Vladimir A. Goodkovsky, Alexei Yu. Kazennov
Moscow Institute for Physics and Engineering
31 Kashirskoe shosse, Moscow 115409, Russia
Abstract
The paper describes a theoretical systems approach to the design of Intelligent Tutoring Systems (ITS). Putting into practice of that approach allowed:
Introduction
The design of intelligent tutoring systems (ITS) is an extremely complex and expensive process so far. It results in unique products, requires unique versatile qualification of its designers. In other words, the creation of ITS is some kind of art now.
The main objective of our work is to transfer this art into craft in order to arrange mass production of ITS in different disciplines. To meet this objective one needs the development of constructive theory and elaboration of the ITS technology.
A working definition of ITSIn compliance with this direction, ITS is a computer - aided instructional system (CAIS), that provides radically new possibilities for the control and individualization of a instructional process without having a scenario, prepared in advance. It is automatically generated in real time of instructional process with a help of models. So defined, ITS is a model-based CAIS of a generative type.
Traditional approach
It is safe to say that a certain approach to the creation of ITS is formed by now (see, for example, a survey of Brusilovsky, 1990). According to it, first an applied expert system (ES) in a subject area of the envisaged instruction shell be designed, using the available tools and following the traditional technology of ES creation. Then a user interface has to be improved. This design process results in ES, intended for learning support. But still it is not an ITS in our sense.
The tutoring functions themselves are built in ES in a last stage. But because of a priori uncertainty, the labor consumption for their synthesis is rather high. That is why it has to be done on a basis of intuition-heuristic considerations by the designers. This reason cannot guarantee a high quality of functioning of ITSs, which in fact are rather expensive.
Theoretical systems approach
As an alternative to the traditional approach, which can be characterized as a heuristic synthesis, a theoretical systems approach is proposed. It is based on well known methodic of systems analysis (see, for example, Pospelov, Irikov, 1976).
Consistent systems analysis of an ideal portrait of ITS as an automatic instructor, enables the designer to reduce the complexity of an entire instructional problem and to transfer it to an assembly of more simple and better defined tutoring tasks (see, for example, Goodkovsky, 1988).
Utilization of this approach for the analysis of an instructional system, instructor's conceptions (mental models) and instructor's tasks in it, gives the results presented in figure 1. It can be seen, in what system, what kind of tasks and on the ground of what conceptual models the instructor can solve his/her tasks. It is likely that in order to solve the same tasks in the same system ITS may have the analogous models. The next chapter is devoted to the systems synthesis of such models.
The theory of ITS
Further elaboration of systems analysis allows getting a wide range of possible variants of more specific models and tasks. An expedient selection from them is specified by the following requirements to the models:
The general abstract form of all models is chosen in a shape of a well known (Pospelov, Irikov, 1976) mathematical model of a system of a general type:
S = < A, B, C, ...; Rl, R2, R3,.. . >,
Where A, B, C,... - sets; Rl, R2, R3,... - relations.
The model PM of a subject of learning P in S-form is suggested as a model of a lowest hierarchical level of instructor internal representations (mental models). If a degree of detailed elaboration of subject P modeling is fixed, then all possible components of PM (sets and relations) for this level can be considered as being elementary and thus be identified simply by elements Aj. In this case, a model of subject knowledge of a student is represented in a form of fuzzy subset K, that is defined on a basic set PM by the membership function M(Ai). Where M(Ai) may be treated as a degree of mastering. The tutoring objective on this level can be represented in a form of ideal knowledge model of a student.
A model of the next level is a theoretical systems model of a student OM. It is suggested in a shape of the following triplet:
OM = < Ml, M2, M3 >.
Where:
Ml - model of knowledge introduction (input);
M2 - model of knowledge derivation (interior);
M3 -model of knowledge manifestation (output).
