Chapter 1
Model facilitated learning
Marcelo Milrad, Michael Spector and Pål Davidsen
Introduction
Technology changes what we do and what we can do. People change on account of technology. Technology in support of learning and instruction is no different. Instructional technology changes what teachers and learners do and can do. This is especially true when the Internet and distributed technologies are taken into consideration. Learning research has also evolved and increased our understanding of how people learn different things in different situations. There has been a trend to apply emerging instructional technologies to support learning and instruction in ever more challenging and complex domains (Spector and Anderson, 2000). Such a trend is quite natural. Once it is understood how to use technology to support mastery of simple skills, it makes good sense to explore more advanced uses of technology. We support this trend and believe, along with many others, that technology can be effectively used in distributed learning environments to support learning in and about complex systems, which is the focus of the discussion in this chapter (Spector and Anderson, 2000).
Modelling and simulation tools are gaining importance as a means to explore, comprehend, learn and communicate complex ideas, especially in distributed learning and work environments (Maier and Größler, 2000). Students are building and using simulations in both guided discovery and expository learning environments (Alessi, 2000). Of particular interest is whether and when one learns by building simulations or by interacting with existing simulations (Spector, 2000). To explore this interest, we provide a framework for the integration of modelling and simulations deployable in collaborative tele-learning environments. We focus on a particular modelling and simulation approach called ‘system dynamics’ (Forrester, 1985).
The system dynamics community has focused primarily on learning by creating simulation models, although some researchers are becoming more sophisticated in recognizing a variety of different learning situations and requirements (Alessi, 2000; Gibbons, 2001; Spector, 2000). The system dynamics community believes in the value of using system dynamics to improve understanding of complex, dynamic systems (Davidsen, 1996; Forrester, 1985; Sterman, 1994). This general commitment allows for both learning with models and learning by modelling.
The ability to model complex systems requires being able to define a model and use it to understand some complex phenomena—to make connections between and among parts and to analyse the model’s ability to represent relevant aspects of the perceived world (Jackson et al, 2000). In the construction of models using systems dynamics tools, learners engage in cognitive and social processes that appear to promote understanding. However, it seems unreasonable to conclude that deep understanding in a complex domain always requires one to become an expert system dynamics modeller (Spector, 2000).
Considerable research has documented a variety of difficulties with learning concepts relevant to understanding complex systems in a variety of disciplines (Dörner, 1996; Kozma, 2000). For example, many people have difficulty with the following:
- understanding the effects of non-linear relationships over time;
- keeping the entire system in mind when trying to resolve an apparently localized problem;
- appreciating the full range of control and influence possible within a complex system; and
- generalizing lessons learnt from a particular problem context to a different problem situation.
How can learners acquire and maintain deep understanding about difficult-to-understand subject matter? How can modelling and simulation in complex domains be best used to facilitate learning? Understanding complex system behaviour involves the ability to provide causal and structural explanations as well as the ability to anticipate and explain changes in underlying causes and structures. This kind of understanding is not acquired easily nor is it likely to be acquired from observations of either real or simulated behaviour (Dörner, 1996). However, an appropriate methodology linked with collaborative and distributed technologies can significantly enhance such learning.
Our motivating concern is to help learners manage complexity in ways that contribute to improved learning and deep understanding. To achieve this goal, learning theory (socio-constructivism), methodology (system dynamics) and technology (collaborative tele-learning) should be suitably integrated (Spector and Anderson, 2000). We call this integration Model Facilitated Learning (MFL) (Spector and Davidsen, 2000).
A theoretically grounded framework
Our understanding of the developmental, cognitive, and social dimensions of learning improved in the last half of the 20th century. Research inspired by Vygotsky and others suggests that recognizing the need for learners to engage peers in dialogue concerning challenging new concepts and to work in collaboration with colleagues on difficult tasks produces desirable and persisting improvements in understanding (Jonassen et al, 2000; Rouwette et al, 2000; Spector et al, 1999; Wells, 1999). Distributed technologies (eg, networked learning communities) are well suited to support such collaboration.
