relationships between roles. In other words, roles are attributed to individuals according to the needs of the (mission, context) pair more than according to the function of the individuals in the underlying organization. This implies that the organization of actors in a task force does not correspond necessarily to the hierarchical organization of the firm. A complete study of the various roles played by the individuals to fulfill the mission has been presented in (Gachet and Brézillon, 2005). In this section, we focus on the exchanges of knowledge happening between the actors, based on the model introduced above. Figure 4 uses the formalism described in Section 3.2 to give an overview of these exchanges during the alert state of the scenario. At the human level (bottom of Figure 4), all the human beings involved in the scenario are individuals. In the context of the (“problem to fix”, alert state) configuration, they all belong to a same social network whose discriminating factor could be “Working for Robots Manufacturers, Inc.” This social network includes not only the employees of Robots Manufacturers, but also individuals working for other companies doing business with Robots Manufacturers (for example, the maintenance company). The contextual knowledge of each individual contains work-related knowledge (such as his professional background), as well as knowledge influencing the individual's working environment (for example, the knowledge that someone in the family circle of the individual is ill can influence the performance of the individual at work). The external knowledge of each individual contains knowledge unrelated to work or without influence on the individual's working environment. For example, the individual loves fishing in his free time and has a lot of knowledge about this activity, but this knowledge is not relevant in the current configuration. Still in the same (“problem to fix”, alert state) configuration, but at the organization level (second level in Figure 4), the same individuals become actors with roles in their respective organizations. The individual Bob becomes a specialized worker for Robots Manufacturers, Inc., William the shop foreman, George the production manager, and James a maintenance expert working for a different organization called General Maintenance, Ltd. At this level, the knowledge of the actors has a better focus on the (mission, context) configuration than at the human level. The contextual knowledge of Bob contains technical knowledge about the failing casting machine, the shop foreman is well aware of the role of this machine in the production line, and the production manager about the role of production line in the enterprise and the impact that a glitch in the production line

could have for the enterprise. Moreover, the contextual knowledge of the maintenance expert contains his expertise knowledge about the failing machine.

Figure 4. Knowledge exchanges in the (“problem to fix”, alert state) configuration Even though all actors in the enterprise are aligned on the vision of the enterprise (which in the case of Robots Manufacturers is to make profit by producing and selling robots), each one follows independent, modular goals in his day-to-day activities. The goal of Bob as a specialized worker is not the same as the goal of William as a shop foreman, nor as the goal of George as a production manager. Because of this absence of shared goal between actors, the organization level as depicted in Figure 3 is not an interaction context for the human level. In other words, individuals at the human level proceduralize chunks of contextual knowledge for their own goals at the organization level, not for a collective goal shared by all actors.

However, this situation changes at the next, community level. For clarity purposes, Figure 4 singles out the PC of the specialized worker and the shop foreman in the manufacturing CoP of the enterprise. Unlike the organization level, Bob and William now have an explicit focus of interest (the failing machine) and a convergent manufacturing goal (to find what is wrong with the machine). From his technical knowledge about the failing machine (organization level) Bob the specialized worker proceduralizes the knowledge that the machine is overheating and must be shut down. From his knowledge about the role of the machine in the production line, the shop foreman proceduralizes the knowledge that the machine is out-of-order and disrupts the production. Both chunks of knowledge enters the PC of the manufacturing community of practice (CoP), whose contextual knowledge already contains knowledge about similar failures in the past. The main role of the community level in that situation is to provide an interaction context for individuals and/or actors sharing a domain interest. Finally, as the “problem to fix” mission is not a pure manufacturing problem, but a problem impacting the entire enterprise, a task force is built at the adhocracy level to accomplish the mission. Based on the knowledge regarding similar failures in the past, the manufacturing CoP proceduralizes the knowledge that the manufacturing team cannot work with the machine. Always in the alert state, the production manager joins the task force and, from his knowledge about the role of the production line in the enterprise, proceduralizes the knowledge that he needs to call a maintenance expert as soon as possible to minimize the impact of the failure on the production activities. The maintenance expert joins the task force and, based on his expertise knowledge regarding the failing machine, proceduralizes the knowledge that the machine is unavailable for about five days. At that stage, the organizational context moves from the alert state to the design state and the task force is reshaped accordingly4. The task force provides an interaction context for the individuals and/or actors trying to accomplish the mission. In the design state, alternatives have to be identified and evaluated. The production manager proceduralizes the sensed knowledge (“machine unavailable for five days”) into more useful knowledge (for example, “30% reduction of capacity for some production processes”) to propose a decision alternative that will be analyzed by the director (this proceduralization follows the horizontal view described in Section 3.1). The director can in turn ask for the advice of other executives (for example, the shop foreman if the alternative changes the production planning, the purchasing manager if the alternative affects purchases, or the inventory manager if the alternative changes the expected storage quantities). The production manager can produce as many alternatives as necessary, which can reshape the task force as often as necessary. However, once one of the alternatives is agreed upon by all the actors of the task force, the director assumes the role of the decision maker and decides to implement the new plan. This moves 4

A complete description of the changes of context is provided in (Gachet and Brézillon, 2005).

the enterprise context from the “design state” into the “implementation state”, thus creating a new (mission, context) pair and, again, reshaping the task force. The shop foreman turns the decision of the director into a reorganization plan and decides which specialized worker will work on what kind of machine tool. Table 1 summarizes the four (focus, context) pairs involved in our scenario, together with the corresponding social networks, communities, and task forces. State/Level

Human (SN)

Organization

Community

Adhocracy

Normal state

(All)

Robots Manufacturers, Inc.

