Our ITForum discussion occurs between one on the topic of ethical research in the information age (Lynne Schrum) and one on the topic of the forgotten art of interactivity (Rod Sims). We thought it might be nice if we could provide a bridge between those two subjects in addition to saying something about our own subject. We could not decide how to provide any thought provoking links short of performing some unethical electronic interaction (hereafter to be referred to as UEIs and clearly distinct from UFOs, unidentified flying objects), but it may be worth mentioning that there is a recent book edited by William Wallace (1994) on Ethics in Modeling that might be a piece of the bridge as it explores the relationships between computer models and decision making, how values become incorporated in models, ethical responsibilities of model builders, and implications for the future. If one substitutes "technology-based learning environments" for "models," then those collected essays can be viewed as relevant to this Autumn's sequence of three discussions on ITForum.
We have two primary goals in this discussion: (1) provide an appreciation for the system dynamics perspective of the relationships between structure and behavior in complex systems and (2) provide a basis for the use of system dynamics as a basic building block in the construction of effective, computer-based learning environments. In order to accomplish these goals, we shall first provide a brief overview of system dynamics, as most ITForum participants may be unfamiliar with this discipline. Then we will be able to indicate some possibilities for the use of system dynamics in the creation of effective courseware. We hope this will provide a reasonable departure point for a fruitful discussion.
We shall make a number of assumptions with regard to the nature of reality, learning, learning environments, learning effectiveness, and instruction. We do not wish to defend these assumptions in the course of this discussion, although we do realize that many or all of these assumptions may be questioned. Our aim is to provoke a discussion about what we believe to be a very powerful way to design learning environments.
For the purposes of this discussion, learning refers to intentional learning and involves persisting, stable, and observable changes in an individual. Learning environments can be described in terms of the following attributes: (1) a variety of actors/agents (e.g., learners, instructors, computer-mediated delivery mechanisms, etc.); (2) appropriate and reasonably well-defined roles for those actors (not necessarily fixed or pre-determined); (3) a setting in which learning occurs (this may involve multiple settings as in distance learning situations or variable settings as in laboratory-based courses); (4) a knowledge base pertinent to a subject matter area, possibly with links to other knowledge bases (e.g., a collection of learning materials or an instructional database); (5) learning goals, objectives, and desired outcomes (these may be negotiable); and (6) a set of possible relationships and interactions among the actors, roles, knowledge base, and settings which facilitate or contribute to accomplishment of goals and objectives. Learning effectiveness refers to the efficacy of an environment in the attainment of those goals. Instructional design (ID) refers to a structuring of the learning environment for the purpose of facilitating learning or improving learning effectiveness. We shall further assume that there are external realities, meaningfully described in terms of causal relationships, which often form the subject matter for many learning situations.
First, we shall provide an overview of a generic system dynamics (SD) perspective. SD suggests that a complex system can be described in terms of stocks (containers) of things or substances, the quantity of which may change over time. Such quantities are called stock levels. The concept "thing" is very broad and covers, in this context, items such as commodities in an inventory, money in an account, animals in a population, molecules in a container, persons infected by a virus, tasks to be accomplished, levels of understanding, and even mental states. "Substance" refers to a single item of continuous nature such as a population, a liquid or a gas often considered to be an aggregate of things in which single items are considered indistinguishable (for the purpose of the model). In SD stocks may be visually represented as containers (usually rectangular boxes).
Stocks can be distinguished in two different ways. On the one hand, we distinguish stocks of different kinds of things. In this case, stocks represent a categorization of things (e.g., commodities, money, animals, molecules, or persons). On the other hand, we distinguish stocks of the same kind of things in different states, viz., a categorization of states that characterize things. For instance, commodities exist in different states along a production and delivery chain, or understanding exists in different states in stages of maturity or expertise. Each of these stages may be represented by a stock and the amount of commodities in that stage by the level of that stock.
