The provision of lifelong learning facilities is considered to be a major new direction for higher and distance teaching educational institutes catering for the demands of industry and society. ICT networks will in future support seamless, ubiquitous access to lifelong learning facilities at home, at work, in schools and universities. This implies the development of new ways of organizing learning delivery that that goes beyond course and programme- centric models. It envisions a learner-centred, learner-controlled model of distributed lifelong learning. We present a conceptual model for the support of lifelong learning which is based on notions from self-organization theory, learning communities, agent technologies and learning technology specifications such as IMS Learning Design. An exploratory implementation has been developed and used in practice. We reflect on the findings and future directions. p. 1

Lifelong learning is the concept that refers to the activities a person performs throughout their life to improve their knowledge, skills and competence in a particular field, given some personal, societal or employment related motives (Aspin & Chapman, 2000; Field, 2001; Griffin, 1999). p. 2

LNs can be studied from at least three perspectives: pedagogical, organizational and technical (Koper, 2004). Pedagogical Aspects An LN is a distributed set of people who interact to create and share units of learning in developing their competence in a particular discipline. It is a ‘two-mode network’ with two types of nodes: the members of the LN and the units of learning in the LN (Wasserman & Faust, 1994; Degenne & Forsé, 1999). The members define the learning community (LC) within the LN. The units of learning (UOLs) define the set of learning activities offered in the LN, for example courses, assessments, workshops or seminars. The core questions in this section are: How can effective LCs be developed? And how can effective UOLs be developed and used in LNs? p. 3

Learning Communities Shaffer and Anundsen (1993) define ‘community’ as a dynamic whole that emerges when a group of people share common practice; are interdependent; make decisions jointly; identify with something larger than the sum of their individual relationships; and make long- term commitments to well-being (their own, one another’s and the group’s). Communities tend to be self-governed, self-organized and decentralized. Common goals and values and communal relationships are important moderators in forming communities. Communities have their own identity, which can change and evolve. Wilson and Ryder (1998) characterize ‘learning communities’ (LCs) as follows: they have distributed control; there is commitment to the generation and sharing of new knowledge; learning activities are flexible and negotiated; community members are autonomous; there is a high level of dialogue, interaction and collaboration; and there is a shared goal, problem or project creating a common focus and incentives to work together. Within the context of lifelong learning it is necessary to have an enduring membership of the community. Competence in a field evolves over a lifetime. An important requirement p. 3

for lifelong learning is that the learning results are stored in a portable, standard way, for example in a portfolio. These learning results can be used to identify the LNs; position the person in the network; and provide a classification of the expertise of the person in the field. As seen from a lifelong learning perspective, a teacher is not a separate entity, but is a role that any lifelong learner can take, depending on their expertise, and relative to the expertise of whoever requires support. Anyone can start in a community as a novice and evolve into an expert. During his/her lifetime the person stays a member and is responsible for sharing knowledge and experience as required. The knowledge and support services in the community and the members’ knowledge also evolve. In a permanent community, the community itself gets a structure and culture independent of the participants. We adopted this idea of lasting, evolving learning communities as a key principle for the design of LNs. Lifelong learners must have easy, ubiquitous access to an LN, which should not be location or technology dependent. It should be accessible from anywhere by standard means of communication. In order to sustain it, it must support, among other things, interoperability standards that have been adopted, defined and agreed upon within the community. p. 4

Units of Learning UOLs are developed and used in LNs and can serve various functions in them depending on the design, for example the introduction to a knowledge domain; acquisition of a skill; or assessment of acquired knowledge. Members of an LN should be able to select the UOLs they need in order to attain certain expertise or competencies, given their pre-knowledge. When a UOL is selected, the person must be able to study it and provide feedback about it. When the UOL is used in practice, additional run time data can be added to the design, for example user and usage information, mail messages and forum contributions. A UOL typically contains a learning design and learning resources. The learning design specifies the workflow in the teaching-learning process (Koper, 2001; Koper & Manderveld, in press; Koper & Van Es, 2004). At the abstract level, the learning design describes the following process: A person gets a role (e.g. lifelong learner) in a learning context. This role entails a set of learning activities for attaining some specified learning objectives. A learning design method, based on a pedagogical approach, determines which roles get which type of activity at a given time. The learning activities are performed in a learning environment provided with resources and communication facilities. The outcomes of the learning activities are also resources that are added to this environment. Properties are p. 4

defined to keep track of learners’ progress. In addition to learning activities, a person can also get a role to perform support activities to help others learn. This abstract learning design model is implemented in the IMS Learning Design specification (LD, 2003) to create interoperable learning designs. The same learning design can be used with different resources and vice versa. The process of building UOLs from learning designs and resources is called ‘aggregation’. Conversely, the process of breaking down the structure of a UOL into learning design and resources is called ‘disaggregation’. These processes support the reuse of learning designs and underlying learning resources. In order to develop effective UOLs, the learning design of the UOL should be based on an appropriate ‘pedagogical model’. A pedagogical model prescribes an effective teaching- learning process for a class of learners to achieve a class of learning objectives in a class of situations. A learning design is an instance of a pedagogical model. It is a concrete application of a pedagogical model for a specific learning objective, a target group and a specific situation. p. 5

