Communication and Knowledge Transfer Within a Community of Practice: an Agent-Based Model
Total Page:16
File Type:pdf, Size:1020Kb
Communication and knowledge transfer within a community of practice: an agent-based model Widad Guechtouli GREQAM – Paul Cézanne University Aix-en-Provence, France [email protected] 1. Literature background The analysis of knowledge formation and diffusion became very important for organization research since knowledge based economy appeared. It has been recognized that it represents a crucial asset that every organization should take care of, just like every other asset and yet, in a quite different way [Foray, 2000]. Knowledge is intangible and therefore is not easy to capitalize within an organization, or share between a set of individuals. Some authors studied the economic features of knowledge, as well as its transfer and sharing among individuals [Witt et al, 2007; Zacklad, 2004; Cataldo et al, 2001; Nonaka and Takeuchi, 1997; etc.] Knowledge is acquired through a learning process, and we can distinguish several levels of learning in the literature [Brenner, 2006; Vriend, 2000; Bala and Goyal, 1998; Dibiaggio, 1998; Cook and Yanow, 1996; Salomon and Perkins, 1998; Lave and Wenger, 1991, etc.]. This process is very important for the sharing and capitalization of knowledge within an organization. However, as stressed by Cohendet et al [2000], “The traditional vision of learning does not fully account for what actually occur within firms. It fails to acknowledge the importance of the complex cognitive structure of the firm from which stems learning processes.” The authors suggest considering a firm as a set of overlapping communities, and they studied the role played by such entities in the process of organizational learning. Starting from this point, it seems important to study the process of learning within these communities. We will focus on communities of practice (CoPs). This notion appeared for the first time in the early 1990’s, in the work of Lave and Wenger [1991]. A CoP is seen as one of the most efficient concepts to study the process of knowledge sharing in groups [Lave and Wenger, 1991; Brown and Duguid, 2000; Lesser and Storck, 2001]. There is a rich literature on CoPs in the knowledge management field. However, this concept has not been much explored in economics so far. Some authors though did work on the subject, focusing on the organizational and social aspects of these communities (see for instance; Cohendet et al, 2000; Dupouët et al, 2003; Dupouët and Yildizoglu, 2003; Diani and Muller, 2004). Economic literature shows that no particular research was led about learning 1 and knowledge transfer within a CoP. We consider that it is an important issue that we wish to address in this work. This paper aims at answering the following question: what impact do structures of communication have on knowledge transfer within a community of practice? In order to do so, we choose to compare two types of communication structures: communication through face-to-face interactions and communication through a web based forum. We want this work to be a contribution to the literature about modeling learning and knowledge transfer within communities of practice. Therefore, we will use agent-based modeling that can be considered as an appropriate tool to investigate learning in a social context. It allows the study of dynamic phenomena and captures complexity in a model [Phan, 2003; Taylor and Morone, 2006; Gilbert et al, 2001]. 2. The model: 2.1. The agents We have a population composed of 110 agents. Each agent is characterized by the following features: - Knowledge vector: Each agent in endowed with a knowledge vector, composed of 100 types of knowledge. An endowment of 0 in a knowledge k means that an agent doesn’t have that particular knowledge; whereas, an endowment of 1 means that he does. Let’s look at the example below: Knowledge 1 2 … 99 100 Knowledge endowments 1 0 … 1 1 Figure 1 Example 1: This agent has knowledge 1, 99, and 100, but doesn’t have knowledge 2. Knowledge is presented as a vector to capture knowledge diffusion. According to Cowan and Jonard [1999], models presenting knowledge as a scalar cannot apprehend this important process. Consequently, we will model knowledge as a vector composed of different sorts of knowledge. - Competence: it is defined by the number of knowledge an agent possesses. Technically, an agent’s competence is the sum of ones in his knowledge vector. This definition follows the work of Cataldo et al [2001]. In their model, these authors model each agent’s knowledge as a mask composed of 0 and 1 for different types of knowledge. They state: “The increasing number of pieces of information of a particular type, the more experienced the individual will be in that area. In addition, experience is represented in the number of ones that the knowledge mask has”. The ௧ . ݉݁ݐ݁݊ܿ݁ܥ competence of agent j changes across time and will be noted 2 - Memory: where an agent stocks information about past interactions (name of agents previously met and the answer given by each one of them). - Availability: defined by the number of questions that an agent can answer per time- step. - Tolerance threshold: defined as the number of unanswered questions that an agent is willing to accept from another agent, before deciding not to ask it anymore. According to these features, every agent is potentially a knowledge-seeker or a knowledge- provider, or both, according to its competence: In terms of answering questions and providing knowledge, we consider that the population of agents is divided in two parts: expert agents and intermediary agents. The members of the former have knowledge about the 100 subjects of an agent’s knowledge vector; they have a competence equal to 100. Whereas the latter members have a competence greater than or equal to a competence threshold (CompMin) defined as the minimal competence required in order to be able to answer questions. This threshold is set to 751. In terms of asking for knowledge, each agent that has a competence less than 100 is a knowledge-seeker. This includes intermediary agents as well. The initial structure of the population is the following: - 1 agent with an initial competence equal to 100; - 9 agents with an initial competence equal to 75; - 100 agents with an initial competence equal to 0. Knowledge- seekers Intermediary agents Expert agents New comers Knowledge- providers Figure 2 The structure of the population 2.2. Several sets of simulations All of these agents have a common goal: to learn by acquiring new knowledge and increasing their individual competencies. They interact in a social environment “consisting of a network of interactions with other agents” [Gilbert and Terna, 2000]. In order to model knowledge diffusion with two different types of communication structures, we will run several sets of simulations. 1 We led simulations for several values for CompMin and 75 is the value where the highest numbers of agents increase their individual competences. 3 - Simulations with face-to-face interactions: where we will consider a situation where interactions are based on a predetermined hierarchy according to a diffusion of competences within the community (simulations based on a predetermined hierarchy), and a situation where interactions are based on agents’ past experiences (simulations based on past experiences). - Simulations with interactions through a forum: where interactions happen on a one-to- all basis. We chose to model this type of interactions because this way of communication joins the definition of indirect knowledge transfer given by Witt et al [2007]. By contrast, interactions in the previous sets of simulations represented face- to-face interactions, or interaction through email or telephone, between two individuals. 2.3. Interacting - In simulations through face-to-to face interactions, an interaction is defined by a question asked by agent a to agent b, and answer given by agent b to agent a. Whereas in simulations on a forum, each agent posts a question on a blackboard, and every other agent will see it. Thus, questions are not addressed to one specific agent, but rather to the community as a whole. - Each agent can only ask one question per time-step. - An agent asks a question about a knowledge it doesn’t have. - An agent answers a question if it has the specific knowledge asked for and if it is available; otherwise, it will ignore the question. 2.4. Learning process Each time an agent gets an answer to a question; it raises its knowledge of that particular subject to 1, and won’t ask questions about this subject anymore. Following example 1, an agent increases her knowledge of subject 2, as shown in Figure 3. Knowledge 1 2 … 99 100 Knowledge endowments 1 1 … 1 1 Figure 3 An agent learns and acquires knowledge about subject 2 2.5. Choosing a knowledge-provider In simulations where agents interact on a forum, agents do not choose a knowledge-provider as they simply post their questions on the forum and wait for an answer. Things are different in simulations with face-to-face interactions. Here, knowledge-seekers must select a knowledge-provider, and this choice depends on the type of simulations running: - In simulations based on a predetermined hierarchy: agents choose the most competent agent in the population. 4 - In simulations based on past experiences: to choose a knowledge-provider, an agent will base its decision on the performance of each knowledge provider towards it. The t+1 performance of agent j towards agent i at time-step t+1 ( perf ij ) is calculated as follows: t +1 t t +1 perf ij = α perf ij + (1 − α ) nbAnswers ij (1) 2 t+1 Where: α = 0.2 and: nbAnswers ij : Number of answers given by agent j to agent i at time-step t+1.