SAFARI: an Environment for Creating Tutoring Systems in Industrial Training

industrial formation requires specific tools and architecture. The Safari project aims at developing various architectures with different levels of complexity and also a methodology. Creating tutoring systems remains difficult for various reasons : the different components of their architecture are not stabilized, the structure and type of knowledge to be used are of different granularity and complex to handle, they require multiple expertise which is difficult to integrate. However, the utility of tutoring systems having more or less intelligent capabilities is obvious. Most of intelligent tutoring systems have been experimented in academic domains such as mathematics, physics, programming. But do we lack of teachers in such domains ? Is it crucial for improving the present training in universities or colleges ? On the other hand, there is a global market competition in the industry, skills are rapidly evolving (sometimes in complexity) with new technical environment and specialized knowledge, and employees must acquire new functions and know-hows in limited time. The need of training tools for this context is high. Experimenting and developing this category of tools appears different from the prototypes developed for academic courses. Scale of implementation, nature of problems, type of knowledge and expected functionalities are different. This is why we have specialized our research on industrial formation with "Safari". Safari is a project under the auspices of Synergie, a programme sponsored by the Ministry of Technology and Science of the Governement of Québec. The main objectives of Synergie are (1) to enhance cooperative research and development between universities and industry, (2) to accelerate the product development cycle, (3) to facilitate the transfer of knowledge between research establishments and the industry, and (4) to educate highly qualified professionals in the domain. Safari involves four Québec universities, two private enterprises and a government agency. The industrial partners are Virtual Prototypes Inc., providing a software package VAPS, and Novasys Inc. VAPS (Virtual Applications Prototyping System) is a high-quality commercial interface-building and simulation system, used in many areas (such as airline cockpit design). The Safari team includes seven professors as researchers on a part-time basis : C. Frasson (project leader), G. Gauthier, I. Gecsei, G. Imbeau, M. Kaltenbach, S. Lajoie, B. Lefebvre; about 20 M.Sc and Ph.D students, two full-time programmers and two engineers from the industrial participants. The implementation time frame spans 4 years (1993-96). Building reasonable tutoring components The main objective of Safari is to develop a methodology and an environment for the creation of tutoring systems to be used in professional formation. The focus is on teaching mostly procedural knowledge concerning the operation of devices such as medical instrumentation, consumer appliances and aeronautical instruments. The basic idea is to add a tutoring component on the top of device models (microworlds) built in VAPS. This permits to use models written in VAPS, instead of real devices, for training and exercising. Thus the expensive machinery can be left for its actual purpose and damages from incorrect manipulations during learning can be prevented. Safari can be seen as a value-adding component attached to the VAPS package, making it into an integrated toolkit for the creation of device models together with a corresponding computerized training course on how to operate the device. Since a large base of VAPS device models are already available, these can be conveniently used to validate the Safari approach. In order to remain within the bounds of practicality, we limited a priori this class to devices which have the following characteristics: they permit to define set of clearly distinguishable interaction procedures (e.g. through control panels, switches). the device functions are decomposable into a number of well-defined tasks, each task corresponding to a meaningful operation in terms of the device’s main purpose. This excludes very complex equipments such as nuclear reactors (although parts of it may be covered), and devices whose operation involves hard to enumerate, "look-and-feel" actions such as handling certain manufacturing machines, or performing surgery. Tutoring systems created in the Safari environment will be capable of : tutoring workers in an enterprise (factory, hospital) through real tasks recreated on device models using the VAPS software, constructing models of individual users, reflecting the user’s reasoning style and state of knowledge, dispensing individualized tutoring based on the user model, diagnosing the actions of a student when asked to perform a task on the visual model, and composing an entire curriculum for teaching to operate the device, taking into account the learner’s expertise and style of learning. Some distinguishing features of Safari are that : it involves contractual cooperation between the industrial and university partners who are committed to a binding schedule including the development of a commercially viable product, the proposed methodology is applicable to a large class of devices, not only to a single device type, it is based on an existing software package VAPS, and it enhances learning by using multiple representations of knowledge, and by encouraging the student to pass between them. We do not aim at deeply intelligent tutoring; this would be somewhat unrealistic, given the limited success and the problems such systems have even within narrow application domains. Instead, Safari provides a combination of some existing tutoring tools, enabling to build easily and rapidly a tutoring system for a given device. A realistic training progression Throughout the project time frame we use the following functional paradigm. In Safari, tutoring functions (modes) are based on the observation that the natural cycle in which most people acquire a given skill is by first observing someone’s demonstration of the skill, then freely experimenting with the device in question (given the availability of the device, and that such experimenting is not hazardous), then executing precise tasks (assignments) in terms of the device functionalities under the guidance of an expert, and finally by communicating the learned skill to another person. In our approach this translates into four distinct tutoring modes: demonstration, free exploration, coaching and critiquing. These modes are presented to the user through an interface which is essentially the same for all modes, and which displays two major windows: the "Device" window showing the simulated device, and the "Plan" window showing an abstract representation of a task called task graph. In this arrangement the learning process involves two related views of the learning space, encouraging the learner to develop his knowledge simultaneously on the concrete and abstract level. This method has its origins in the "PIF" approach to tutoring, described in (Frasson, C. & al, 1992). PIF supports three interrelated "worlds" (i.e. levels of representation): the Physical, Intentional and Functional. In Safari we retained the P and I worlds, corresponding to the Device and Plan windows. In demonstration mode, the student can observe the execution of various tasks applicable to the device, somewhat like with a VCR based demonstration. However, in Safari the demonstration is based on the VAPS model, executing automatically a task graph (TG) which had been edited by a human expert. A demonstration scenario is a TG augmented by hypermedia comments and explanations attached to certain points of the trace. The student can navigate in the scenario by stopping, repeating, jumping to an arbitrary point (i.e. by clicking on an action in the TG), and by asking for hypermedia explanations. However, he cannot interact directly with the simulated device, only with the scenario. As the demonstration progresses, actions in the TG are highlighted along with the corresponding simulated manipulations of the device (such as operating the controls). The purpose of exploration mode is to permit direct interaction with the microworld (simulated device). Here the student is allowed to freely experiment with the device, to observe the consequences of his actions in a realistic environment, but without the inconveniences of experimenting with the real device (Fath, J. L., Mitchell, C. M., Govindraj, T., 1990). The system may attempt to determine (or guess) which task the learner is trying to execute; in this case the TG of this task is gradually displayed in the task window. Exploration can also be done within the context of a given task; then the student is allowed to manipulate both windows, with automatic highlighting in both windows. In coaching mode the student is asked by the system to perform a given task or to solve a given problem scenario (=sequence of tasks) by interacting only with the Device window. Initially the correct TG is hidden; it is gradually revealed as the learner’s solution is being built. The process is supervised by a coaching component, essentially analyzing the difference between the correct solution(s) from the knowledge base and the student’s construction, and generating appropriate coaching interventions. The fourth tutoring mode iscritiquing, where the student has to prove his knowledge by constructing a solution (in the form of a TG) for a given problem. The learner can "experiment" in the device window as in exploration mode, and build the solution in the task window. In this way he is encouraged to pass frequently between alternative representations in the two windows; in other words, the student has to prove not only that he can correctly execute a task, but also that he understands (to a certain degree, of course) what he is doing. By requesting the student to construct a TG, he is effectively asked to communicate, or to teach his knowledge in an abstract form. When the proposed solution is complete, a critiquing component (similar to the coach) will comment on the solution. At any time during work in the above tutoring modes, the student can ask for help (noncontextual) or contextual advice.