Learning in complex and ill-structured domains places significant cognitive demands on learners, as appropriately recognized by the medical community. Feltovich et al (1996) note that one of the difficulties involves the misunderstanding of situations in which there are multiple, co-occurring processes or dimensions of interaction. In these kinds of situations, learners often confine their understanding to one or a small number of the operative dimensions rather than the many that are pertinent (see also Dörner, 1996). Technology that depicts dynamic interactions can be of particular help in this area. The learning perspective we find most appropriate is based on notions derived from situated and problem-based learning (Lave and Wenger, 1990), especially as informed by cognitive flexibility theory (Spiro et al, 1988). Instructional design methods and principles consistent with this learning perspective can be derived from elaboration theory (Reigeluth and Stein, 1983) and from cognitive apprenticeship (Collins et al, 1989). MFL is derived from these learning and instructional theories. That these theories are reasonably well established but not embraced by the system dynamics learning community is somewhat disturbing.
Situated learning (Lave and Wenger, 1990) is a general theory of knowledge acquisition based on the notion that learning (stable, persisting changes in knowledge, skills and behaviour) occurs in the context of activities that typically involve a problem, others, and a culture. This perspective is based on observations indicating that learners gradually move from newcomer status (operating on the periphery of a community of practitioners) to more advanced status (operating at the centre of the community of practitioners). As learners become more advanced in a domain, they typically become more engaged with the central and challenging problems that occupy a particular group of practitioners.
Cognitive Flexibility Theory (CFT) (Spiro et al, 1988) shares with situated and problem-based learning the view that learning is context dependent, with the associated need to provide multiple representations and varied examples so as to promote generalization and abstraction processes. Feltovich et al (1996) argue that CFT and related approaches can help learners develop skills for thinking and learning about complex subject matter. Multiple representations naturally emerge in collaborative and group work. When learners are distributed in various settings and circumstances, it is essential to support multiple representations; CFT suggests this is important even for individual learners. Moreover, learning should be supported with a variety of problems and cases, which is especially important in distributed learning environments. However, people seem to prefer single and simple models. These restricted perspectives may be detrimental to learning (Feltovich et al, 1996; Kozma, 2000). As knowledge is used and represented in many ways it becomes more meaningful and more powerful. Towards this end, CFT advocates multiple types of models, multiple representations, alternative conceptualizations, varying levels of representational granularity, and so on. Additionally, CFT places particular emphasis on the importance of learner-constructed and learner-modifiable representations.
MFL, as a realization of CFT through system dynamics and distributed technology, provides learners with the opportunity and challenge to become model builders, to exchange and discuss models with peers, and to experiment with models to test hypotheses and explore alternative explanations for various phenomena. We believe that such modelling activities are often appropriate activity for advanced learners, but model building is not always required in order to understand some aspects of a complex and dynamic system. Moreover, we believe that other activities, including interacting with existing models and simulations, are often appropriate precursors to model building activities. MFL advocates a sequence of learning activities that begins with some kind of concrete operation, manipulating tangible objects in order to solve specific problems (Milrad et al, 2000). As these operations are mastered, learners can then progress to more abstract representations and solve increasingly complex problems. A set of principles to guide a MFL elaboration sequence is:
- Situate the learning experience. Provide an opening scenario or a concrete case to familiarize learners with the complexity of the domain and with typical problems encountered in that domain.
- Present problems and challenges of increasing complexity related to the opening scenario. For instance, suppose the initial situation involves managing a production plant. A problem sequence might be to determine existing inventory, predict future orders and provide a plan for maintaining a stable inventory. As participants gain expertise, other aspects of the enterprise can be brought into consideration, such as the effect of overtime on workers as they try to keep up with orders or the effect of backlogged orders on future orders and so on.
- Involve learners in responding to a set of increasingly complex inquiries about the problem situation. For example, suppose that the sales force has predicted a seasonal increase in orders. A number of inquiries about the effect on existing inventories can be constructed and used to stimulate individual and small group discussion and experimentation in order to provide answers about predicted system behaviour.
