Proc. of the 2005 Westen Simulation Multiconference SIMCHI 2005, International Conference on Human-Computer Interface Advances for Modeling and Simulation, January 2005 Combining Intelligent Tutoring and Simulation Systems Alke Martens University of Rostock Albert-Einstein-Str. 21 alke.martens@informatik.uni-rostock.de Keywords educational computer simulation, intelligent tutoring system Abstract Intelligent Tutoring Systems (ITS) are well established in educational settings for more than 30 years. To use simulation systems for training is also standard in a lot of domains. Usually, the purpose of simulation in educational settings is the training of certain aspects of an application domain. The system described in this paper combines ITS and simulation in a casebased training environment. In contrast to traditional ITS or simulation systems, the purpose of the described approach is to help the students to learn modeling and simulation in exemplary training cases. The so called Intelligent Tutoring System for Modeling and Simulation, abbreviated as TutMoSi, can be perceived as an extension of JAMES II (Java Agent Modeling Environment for Simulation). JAMES II allows the usage of different modeling formalisms. Extended with an ITS interface, JAMES II shall allow for teaching and training of modeling and simulation in a case-based manner. In TutMoSi Students should interact with a customer, who provides some aspects of the model to develop. After the acquisition of the required facts, the student has to choose an adequate modeling formalism and to construct the model. After creating the model the student has to validate the model by making simulation runs and by discussing the evaluated results with the customer, to check whether he has met the requirements. The result of this may be that the student has to refine or change his model. Thereby the ITS shall support the student with feedback, correction, and help in each of the described steps. Application domain of the training cases is life science. Jan Himmelspach University of Rostock Albert-Einstein-Str. 21 jan.himmelspach@informatik.uni-rostock.de 1 INTRODUCTION Intelligent Tutoring Systems (ITSs) can be traced back to the early 1970s, when the upcoming Cognitive Science in combination with Artificial Intelligence has led to a new type of teaching and training system (e.g. [4]). In contrast to the traditional Computer Based Training Systems (CBT), these ITSs are based on expertsystems, containing the knowledge of the application domain. Moreover, nowadays ITSs usually contain a learner model. The system’s flexibility in combination with the learner model allows for adaptation to the learner’s performance, the learner’s expertise and the learner’s needs. For more information about ITS see e.g. [5, 15]. Simulation systems have always been used to support teaching and training, e.g. flight simulators. Here, the simulation itself is the medium to teach and train. To use simulation systems in ITS is also not a new approach. Here, the simulation takes over one of two roles. On one hand the simulation is used to steer a certain component of the ITS, e.g. the simulation of a virtual tutor. On the other hand, the simulation is used to train a certain aspect of the application domain, e.g. simulation of electrical circuits or medical simulations. Here, additional to the ”pure” simulation, the ITS provides knowledge about pedagogical approaches and about problem solving strategies. In contrast to this usage of simulation, almost no systems actually train modeling, simulation and evaluation of the simulation’s result. Thus, whereas simulation is a good medium for teaching and training, the teaching and training of modeling and simulation has seldom been realized. Case based training, where the learner takes over a role in a training case, provides the adequate environment for establishing teaching and training of modeling and simulation. To develop an ITS, which allows for simulating models the learner has modeled, techniques of ITS have to be combined with a simulation system. Here, we use the system JAMES II, which provides the basis to simulate models based on different modeling approaches. For example, JAMES II is able to run simulations in a discrete event, and even in a discrete time stepped way. It soon will be possible to provide support for continuous simulations as well. Thus, the system is able to support every model the learner thinks is appropriate for the given task. In the following, we will start with a description of the different ways of using simulation in ITS. We will throw a short glance at how case based training can be realized to support teaching and training of modeling and simulation. After that, an introduction into JAMES II will be given. Finally, we will describe, how JAMES II can be used as a constituent of an ITS. 2 SIMULATION IN ITS Simulation systems have always been used for teaching and training. However, the usage of the simulation in a training environment is quite heterogeneous. In the field of educational computer simulations, we can distinguish between at least