Another reason for introducing self-organization in LNs is to increase the efficiency of the learning support structure in LNs. Active learners in an LN produce work such as written contributions to discussions and research reports. These have to be read, reacted to or reviewed. In a traditional setting, there is a danger that these tasks will be assigned solely to the teacher, whose workload will then increase considerably. Our assumption is that the application of self-organization theory can be a foundation for the establishment of efficient systems with a minimum of planning and control, while maintaining maximum flexibility to adapt to learners’ needs. This will reduce overhead costs for maintenance, planning, control and quality. This assumption is based on research into self- organization theory (Varela, Thompson & Rosch, 1991; Maturana & Varela, 1992), which is grounded in complexity theory (Waldrop, 1992; Kauffman, 1995) and studies the characteristics of the social organization of communities that ‘emerge’ from the interactions of lower level actors. It deals with the way macro-phenomena occur as emergent behaviours from the activities of the subsystems at the micro-level (Prietula, Carley, & Gasser, 1998, p.14). The social organization that emerges (e.g. trust, grouping, role specialization, action coordination, distribution of tasks and resources, conflict resolution, quality norms and interaction standards) in its turn imposes behavioural constraints on the actors and provides for social objectives (Ferber, 1999; see Figure 1). p. 6

Using this perspective, the organization of lifelong learning can be realized by installing technical facilities that enable distributed interactions among participants directed at a common purpose (e.g. competence development in a disciplinary field), governed by policies that stimulate participants to learn, share knowledge and support each other. The management and application of policies in an LN is termed ‘sociability’ (Preece 2000, p. 26-17). Sociability governs social interaction in a community. It cannot be controlled directly, but can be supported by carefully communicating the purpose and policies of the community. Preece (p. 95-96) identifies several policies in a community: joining or leaving requirements; by-laws; codes of practice for communication; rules for moderation; issues of privacy and trust; practices for distinguishing professionally contributed information; rules for copyright; and democracy and free speech in the community. We identify the policies in LN in terms of: objectives and values; terms of use; membership/role policies; standards and quality policies; and reward policies. p. 7

An important factor in establishing self-organization is the creation of first-order and second-order feedback mechanisms. First-order means that people in the community know what their counterparts are doing or have done regarding the UOLs in the network. This provides information for navigation and behavioural models within the community. Second- order refers to feedback about the emergent properties in the system: what is the performance of the community and how is it organized (Gilbert, 1995)? For instance, there is no centralized quality control in the LN. It is expected that the network will uphold a range of quality levels, but that the feedback mechanisms (e.g. reviews and ratings) will ensure that on average satisfactory quality is maintained. Thus, factors such as development costs, frequency of use, incentives, price and satisfaction may be dynamically balanced. p. 7

role-specialization is functional. Individuals can change roles when the demand for a certain activity increases (Bonabeau, Dorigo & Theraulaz, 1999). The activities of persons in an LN are influenced by the reward system established in it (e.g. personal need, reputation, money). A theory about reward is elaborated in social exchange theory (Thibaut & Kelly, 1959; Constant et al., 1994). The reward system is typically implemented in the policies of an LN. We want to stress that self-organization implies organization. It is not a synonym for chaos, anarchy or disorganization. The structures that result from self-organization can, in principle, be the same as those proposed by central agencies, except that democratic principles determine the hierarchy and organization. p. 8

. Several open interoperability specifications have been developed, most of them by IMS (imsglobal.org), IEEE (ltsc.ieee.org) and AICC (aicc.org). p. 8

A critical specification for LNs is IMS Learning Design (LD, 2003). LD implements the abstract learning design model discussed above. It enables the representation of the learning and teaching processes in a UOL to be interoperable and machine interpretable. It provides a framework for including learning activities, support activities, assessment and learning or knowledge resources. LD can express the pedagogical approach taken in the UOL, and supports personalization of learning routes and reusability (Koper & Van Es, 2004). p. 9