The model of knowledge introduction Ml is intended for the solution of tutoring tasks of student knowledge supply and mastery prognostication:
MI = < PM, T; Rl >,
Where:
PM - set of elements Aj of a model of a subject of learning;
T - set of tutoring (learning) materials Tl, aiming to introduce knowledge K about P;
Rl - in a general case is a fuzzy relation, which is given by the membership function MPjl.
MPjl measures a possibility of introducing knowledge element Aj by a tutoring material Tl.
The model of knowledge derivation M2 serves for the determination of a succession in the resolving of tutoring task of the subject knowledge introduction to the student:
M2 = < PM; R2 >,
Where:
R2 - fuzzy relation, given by a membership function MNjh.
MNjh has a meaning of a degree of element Aj knowledge necessity for the introduction of element Ah knowledge.
The model of knowledge manifestation M3 is intended for the solving of knowledge testing and diagnosing tasks:
M3 = < PM, X, Z, V; R3 >,
Where:
X - set of assignments (testing and diagnostic tasks and questions ) Xi;
Z - set of all subsets Zi of possible (right - ZRi, wrong - ZWi and mixed ZMi) answers Zik of a student on the assignment Xi;
V - set of all subsets Vik of possible ways Viks for the students to fulfill the assignment Xi and get an inswer Zik;
R3 - fuzzy relation given by the following membership functions:
- MNjiks - has a tutoring meaning of a degree of necessity, that a student should know element Aj :for getting answer ZRik or ZMik by using Viks;
- MPjiks - has a tutoring meaning of a degree of possibility, that not mastering the element Aj by the student would cause getting answers ZWik or ZMik on the assignment Xi by using Viks.
It is worth noting, that terms "necessity" and "possibility", which have been used here in the tutoring sense, in fact have the strict mathematical interpretation in the theory of possibilities (Dubois, Prade, 1990).
The tutoring objective on this level of modeling (that is model of ideal OM) can be represented in the form M3, that has only right answers Z. In this case M3, treated as a model of tutoring objective, is presented in a declarative mode. A procedural model of a tutoring objective on this level can be represented by the appropriate ES, which can fulfill the assignment X in a way, that a well trained student can do.
A treatment of a suggested OM may be broadened if necessary. For that purpose, besides cognitive (knowledge and skills) characteristics, the psychological characteristics of a student, that has an internal representation (PM) and external manifestation (Z), shall be considered as well.
The models of the next level (these are learning system LS models) can be designed on a base of model OM, mentioned above, by extending the margins of the definition of its inputs (T,X) and outputs (Z). The extended T and X include not only direct supply (learning materials) and assignments (questions and tasks) from the instructor to the student, but the possible control influences of the instructor on the learning subject itself as well. Similarly, the extended Z may include not only immediate answers of a student on the assignment X, but also the other actions of the student, observed by the instructor in a learning process, and the reaction of a learning subject as well. The model of a tutoring objective on this level is an extended M3, which describes an ideal behaviour (Z) of LS as a response on the instructor's assignment (X).
LS models are used in those cases, where the instruction process is not as simple as a dialogue between the instructor and the student about an abstract subject, but has a form of supervision of the process of a student interaction with a real (controlled and/or monitored by the instructor) subject or with its model. This is typical for different kind of simulators and games. Naturally LS models have higher dimension than OMs have. LS models design is rather serious problem, but it can be solved successfully in some cases.
All models listed above, are directly related to a specific subject, particular features of an instructor activity and provide all necessary, basically declarative, knowledge. It gives an opportunity to derive procedural knowledge that is also needed, independently. ITS with such decoupling of tutoring knowledge may be accomplished in a shape of a multi-purpose procedural shell, which is filled in by a specific declarative content.
Utilization of models Ml, M2, M3 on a stage of applied ITS design allows easily formulating and efficiently solving the following tutoring tasks:
On the stage of ITS practical implementation, those models allow easily to solve current tutoring tasks of choosing the next:
As well as to solve current tasks of passive knowledge assessment:
Proper algebraic operations for the solution of the above tasks are not unique and can be determined by a selection of a scale for presentation and measuring of membership function, i.e. degrees of necessity MN and possibility MP.