System dynamicists often distinguish between real, perceived, and desired levels of stocks. Frequently, the emphasis in a dynamic system is placed on the perceived and desired levels as it is often people's perceptions and goals that give rise to dynamic behavior. These distinctions make it possible to create representations of and speak meaningfully about mental models. We shall suggest that a particularly powerful aspect of SD is that it provides learners with tools and techniques that allow them to represent and explore the implications of their own mental models.
The levels of all stocks in a system at a particular point in time constitute the state of the system at that point in time. As already suggested, we may refer not only to material but also to mental states by stock levels. We may represent a level of aggression, a level of frustration, and so on. In such cases, it is extremely important to agree upon an operational definition underlying the term, although the definition may involve some fuzziness or uncertainty. The state is but one of two fundamental aspects of dynamic systems--the other one being the change of state. Since the state is represented by stock levels, the change of state may be conveniently represented by flows, and the rates of change by flow rates. Flows accumulate in stocks, and the rate at which a stock level changes is determined by the net rate of the flows accumulating in the stock. If a stock level is characterized in terms of units, the associated rates of flows are all characterized in terms of units per time.
For instance, a population (stock) is increased by an inflow of individuals being born and (simultaneously) decreased by an outflow of individuals dying per year. We call these the birth rate and the death rate, respectively. It is often possible to specify rates which at least partially determine how fast the things or substances in a stock increase or decrease in quantity, such as order and sales rates, production and delivery rates, birth and mortality rates, infection rates, and so on. These rates can be thought of as the magnitudes of flows into or out of a stock and are usually depicted as arrows into or out of a rectangular box with a control valve to represent the notion of a variable flow rate, as in a faucet controlling the flow of water into a bath tub or water basin--the faucet represents a variable flow rate, in the familiar sense that opening the valve more allows more water to flow into the basin (i.e., creates a higher flow rate). The tub or basin can be thought of as the stock in this case, and the drain might be thought of as a flow out of the stock. The drain also has an associated flow whose rate depends in large part on the amount of water in the basin and the size of the drain pipe.
In addition to stocks and flows, in understanding complex systems we also need to take into account a variety of variables which ultimately influence the flow rates associated with the stocks in the system. For example, resistance to drought may be thought of as a variable which may influence the mortality rate of a particular species in an ecosystem or perceived wealth may be a factor which may be thought of as influencing the investment rate in an economic model. In SD, variables represent causal relationships and are depicted as circles with links or curved arrows to indicate the lines of influence. The totality of all the causal relationships among stocks, flow rates, and variables in a system is called the structure of the system.
At this point we want to make two important comments. First, we have only introduced a small portion of the basics of system dynamics, but what we have discussed has a very strong visual component (rectangles, circles, arrows, etc.). We have not discussed the notions of delay, nonlinearity, and uncertainty which are central notions in SD--indeed, it is these notions which distinguish SD from other modeling and simulation techniques. Second, we are unable to provide these important visual components in a strictly text-based presentation on ITForum. Consequently, we would like to direct interested readers to a web page with an expanded version of this paper containing the appropriate diagrams and references (URL: http://www.uib.no/svf/ifi/emp/mike.html, click on ITForum).
In addition to constructing stock and flow diagrams to represent complex systems, system dynamicists often construct causal loop diagrams to make explicit causal relationships among the various components of a system. Of particular interest in such causal loop diagrams are positive and negative feedback loops. An example of a positive feedback loop might consist of a savings account (a stock of money), an earnings rate on our savings (the rate at which that stock of money increases, partly determined by what might be a constant interest rate and partly by the balance in the savings account), and a savings factor (which represents how much of our total earnings we can redirect back into savings). If we make a number of simplifying assumptions (such as zero inflation, constant other earnings, constant expenditures, etc.), we might say that as the earnings rate on savings increases, our savings increases, which has a positive influence on total earnings, which has a positive influence on how much can be diverted back to savings, which means that the earnings on savings will continue to increase.