three different ways: 1. Interactive training simulations 2. Demonstrative training simulations 3. Character simulations In interactive training simulations, the learner uses the simulation system to train a certain aspect of the application domain. The learner is not necessarily aware of the simulation itself. He interacts with an offered environment and e.g. solves the described problems (e.g. [19] and [14]). One of the first interactive training simulations is the system SOPHIE (SOPHIsticated Instructional Environment), developed by Brown, Burton and Bell [2] and Burton, Brown and de Kleer [3]. Using SOPHIE, the learner shall learn the functionality of electrical circuits. SOPHIE offers an electrical circuit containing a bug. This bug shall be located and repaired by the learner. Via continuously simulating the circuit, SOPHIE checks the repair strategy of the learner. Also the well-known flight simulators are part of the category of interactive training simulations. Sometimes, as described in [7], the interactive training simulation can also be used to learn something about the learner himself. Thus, it can provide the basis of psychological and cognitive research. In the demonstrative training simulation, the learner has no possibility to interact. Here, the visualization of aspects of the training domain is the focus of interest. Simulations can for example easily be embedded as subprograms in case based ITS in the application domain of clinical medicine, like [6]. To embed a character simulation in an ITS most of the time means to embed a virtual persona in the ITS. The aim of such an approach is to support the learner and to personalize the training with the ITS. The virtual persona is often realized as a virtual tutor, offering feedback and help on demand or proactively. Two virtual tutors are well known in this area: STEVE and ADELE [13, 12, 17]. STEVE is the ”Soar Training Expert for Virtual Environments”. Realized as a three dimensional agent, STEVE is capable to virtually interact with the training environment. STEVE is not only based on a simulation, moreover it is embedded in an interactive training simulation. ADELE, the ”Agent for Distance lEarning, Light Edition”, is the two dimensional pendant of STEVE. In contrast to STEVE, ADELE acts as virtual tutor in a case based medical ITS, and not in an additional simulation environment. 3 CASE-BASED TRAINING FOR MODELING AND SIMULATION The aim of the system described in the following is to teach and train modeling, simulation and to evaluate the results of the simulation. The system intended is a combination of a classical ITS based on a formal tutoring process model [15, 16] and the simulation system JAMES II which is able to run models of different types. Result of this combination is an interactive training simulation, referring to the categories described in the prior section. In contrast to the categories described above, the simulation is not only the medium but the subject of teaching and training. Case-based training is famous in a lot of domains like clinical medicine, education in law, and in business education. Classroom teaching of modeling and simulation often takes place in a case-based or problem-based manner. Students learn modeling and simulation techniques based on examples. However, there seem to be no systems yet, which allow for web-based and case-based training of different modeling and simulation techniques in an ITS. In [1], an approach of using a multimedia ITS for simulation modeling has been described. The paper focuses on the multimedia aspects of simulation modeling, thus it lacks a description of how modeling and simulation should be trained with the system. In the casebased approach, pursued in our system, the learner takes over the role of a computer scientist in a scenario, where a client wants to have a simulation. Via text bricks, the client ”tells” the computer scientist, which kind of system he wants to have simulated. Thus, our system starts with the data collection, given an unordered and not scientifically specified amount of information. The learner has to decide, which kind of modeling approach is appropriate for the system described. He must decide whether information is missing and how to structure the information given. After the learner chose a modeling formalism, he has to develop a model. The ITS, equipped with an expert knowledge base, can compare the learner’s modeling approach with an expert’s solution. If the model is completed, the learner starts the simulation and can have a close look at the simulation results. He can decide whether the simulation is appropriate to the client’s demands, or whether some re-modeling has to be done. The simulation run and the simulation results can be evaluated by the ITS. The graphical user interface of the ITS should allow for different modeling techniques. For example, modeling of UML diagrams, and of Petri nets should be possible, as well as modeling of data based only on XML or on formal descriptions. However, for the first system developed, a simple modeling dialog has been constructed. The learner can decide, which parts of the information given in the text shall constitute a model. He can decide, which properties are required and how these properties shall develop over time. Furthermore, the learner must decide about the models’ coupling structure. For this purpose, in the first training case realized, a simple grid of atomic models is displayed to the learner. Using this grid, the learner can define model positions, neighborhoods, and influences. This data is stored in XML and used to run the simulation. Via the interface, the learner can pursue the simulation, stop it and view the simulation results. The initially developed models are re-visitable and correctable. 