An LN is usable when it supports rapid learning, high skill retention, low error rates and high productivity. It is consistent, controllable and predictable, making it pleasant and effective to use (Preece 2000, p. 26-27). p. 9

Like any network, an LN can be represented as a graph with nodes. However, an LN is a two-mode network, with the nodes being LN members and UOLs. In the following sections we will aggregate the two modes into a single node, called an Activity Node (AN). An AN contains all the runs of all the versions of a UOL, including information about the members who are (or have been) active in it and the information the members have produced about it (e.g. feedback, completion data). Moreover, it contains a set of rules that govern its lifetime, specifically its ‘fading out’ and ‘staying alive’ behaviour. There are subtle differences between a UOL, a UOL run, and an AN. A UOL is the learning facility that is abstractly defined for any set of learners at any time. A UOL run is its instantiation for a specific set of learners in a certain time frame (e.g. a class, the actual run of a workshop). An AN is the set of all possible runs for different versions of the UOL. p. 10

Actors travel in an LN from AN to AN, leaving their tracks in each. The sequence of ANs completed by learners is a ‘learning track’ (see Figure 3a). The information in any two ANs must be designed so that it can be analysed. (How many learners travelled from the first to the second AN or vice versa? What are the usage profiles of the learners and the perceived quality of the path?) Learning ‘routes’ are paths that are planned (Figure 3b). In traditional educational settings, teachers or instructional designers are responsible for this. A different approach to creating routes is to share learning tracks and experience of them between participants in an LN. This can be a single track or an analysis of the aggregated, collective tracks from a set of participants to determine the most successful routes. This data is expected to help actors navigate in the LN, specifically when additional maps and facilities are created, for example the possibility of matching a user’s current usage profile with the average profiles of users who followed a particular track. In this approach, routes are based on real practice rather than theoretical instructional design principles. It opens the possibility of adding this inductive route design to the current theoretical, deductive design approaches (Koper & Sloep, 2003). p. 14

We created a prototype based on the principles discussed above. In order to save costly programming time, we selected a platform we could customize to fulfil the requirements as far possible. We selected the Groove (Groove.net), a fairly easy to customize, peer-to-peer collaborative environment. p. 15

An analysis of the implications of theories such as the social exchange theory (Thibaut & Kelly, 1959; Constant et al., 1994) for LNs is required. p. 22

Aspin, D. N., & Chapman, J. D. (2000). Lifelong learning: concepts and conceptions. International Journal of Lifelong Education, 19(1), 2-19. p. 24

Cross-Durrant, A. (1987). Basil Yeaxlee and the origins of Lifelong education. In P. Jarvis (Ed.). Twentieth century thinkers in adult education (pp. 327). New York: Crook Helm. p. 24

Field, J. (2001). Lifelong education. International Journal of Lifelong Education, 20(1/2), 315. p. 24

Griffin, C. (1999). Lifelong learning and social democracy. International Journal of Lifelong Education, 18(5), 329-342. p. 25

Koper, E. J. R., & Sloep, P. (2003). Learning Networks: connecting people, organizations, autonomous agents and learning resources to establish the emergence of effective lifelong learning. Heerlen: Open University of the Netherlands.Retrieved www.learningnetworks.org Koper, E.J.R. (2001). Modelling Units of Study from a Pedagogical Perspective: the pedagogical metamodel behind EML. Heerlen: Open Universiteit Nederland. Retrieved October 3, 2003, from http://eml.ou.nl/introduction/docs/ped- metamodel.pdf Koper, R., & Van Es, R. (2004). Modeling units of learning from a pedagogical perspective. London: RoutledgeFalmer. p. 25

Merrill, M.D. (2003). First Principles of Instruction. Retrieved October 3, 2003, from http://www1.moe.edu.sg/itopia/download/abstracts/Applying%20First Principles of Instruction to Technology-Based Education.pdf p. 26

Reigeluth, C.M. (1999). What is Instructional-Design Theory and How is it Changing? In C.M. Reigeluth, Instructional-Design Theories and Models (pp. 5-30). London: Lawrence Erlbaum. Retallick, J., Cocklin, B., & Coombe, K. (1999). Learning Communities in Education. London: Routledge. p. 26

Tuijnman, A. v. d. K. M. (1992). Learning across the lifespan, theories, research, policies. Oxford: Pergamon Press. p. 27

Wilson, B., & Ryder, M. (1998). Distributed Learning Communities: an alternative to designed instructional systems. submitted to Educational Technology Research and Development, from http://carbon.cudenver.edu/~bwilson/dlc.html p. 27