We might all like to believe that our savings could enjoy such sustained growth, but we know that there are moderating or balancing influences which make it very unlikely that we shall enjoy such periods of expansion in our savings. These moderating influences might be represented as negative feedback loops, causal loops that have one (or an odd number of) negative link(s). For example, as our savings grow, we might spend more on consumable goods (making the assumption about constant expenditures no longer applicable), or we make more risky investments some of which will result in a loss of income (making the assumption about constant non-savings earnings invalid), or we are taxed at a higher rate (at least in some countries), and so on. In short, these and many other factors have a very real and balancing effect on our earnings.
SD makes it possible to make all of these relationships explicit and then to study their interactions over time. The change in the levels of the stocks over time constitutes the behavior of a system. With complex systems, it is often less than obvious how the structure of the system (the various stocks, flows, and variables) determines the behavior of the system (changes in the levels over time). SD can be thought of as a systematic approach to understanding the complexities of dynamic behavior (e.g., in systems with a dominant positive or growth loop which, as time passes, becomes less influential, while a negative or balancing loop gradually becomes more influential).
SD also makes it possible to explicitly represent factors outside the system (exogenous variables) which influence system behaviorÑsome of these factors represent things which might have been empirically established while others are established by consensus or assumption. Since one can never model everything, it is important to make simplifying assumptions and to make those assumptions explicit. We believe that the activities of making assumptions explicit, exploring the dynamic implications of assumptions, and examining alternative assumptions are critically important activities in many learning environments.
We began with a monetary example because we thought most ITForum readers would be familiar with the concepts of earnings, savings, and interest and, as a consequence, would be able to follow the discussion (cf. Reigeluth's elaboration principle of using epitomizing examples and Gagne's third instructional event of stimulating recall of prior knowledge). However, we expect that most readers are more interested in improving the quality of instruction and learning than in bank accounts. In order to push our discussion of system dynamics into the domain of learning, we'll consider a simplified ecosystem as the subject of a learning environment that we want to construct for a group of learners. We might start with a rather vague learning goal, such as understanding how the population of a particular species is effected by a predator species in that environment. In other words, our task is to design a learning environment to facilitate learning about, for example, how a prey population fluctuates in relationship to a predator population. To make this exercise more personally meaningful and motivating, we leave it to readers to supply their favorite prey/predator pairs (e.g., rabbits and cougars, rats and cats, little fish and big fish, etc.). Now, imagine an area where these two species exist in reasonable abundance. It will be especially helpful if this area is near a human population which will develop a preference for more or fewer of one of the species (e.g., rabbits may become a bothersome pest to farmers' crops, rats may begin to spread diseases, there are too few big fish to make for worthwhile fishing excursions, etc.).
In designing courseware about this subject we will have to address the following kinds of issues: (1) Learners: Who are the learners? What are they like? What do they like? What do they already know? etc.; (2) Subject Matter: What is the subject matter? What kinds of things are to be learned? What links exist to related subjects? etc.; (3) Historical Value: Why do people study this subject? Who has studied this subject? What kinds of things have they said? Why? etc.; (4) Learning Support: What alternative learning settings exist? How can learning about this subject for the indicated purposes best be supported? What roles can be supported for learners, peer groups, tutors, teachers, computers? etc.; and, (5) Learning Effectiveness: How can progress in this learning situation be evaluated? How can students learn to be reasonably accurate in self-assessment?