- Challenge learners to develop decision-making rules and guidelines for a variety of anticipated situations. In this case, a great deal of experimentation with models and simulations is appropriate. As the challenges increase in complexity, it is at this stage of learning that it is appropriate to provide opportunities for learners to modify models or create new models.
To summarize, we accept the notion that complex concepts are best learnt in context—a problem setting in which the learner must apply and use the relevant concepts and knowledge to solve meaningful problems. Such learning should improve both retention (by providing a relevant context) and transfer (by providing multiple representations). The principle of graduated complexity (Spector and Davidsen, 2000) is used to guide the design of learning sequences. In addition, the notion of socially-situated learning experiences threads throughout such a sequence. Such learning principles suggest that the coupling of system dynamics with collaborative and distributed technologies has strong potential. Next we examine the role of models in learning.
The potentials of models in learning
In this section we illustrate how models can be used to represent complex subject matter. It is worth emphasizing that the steps in a graduated complexity model should not be considered fixed or rigid. The model we advocate recognizes individual and group differences and supports the notion of iterative development of learning, understanding and expertise.
Learning with models and learning by modelling are discussed separately here, but in a learning or problem-solving environment it is conceivable that both might be involved (albeit for different purposes and in different ways). In MFL, there are three stages of learner development with associated instructional approaches (Spector and Davidsen, 2000):
- problem-orientation (problem confronting and problem solving), in which learners are presented with typical problem situations and asked to solve relatively simple problems;
- inquiry-exploration (hypothesis formulation and experimentation), in which learners are challenged to explore a complex domain and asked to identify and elaborate causal relationships and dominant underlying structures; and
- policy-development (decision-making rule and global system elaboration), in which learners are immersed in the full complex system and asked to develop rules and heuristics to guide decision making in order to create stability or avoid undesirable situations.
The stages and principles of MFL correspond with major components of van Merriënboer’s (1997) 4C/ID model and Dreyfus and Dreyfus (1986) (see Table 1.1). Interestingly, the methods in the 4C/ID model are primarily focused on an analysis of the subject domain whereas Dreyfus and Dreyfus focus primarily on the learner. Naturally, both are important considerations for an instructional designer.
The principle of graduated complexity in MFL suggests a sequence of learner challenges:
- Challenge learners to characterize the standard behaviour of the complex system (how the system behaves over time with an indication of how components are interrelated).
- Challenge learners to identify key variables and points of leverage with respect to a desired outcome.
- Challenge learners to identify and explain the causes for observed system behaviour, especially in terms of key influence factors that might be subject to control and manipulation.
- Challenge learners to reflect on the dynamic aspects of the system in the context of decision and policy guides to achieve desired outcomes.
- Challenge learners to encapsulate learning in terms of a rationale for system structure, decision-making guidelines, and an elaborated strategy for policy formulation.
- Challenge learners to diversify and generalize to new problem situations. (To assess deep understanding one might ask learners to create a dynamic model relevant to an apparently new problem situation that is likely to have an underlying structure similar to a problem situation already resolved by the learner.)
Table 1.1 MFL, learning development and related models
Throughout the various stages learners are challenged to start meaningful discussions with peers about problems, models and proposed solutions, all of which are well supported by available Web-based technologies. Such discussions help learners reflect about the subject matter and encourage peer-peer learning and group collaboration.
Next we shall provide examples. We follow Alessi (2000) in distinguishing learning with models from learning by modelling. We believe that learning with models is generally well suited for the earlier learning stages that often involve simple procedural tasks and simpler conceptual foundations (similar to algorithmic-based learning in 4C/ID), whereas learning by modelling is generally better suited to more advanced stages of development targeted at causal understanding and mastery of complex procedures not amenable to formulaic or standard solution (similar to heuristic-based learning in 4C/ID).
MFL emphasizes socially-situated learning processes. A suggestion of how to support collaboration with modelling tools in a discovery setting has been made by van Joolingen (2000). In the construction of models using systems dynamics tools, learners engage in cognitive and social processes that promote collaborative knowledge building. Rouwette et al (2000) argue that a collaborative approach to model and policy desi...