4 JAMES II JAMES is a modeling and simulation framework created for being applicable to a broad range of diverse applications. For meeting this challenge JAMES had to be designed in a highly flexible manner. Thus, everything in JAMES is considered to be a component. The component design includes communication via interfaces and allows for exchangeability. The design of the system resembles the idea that a modeling and simulation framework should clearly separate between the modeling and the simulation layer. The advantages of this separation are the following. • There may be more than one simulator for a particular model, which adds to the system’s flexibility. • There are no side effects between model and simulator. The current version of JAMES, i.e. JAMES II, supports any number of modeling formalisms and any number of execution mechanisms. So far we have realized support for PDEVS, dynPDEVS, epiPDEVS, epiDynPDEVS and CA (cellular automatons). Thereby the PDEVS formalisms can be executed either in a sequential, parallel or in a combined parallel sequential manner. In addition they can be executed paced, i.e. the simulation time relates to wallclock time during execution, or unpaced, i.e. as fast as possible [20, 9, 18, 10, 11]. The epi-formalisms provide support for the integration of external processes, the dyn-formalisms provide support for Design Visualization Runtime Data Analysis Model access data flow Simulator Figure 1: JAMES dynamic structures [11]. We are currently planning to create a set of components to support pure continuous ITS Design as wellVisualization as hybrid models. Runtime Figure 1 shows a top level view of the Analysis components of JAMES. These top level components are the simulator, the model, the visualization, and the data. Data Model Simulator The simulator package contains simulator components. access A simulator maybe specific for one model Simulator type or it may be able to compute a broader range datatypes. flow In addition there may be more of model than one simulator per model type. Furthermore the simulator package contains components for supporting distributed model execution (sever management, partitioning). A concrete simulator component for running a simulation is build based on the components of this package. Model The model package contains the basic executable model components. It also contains the items which are required in each model, e.g. couplings for DEVS models. Each model to be executed within the framework must support the basic interfaces defined in the corresponding base model package. The model components provide support for the observer pattern [8], i.e. data produced by models is not directly stored by the models themselves. Observers are listening for changes and forward them to the data component. A concrete model component (a model) is build based on the components defined in here. Visualization The visualization component is currently under development, but it already provides a basic GUI (graphical user interface) which allows to parameterize and start a model. Currently the focus for further development is on how to model which formalism visually, e.g. with statecharts. In principle a GUI should provide support for modeling, runtime (i.e. control, visualization) and analysis. access Simulator data flow ITS Visualization Data access data flow simulation run, the user has access to the data component’s entries, which are displayed in the ITS. Thus, he has information about the simulation run available, and can revisit and correct his model, if required. Design Runtime Analysis Model 6 Simulator Figure 2: ITS and JAMES II Data The data component is used to provide support for storing the data generated by a simulation run. There is no restriction on how the data has to be stored - the data will be delivered by observers and can be written into plain ASCII files, XML files or even a database (as long as a proper storage component exists). 