An Intelligent Tutoring System Education Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Intelligent Tutor Systems have been around since the early 1970s, but has suffered an increased in popularity in the 1990s. In the early 1970s a few researchers defined a new and determined goal for computer-based teaching. They implemented the artificial intelligence techniques in computer- based system to make it intelligent like a human tutor. Research efforts over twenty-five years have generated some notable successes in achieving the promise of intelligent tutoring (Lesgold, Eggan, Katz and Rao, 1992; Koedinger, Anderson, Hadley and Mark, 1995). Personal human tutors provide a highly effective learning environment (Cohen, Kulik and Kulik, 1982). The goal of intelligent tutoring systems (ITSs) is to involve the students in continued cognitive activity and to interact with the student based on a deep understanding of the student's behaviour. This review will be based on recent research on ITS and describe briefly how ITS is going to be used in different fields of life and latest research on emotions involve in ITS; it shall be centred around keywords: 'intelligent tutoring system', 'expert system' ,'domain knowledge', 'misconception', 'student knowledge' and 'affective intelligent tutoring system'.

Today ITSs are progressively used in various academic institutions in a number of courses, such as learning database concepts (Mitrovic, 2006), language learning, physics (VanLehn, Lynch, Schulze, etc., 2005), geometry, algebra, and programming languages. The first intelligent tutoring program, SCHOLAR (Carbonell, 1970) was implemented in the field of education. This program attempted to involve the student in a mixed initiative dialogue on South American geography by communicated through a sequence of natural language questions and answers. This tutor was based on a semantic network model of domain knowledge which could both ask and answer questions and keep record of the on-going dialogue structure. The ITS generally consists of four main components: the domain knowledge module, student model and modeller, pedagogical module, and the communication module. Sometimes, a fifth component, the expert module containing an expert's knowledge, is also included (Beck, Stern and Haugsjaa, 1996).

Figure 1: Component of intelligent tutor architecture

The Problem-Solving Environment consists of an editor that accepts and represents student problem solving activities. For example, in the case of programming tutors, the interface may be a text editor (Johnson and Soloway, 1984), structure editor (Anderson and Reiser, 1985) or graphical editor (Reiser, Kimberg, Lovett and Ranney, 1992). An imitation's response to student actions provides feedback that enables the student to reason about the problem solving situation. Finally, simulations can return significant savings in learning time.

The Domain Expert module is at the heart of an intelligent tutoring system and provides the basis for construing student actions. It works as an expert system that can generate solutions to the same problems the student is solving. It is important to develop a true cognitive model of the domain knowledge that solves problems in the same way students solve them.

The Student Model is a record of the student's knowledge state. It consists of two components: an overlap of the domain expert knowledge and a bug catalog. The first component is a copy of the domain expertise model in which record of how well the student has learned it, is kept. The bug catalog is set of misconceptions or incorrect rules that are acquired by the student. Much of the early interest in intelligent tutoring systems was motivated by diagnosing student misconceptions, but this early interest has faded for two reasons. First, continuing misconceptions are less frequent than assumed (VanLehn, 1990). Second, a persuasive paper by Sleeman, Kelly, Martinak, Ward and Moore (Sleeman. Kelly etc., 1991) indicated that providing remedial feedback is no more effective than instruction that simply reteaches the correct approach.

The Pedagogical Module is responsible for structuring the instructional interferences. This module operates at two levels (Pirolli and Greeno, 1998). At the curriculum level it can sequence topics to ensure an suitable precondition structure is observed (Capell and Dannenberg, 1993) and modify the amount of repetition at each level to ensure the students master the material (Corbett and Anderson, 1995b). It is also interfere to advise the student on problem solving activities at the problem solving support level.

On the basis of this architecture, significant research has been done and number of ITS are implemented in academic institutions. Barnes and Dr. Donald Bitzer introduced a 'Fault Tolerant Teaching System' in their paper of "Fault Tolerant Teaching and Automated Knowledge Assessment" which is automatically assessing student knowledge and uses this assessment to direct remediation of knowledge by using the statistical techniques (Barnes and Bitzer, …). FTT allow usual errors that occur during testing - such as a student answering a question correctly without knowing how, or accidentally missing a question they understand well and effectively guide the teaching process. The Q-matrix method is used to provide distance education tools with the ability to get tutorials for any subject to all types of students. This method also helps to measure and adapt the effectiveness of a variety of teaching strategies. In their paper they introduced the Data Mining techniques to determine what a student understands during the course of instruction. Human get things wrong for the right reasons, and get things right for the wrong reasons. Data mining skills use to evaluate the student behaviour, assignments, and tests to assess each student's level of knowledge and expertise. The complex task of assessment of student knowledge requires tolerance of such errors which FTT support in well manners. The results of a good test or assessment do not mean just report those questions a student missed, but its reflection of the skills and understanding of student's performance. The results of a good assessment offer reliability and heftiness across time and circumstance (Dietel, Herman and Knuth, 1991).