Several (based on the company organization chart)

(None)

State/Level

Human (SN)

Organization

Community

Adhocracy

Alert state

Company + Maintenance contractor (discriminating factor: “Working for Robots Manufacturers, Inc.”)

Robots Manufacturers, Inc.

Manufacturing CoP (including employees on the production line and representatives of the manufacturer of the failing machine tool)

Specialized worker Shop foreman Production manager Maintenance expert

State/Level

Human (SN)

Organization

Community

Adhocracy

Design state

Company + Supplier (discriminating factor: “Supplying Robots Manufacturers, Inc.)

Robots Manufacturers, Inc.

Executive committee of the company Manufacturing CoP

Maintenance expert Production manager Shop foreman Purchasing manager Inventory manager Director

State/Level

Human (SN)

Organization

Community

Adhocracy

Implementation state

Company (discriminating factor: “Working in Robots Manufacturers' Production”)

Robots Manufacturers, Inc.

Manufacturing CoP

Director Shop foreman Specialized workers

General Maintenance, Ltd.

RS Electronics, Ltd.

Table 1. Organizational structures of the example scenario

In the normal state, there is no specific problem to solve and all available social networks in the context of the organization Robots Manufacturers are involved in the normal flow of activities. No particular mission requires the creation of an adhocratic structure. When the problem with the failing machine tool materializes, however, all four levels are reorganized toward its resolution. The alert state has been fully described in the previous sections. In the design state, the company needs to check if the new plan can be properly implemented without breaking existing agreements with partners. For example, it must check with its components suppliers if the quantities of components ordered can be modified or not. In the implementation state, the production line is reorganized to carry out the new plan, under the supervision of the shop foreman. 5. Concluding Remarks and Future Work This article has introduced a context-based representation of knowledge flows in dynamic organizations. One purpose of this representation is to analyze how knowledge flows and organizational structures are related in terms of knowledge elicitation, sharing, exchange, evolution, management, and preservation at the individual and group levels. It showed that the degree of contextualization of the knowledge flows increases along the bureaucracy-adhocracy axis of the four-level organizational model. This representation also sheds a different light on the field of knowledge management, moving the focus from static issues (knowledge storage) to dynamic issues (knowledge contextualization according to changing contexts and focuses of attention). A next step in this research will be to leverage the proposed representation in order to infer a framework of support technologies adapted to the different forms of knowledge flows. We believe that traditional knowledge management tools and systems do not pay enough attention to the dynamics of the underlying organizational structures and end up recommending systems that are not adapted to all contexts of the organization. Equipped with the representation proposed in this article, the multi-level model of organizational structures presented in (Gachet and Brézillon, 2005), and with a framework of technology support adapted to the different structures, organizations in general – and enterprises in particular – would benefit from a powerful tool to describe and understand the structures being built and dissolved inside the organization according to varying contexts and focuses of attention, as well as to describe and understand where the needs for new support technologies and/or systems integration lie. It is our hope that the model presented in this article provides a vehicle for researchers to develop new forms of dynamic organizational models.

Bibliography Beairsto, J. A. B., “Leadership in the quest for adhocracy: new directions for a postmodern world”. Ph.D. Dissertation, University of Tampere, Finland, 1997. Bennis, W. G. and Slater P. E., The temporary society. New York,, Harper & Row, 1968. Brézillon, P., “Context dynamic and explanation in contextual graphs”. Modeling and Using Context (Context-03), Palo Alto, CA, Springer Verlag, 2003b. Brézillon, P., “Representation of procedures and practices in contextual graphs”. Knowledge Engineering Review 18(2): 147-174, 2003a. Brézillon, P. “Contextualizations in a Social Network” Revue d'Intelligence Artificielle [this issue], 2005. Brézillon, P. and Araujo R. M. “Reinforcing Shared Context to Improve Collaboration.” Revue d'Intelligence Artificielle [this issue], 2005. Brézillon, P. and Pomerol J.-C., “Contextual knowledge sharing and cooperation in intelligent assistant systems”. Le Travail Humain 62(3): 223-246, 1999. Gachet, A. Building Model-Driven Decision Support Systems with Dicodess. Zurich, vdf Hochschulvlg, 2004. Gachet, A. and Brézillon P., “Organizational Structures and Decision Making Processes: A Multi-Level Model”, Journal of Decision Systems [forthcoming], 2005. Granovetter, M. "Economic Action and Social Structure: The Problem of Embeddedness." The American Journal of Sociology 91(3): 481-510, 1985. McCarthy, J. “Notes on Formalizing Context”. 13th International Joint Conference in Artificial Intelligence (IJCAI'93), Chambéry, France, 1993. Mintzberg, H., The structuring of organizations: a synthesis of the research. Englewood Cliffs, N.J., Prentice-Hall, 1979. Mintzberg, H. and Quinn J. B., The strategy process: concepts, contexts, cases. Upper Saddle River, N.J., Prentice Hall, 1996. Orlikowski, W. J., “Integrated Information Environment or Matrix of Control? The Contradictory Implications of Information Technology”. Accounting, Management and Information Technology 1(1): 9-42, 1991. Schank, R. C., Dynamic memory: a theory of reminding and learning in computers and people. Cambridge [Cambridgeshire] New York, Cambridge University Press, 1982. Toffler, A., Future shock. New York,, Random House, 1970. van Aken, J. E., On the control of complex industrial organizations. Boston, Kluwer-Nijhoff Pub, 1982. Waterman, R. H., Adhocracy: the power to change. New York, W.W. Norton, 1992. Weber, M., Henderson A. M. and Parsons T., The theory of social and economic organization; being Part I of Wirtschaft und Gesellschaft. London,, W. Hodge, 1947.