5 COMBINING JAMES II THE ITS AND Figure 2 shows how JAMES II can be integrated into the ITS system by adding a ITS layer to the JAMES II architecture. Combining the ITS system with JAMES II is a rather simple task due to the component oriented design of JAMES II. Each part of the framework is a component composed of other components. Thus, exchanging complete parts of the framework or reusing any combination of parts in any other system is possible. From the framework’s point of view the ITS will simply replace the visualization part. From the ITS’s point of view the components of the visualization component are reused. The reminder of the framework is just used as it is. Thus, given the description of JAMES II in the former section, the ITS uses the simulation package, the model package and the data component. The ITS is responsible to display information to the learner. The learner abstracts the information of a model, i.e. he uses the modeling component of the ITS. The information recorded by the learner is internally transformed to an XML model. Also the model structure, the neighborhood and the couplings are stored in XML. These learner-made models must meet the requirements for executable models, defined by JAMES II. Thus, if information is missing, the ITS will correct and help the learner to complete his model. If the learner decides to start the simulation execution (and if the models are ready for simulation), he selects a simulator component of the simulation package. After the FIRST IMPLEMENTATION In our first implementation, the learner is confronted with a classical modeling and simulation training case: he should develop a model and run a simulation of the ”forest fire” scenario. In our training case, the client provides background information, like: ”There is a river crossing the forest. The trees in the forest are quite old, but there are some younger birches at the river bank.” The learner has to decide, which parts of information he will use to develop an appropriate model. He uses a modeling editor to note each model’s attributes and properties. The transfer of the learner’s model into JAMES is realized with a model description in XML. This XML description is also used to compare an optimal ”expert-”model with the learner’s model. The model itself is distinguished in essential parts, which must be part of the resulting model, and optional parts, which might be embedded. Thus, the learner will be immediately corrected if he wants to start a model with missing essential parts. The result of the missing optional parts might influence the simulation, but will not be immediately commented by the program. The first implementation of our teaching and training program uses a simple grid as basis for the spatial distribution of the different models. Via marking regions in the grid and associating models to the regions, the learner models the forest with the parts necessary for simulating the described training case. If the modeling is finished, the learner can run a simulation, and have a look at the results. If he feels that the model is not appropriate, he might revisit the modeling part of the program and make model adjustments. 7 CONCLUSION To combine simulation and ITS has been realized for quite a long time. The combined systems can be categorized in three groups, i.e. interactive training simulations, demonstrative training simulations and character simulations. In the first kind of teaching and training simulation, the simulation steers the interactive environment of the training domain, e.g. a flight simulator or a simulation of an electrical circuit. The second kind of simulation is only used for demonstration purposes - the learner has no possibility to interact. In the third kind of simulation, a virtual persona is simulated. This kind of simulation is an add-on to an ITS. Thus, a character simulation can be part of an interactive training simulation, like STEVE. Whereas the learner steers the sim- ulation in the first group, in the third group the learner interacts with the virtual character, but not with the simulation itself. Another kind of interactive training simulation is described in this paper. Here the learner starts his work with an ITS without simulation. He has to solve the task to develop a model, given a large amount of unstructured, relevant and irrelevant knowledge. After the learner has finished the model development, he has to start a simulation. This simulation of his model can be interrupted by the learner at any point in time. The learner can revisit his model, make corrections, and start the simulation again. He also has the possibility to completely change between modeling formalisms. For example, the learner might start with a discrete-step wise model, and learns that this does not really meet the requirements of the client, then he might decide to develop a discrete event model and re-run the simulation. After the simulation run, the learner has to interpret the results. This kind of interactive training simulation does not simulate a certain aspect of the training domain. In the system type described in the categorization above, the learner has to interact and the simulation steers the training environment. In the system TutMoSi, simulation is one embedded part of an ITS. The simulation itself is steered by the learner. The system TutMoSi that is based on a combination of an Intelligent Tutoring System (ITS) and a Modeling and Simulation environment, JAMES II, allows for interactive training of modeling and simulation, based on simulation. 