Knowledge assessment is main task of creating tutoring systems online (Stauffer, 1996). A computer tutor must be able to identify and correct student misconceptions, and should be able to differentiate these from careless errors or guesses. Once it has been done appropriately then a tutor that may be human or computer must determine how to best lead a student to reach her educational goals. However, much of the research in analysing misconceptions concedes the importance if distinguishing 'slips' from true misconceptions (Tatsuoka, 1983; Birenbaum, Kelly, & Tatsuoka, 1993). An example of a binary Q-matrix is given in Table 1. In Tatsuoka's work (Birenbaum, Kelly, & Tatsuoka, 1993), a Q-matrix, also called an attribute-by item incidence matrix, it contains one if a question is related to the concept, and a zero if not. Brewer comprehensive these to values ranging from zero to one, signifying a probability that a student will answer a question incorrectly if he does not understand the concept (Brewer, 1996).


















Table 1: Example binary Q-matrix

The latest research has been done on the role of emotional agents integrated in ITS which is motivated by the social cognitive theory signifying that learning takes place through a complex interaction between both cognitive and affective dimensions (Kapoor and Picard, 2005). Research works in cognitive sciences show that emotions enable people to communicate efficiently by regulating social interaction (Damasio, 1999) or by evaluating and modifying emotional experiences (Thompson, 1994).

Roger Nkambou introduced emotions cohesive in ITS in their paper of 'Towards Affective Intelligent Tutoring System' (Mendez and Miron, et al. 2006). They integrate two adaptive emotional agents in a multi-agent AITS to promote a more vibrant and flexible communication between the learner and the system. The first one allows the tutor to express emotions in reply to the student's actions. An emotional tutor, called Emilie-1 has been instigated in a learning environment for on-line teaching of science. The second emotional agent Emilie-2 captures and manages the emotions expressed by the student during a learning session.

Figure 2 describe the multi-agent architecture of AITS which shows those two emotional agents. Figure shows that the learner's model represents both his cognitive state (knowledge, skills and performances' history) and his affective state (mood, emotions and psychological profile). Profiler manages the cognitive state which updates the acquired knowledge, skills and performances of the learner, and maintains the cognitive model integrity. It also indicates knowledge or skills incorrectly learned, or missing, and permits remediation with the help of tutoring agents.

Figure 2: Architecture of AITS taken from (Mendez and Miron, 2006)

The affective state contains short-term information (resp. medium and long-term), which corresponds to emotions (resp. mood and psychological profile) of the learner. Emilie-2 manages this part of the model. Tutoring agents participate to the training like selection of relevant learning activities, help students during problem solving activities, etc. The emotional state of Emilie-1 represents the current emotional reaction. The events are sent out through the virtual laboratory (learning interface) which contains interactive tools (objects and resource) related to the content. These events are logical representations of specific user actions.

Basic architecture of Emilie-1 based on 4 layers, Emotion Generator, OCC layer, Motor Layer and interface layer. The OCC Layer uses intervals and sign algebra to signify the state of the agent toward the emotional couples which is stated in the OCC Model at a given time using an expert system. The Emotion Generator is a set of associations on events, plans and records of past events that tempts variations in the OCC Layer. The motor layer defines how emotions expressed in the OCC Layer are translated into a demonstration in the agent's interface. The Interface Layer is a definition of the agent's appearance free of geometrical considerations which is used to produce a visual output of the agent. The advantage of this architecture is that the information to produce emotional responses for Emilie is very limited.

Figure 3: Architecture of Emilie-1 from (Mendez and Miron, 2006)

Emilie-2 is made of three layers as shown in Fig. 4:

Figure 4: Architecture of Emilie-2 from (Mendez and Miron, 2006)

The first one (perception layer) captures and extracts the facial expressions (image acquisition and face tracking) and proceeds to its categorization (classification); the second one (cognition layer) analyses and diagnoses the learner's emotional state and the third one (action layer) takes decision on mixture pedagogical actions that is carry out in response to the actual emotional state (Chaffar and Frasson, 2005). Tutoring agents are then get information in the new affective state (updated by Emilie-2) to adjust the current tutoring flow accordingly. But result shows some problems with the projected perception component: 1) Face detection using Eigen face method is too limited and less accurate. 2) The facial expression extraction mostly based on variation of the face. 3) The perceptron neural net is very sensitive to the entry variations. 4) The current version of the perception component doesn't takes into account the emotion valence.

Intelligent tutoring systems have provided a luxuriant ground for artificial intelligence research over the past twenty-five years. Some of these systems have left large impact on educational outcomes in field tests, including effective learning rate, asymptotic learning levels and motivation. The future of ITSs and learning looks very promising and electrifying. For the time being, although evaluation studies have shown very high learning rates in ITSs (Mitrovic and Ohlsson, 1999; VanLehn, Lynch, Schulze et al., 2005), but still there is more research is required to work on different aspects of ITSs to increase the rate, depth, and effectiveness of learning to match human one-to-one tutoring.

In the face of the disappointing results of ITS, experts suggest that, appropriate role for a computer is a mind-extension rather than of a teacher/expert (Derry & Lajoie, 1993). Cognitive tools are not intelligent tools which depend on the learner to provide the intelligence, not the computer. This shows that planning, decision-making, and self-regulation are the accountability of the learner, not the technology. Intelligent Tutoring System must enhance student-tutor interactions with more personalized communications (explicit and/or implicit) that is based either on cognitive and affective behaviour. This provides flexible and thrilled tutoring process during learning process, help and explanation.