At this point we can make explicit some of the ways in which we believe that SD can be used to support the creation of effective learning environments. Since space is limited and our purpose is to stimulate a discussion, we shall take a rather abbreviated approach to indicating how SD might be linked to a set of principles for designing effective learning environments. Let's accept for the time being a small set of more or less well-established ID principles (e.g., initiate the learning experience with an epitomizing and motivating example, provide the means to elaborate each aspect of the epitome, provide a variety of learning support mechanisms such as visual representations of key relationships, provide meaningful feedback with regard to learner performance, encourage and support self-assessment, etc.). The problem then arises how best to implement such principles. Our discussion of SD should suggest a strong possibility of supporting such ID principles with interactive models and simulations as the basic building blocks of interactive courseware. We believe that these SD building blocks are especially promising as they actively engage learners in the processes of making explicit their own mental models of complex systems, of identifying their own assumptions, of exploring implications of those assumptions, of examining alternative assumptions, etc. Learners can create and modify SD models, and these models will reflect changes and growth in their knowledge about a subject.
Suppose, for example, that in our imaginary learning environment we want to initiate learning about predator/prey dynamics with an epitomizing and motivating example. We may have a ready made SD model concerning some particular predator/prey situation that developed counter-intuitively. This model could be presented to learners, identifying key components of the model (not all of the model components and certainly not all of its complexity). Learners might then be asked to predict the behavior of this model over time. Then, we run the simulation based on the underlying model to show how the model actually behaves over time. If the outcomes of the simulation are significantly different from expected outcomes (the learner has already been asked to predict the behavior), and especially if the particular example has some recency or currency in terms of general topics of concern, then we will have implemented an important ID principle using SD as a methodology. The model can then be left in a library of relevant materials that may be revisited later or which learners may decide to explore in further detail at their convenience.
Now let's tackle the heart of our planned learning experience. Our nominal problem is to understand how the populations of two animal populations are related with the purpose being to explain and perhaps control the spread of some disease ostensibly spread by one of the animal populations. Students will be asked to construct an initial model of the situation, and the courseware will provide two kinds of support: tools for building the model (such as those available in PowerSim or STELLA), and hints for using the tools for the specific purpose at hand (such hints can be provided under learner or system control and might even form the basis for an adaptive interface, automatically adjusting to perceived level of learner understanding).
This process of building an initial model can be collaborative and staged (consistent with ID principles concerning chunking and graduated complexity) so as to deal with the following questions: (1) What factors contribute to the level or state of each animal population at any given time (e.g., birth and mortality rates)? (2) What is a reasonable estimate of the specific influence (e.g., each individual animal lives 77 years)? (3) What simplifying assumptions are reasonable with regard to factors within the model (e.g., the predators become less effective as the prey population decreases)? (4) What simplifying assumptions are reasonable to make with regard to outside influences (e.g., no more than three natural disasters per work day--similar to an ordinary university working environment)? and (5) What can be done to defend these initial assumptions (e.g., consult other references). The value of SD in this staging process (the process of constructing an explicit model of a complex situation) is that it requires learners to make assumptions explicit, to simplify the problem initially, to consider evidence for initial biases, and so on.
Next, let us assume that learners have progressed to a point where they have constructed reasonable models (i.e., models which, when run over a period of time, appear to learners to produce approximations of actual fluctuations in the pertinent animal populations). Our learning goal involved the possibility of introducing a new factor into the model--a policy aimed at controlling one of these populations so as to moderate or eliminate the spread of a disease, for example. Learners are now in a position to modify their models by introducing additional factors of influence to see how they might effect the behavior of the model over time. This can be thought of as a process of hypothesis formulation and testing (as could the model construction process), and it should now be obvious that SD provides very powerful support for this process (e.g., learners can now interact with models they have created and observe consequences).
The point of this example has been to suggest that using SD in a systematic way to implement ID principles is likely to lead to the construction of highly engaging and meaningful learning environments. This has been tried in some selected situations within the system dynamics community and in secondary education settings, and the results are generally very promising.
Since we expect much fruitful discussion and enlightenment to come from ITForum respondents, we have chosen to postpone writing a concluding section until after the ITForum discussion.
The buttons that appear below will be found at the bottom of each page of the discussion. The first button will take you back to the previous page (in this case, to the beginning of paper #9). The middle button will take you to the ITForum home page. The last button takes you forward into the discussion as it progressed on-line.