8 OUTLOOK The system described in this paper is work in progress. Whereas the ITS and the system JAMES II exist and provide full functionality, the combination of both is currently realized. Our first training case is the classical ”‘forest fire”’ simulation. Other exemplary training cases, e.g. in systembiology and medicine, are currently developed. However, the main problem in developing an ITS for Modeling and Simulation can be seen in the interpretation and the validation of the learner’s modeling success. One big problem is for example to decide what makes a model optimal. Here, a didactial solution is necessary. Development of training cases in this domain is a very demanding task. Thus, the first application realized is on a rather coarse level, regarding the potential abilities of an ITS. References [1] T. A. Atolagbe and V. Hlupic. Simtutor: Am multimedia intelligent tutoring system for simulation modeling. In S. Andradottir, K. J. Healey, D. H. Withers, and B. L. Nelson, editors, Proc. of the Winter Simulation Conference, 1997. [2] J. S. Brown, R. R. Burton, and A. G. Bell. Sophie: A Sophisticated Instructional Environment for Teaching Electronic Troubleshooting. Technical Report 2790, BBN Report, 1974. [3] J. S. Brown, R. R. Burton, and J. de Kleer. Pedagogical, Natural Language and Knowledge Engineering Techniques in SOPHIE I, II and III. In D. Sleeman and J. S. Brown, editors, Intelligent Tutoring Systems, Computers and People Series, pages 227–279. Academic Press Inc., London, 1982. [4] J. R. Carbonell. AI in CAI: An Artificial Intelligence Approach to Computer-Assisted Instruction. IEEE Transactions on Man-Machine Systems, 11(4):190–202, 1970. [5] A. T. Corbett, K. R. Koedinger, and J. R. Anderson. Intelligent Tutoring Systems. In M. Helander, T. K. Landauer, and P. Prabhu, editors, Handbook of Human-Computer Interaction, pages 849–874. Elsevier Science B.V., 1997. [6] Docs ‘n Drugs. - Die Virtuelle Poliklinik. http://www.docs-n-drugs.de, 2000. [7] D. Doerner. Die Logik des Misslingens. rororo science. Rowohlt Taschenbuch Verlag, Hamburg, Germany, 2003. [8] Erich Gamma, Richard Helm, Ralph Johnson, and John Vlissides. Design Patterns: elements of reusable object-oriented software. Addison-Wesley, Reading, MA, USA, 1994. [9] J. Himmelspach, M. Röhl, and A.M. Uhrmacher. Simulation for testing software agents - an exploration based on JAMES. In Winter Simulation Conference, New Orleans, 2003. SCS. [10] J. Himmelspach and A.M. Uhrmacher. A Component-based Simulation Layer for James. In 18th Workshop on Parallel and Distributed Simulation, Kufstein, 2004. IEEE Computer Society Press. [11] J. Himmelspach and A.M. Uhrmacher. Processing dynamic PDEVS models. In MASCOTS conference, Volendam, 2004. IEEE Computer Society Press. [12] W. L. Johnson, J. Rickel, and J. Lester. Animated Pedagogical Agents: Face-to-Face Interaction in Interactive Learning Environments. International Journal of Artificial Intelligence in Education, 11, 2000. [13] W. L. Johnson, E. Shaw, and R. Ganeshan. Pedagogical Agents on the Web. In Proc. of Intelligent Tutoring Systems, ITS98, 1998. [14] Kinshuk, R. Oppermann, R. Rashev, and H. Simm. Interactive simulation based tutoring system with intelligent assistance for medical education. In T. Ottmann and I. Tomek, editors, Proc. of the EDMEDIA / ED-TELECOM, pages 715–720. AACE, 1998. [15] A. Martens. Ein Tutoring Prozess Modell fuer fallbasierte Intelligente Tutoring Systeme. PhD thesis, Rostock, Germany, 2003. [16] A. Martens and A. M. Uhrmacher. A Formal Tutoring Process Model for Intelligent Tutoring Systems. In Proc. of the European Conference on Artificial Intelligence, Valencia, Spain, 2004. [17] J. Rickel and W. L. Johnson. Animated Agents for Procedural Training in Virtual Reality: Perception, Cognition, and Motor Control. Applied Artificial Intelligence Journal, pages 343–382, 1999. [18] A.M. Uhrmacher, M. Röhl, and J. Himmelspach. Unpaced and paced simulation for testing agents. In Simulation in Industry, 15th European Simulation Symposium, pages 71–80, Delft, 2003. SCSEuropean Publishing House. [19] P. van Schaick Zillesen. Using Educational Comuter Simulations. Van Hall Institute, http://www.xs4all.nl/ eszet/personal/educsim.html, downloaded Sept. 2004, 1998. [20] B.P. Zeigler, H. Praehofer, and T.G. Kim. Theory of Modeling and Simulation. Academic Press, London, 2000. AUTHOR BIOGRAPHIES Alke Martens is Dr.-Ing. of computer science. She made her phd in the area of formal methods for intelligent tutoring systems. Her research areas are modeling and simulation, teaching and training systems, and a combination thereof. Current application domains of her research are medicine and education of computer scientists. She is currently a research scientist at the Modeling and Simulation Group at the University of Rostock. JAN HIMMELSPACH holds an MSc in Computer Science from the University of Koblenz. His research interests are developing methods for modeling and simulation, with a focus on effective and efficient simulation mechanisms. He is currently a research scientist at the Modeling and Simulation Group at the University of Rostock.