A local optimization, which uses a selection of the next step (Tl, Xi) by the criterion of minimum of efforts/time that a student spends on reaching the learning objectives under a set of pedagogical limitations, is often sufficient for instruction on the base of these models.
The technology of ITS
The technology of instruction, implemented in ITS, facilitates the general goal of a student to accomplish the learning objectives by the use of a wide spectrum of possible ways. A student, besides an instructional unit, can choose an operational mode of ITS (for example, an electronic textbook, testing, diagnosing, debugging, and practicing), the form of subject material presentation (uniform, individual, self-dependent program) and the ways of testing (initial, intermediate, final). Besides, the following items may vary in a wide range: the desired redundancy of instruction, the frequency of testing, reliability of testing, the tolerance of ITS to the student mistakes and the reliability of their causes diagnosing. All these parameters of ITS are set up by the author-instructor.
The most general mode of ITS operation is an instruction by the individual program with a current progress testing. Its algorithm is based on principles of dual control and includes: a cycle of active supply with a passive prognostication, testing and diagnosing, a cycle of active testing with a passive instruction (commenting) and diagnosing, and a debugging cycle. A debugging cycle, in turn, includes: a cycle of active diagnosing with a passive instruction and testing, a cycle of active correction (or explanation) and testing its success.
The design technology of applied ITS for the specific domain of a subject matter includes the following phases:
The practice of ITS
Within a described technology a number of applied ITS have been implemented. The content was taken from the following subject areas: mathematics, physics, chemistry for schools, and other subjects for instruction of students, operators, and workers. The most advanced commercial applied ITS have been managed to be implemented for training of Nuclear Power Plant personnel.
Using theoretical systems approach as the base, some special independent software tools have been realized: the expert knowledge extracting system, authoring tools, the interactive system of qualification assessment, decision making system for control over training process, system for the diagnostics and individualization of training, diagnostic scenario generator, etc. Those software tools were used for the creation of commercial ITS, such as "Decision making simulator", "Main circulation pump intelligent simulator"(Kazennov, Kucherenko, 1991).
At present an integrated-software-tools-package is being designed on that base with the purpose to support all phases of the ITS creation.
Conclusion
The presented approach allowed to treat the problem of ITS creation not traditionally, that means not starting with the development of applied ITS in a subject area, but originating modeling of knowledge and automation of ideal instructor functions. The results, received in this case, have original importance and may be applied to more simple ITSs in particular, which don't have expensive ES.
In the last case the suggested models and methods enable to reduce a sophisticated intellectual process of ITS creation to a simple filling in generic frameworks with the specific expert knowledge. This lessens the labor consumption significantly, improves the quality of knowledge, received from experts, and makes the feasibility of creation of this kind of ITS quit real.
Acknowledgements
The authors are thankful to the colleagues from ENIKO 14 for their support, assistance and friendly cooperation in every day work.
References
Brusilovsky P.L. (1990) Intelligent Tutoring Systems, Informatics, v.l, n.2, p.3-22
Goodkovsky V.A. (1988) Automation of Cognitive Diagnostics for Adaptive CBT: Theory and Application. Ph.D. Thesis. Bauman Moscow State Technical University. - 295 p.
Dobois D., Prade H. (1988) Theorie des possibilites. MASSON Paris Milan Barcelone Mexico, 1988. - 286 p.
Kazennov A.Yu., Kucherenko A.A. (1991) Systems approach to NPP personnel training process design: methods and software tools//IAEA Specialists' Meeting on "Training Simulators for Safe Operation in Nuclear Power Plants". 24-27 September 1991, Balatonfured Hungary.
Pospelov G.S., Irikov V.A. (1976) Program-purpose planning and control. - Moscow: Sov.Radio, 1976. - 440 p.