AI-based Edutech for Adaptive Teaching and Learning

Abstract

The artificial intelligence (AI)-based problem learning system quickly and accurately performs problem setting and scoring using algorithm. In this process, the learner’s level of prior learning is identified, the subject and quantity to be learned are determined and problem learning is provided for each learner. The basic use of AI-based problem learning enhances ease and fairness in performing assignment and evaluation and provides data that can strengthen interactions between instructors and students. Above all, the biggest advantage is the possibility of helping individual learners with different levels of prior learning to strengthen basic learning. To this end, instructors need to understand the technical aspects of the system, check the content system as an educational goal set by the instructor, and make efforts to supplement the necessary parts. When AI-based problem learning is used in connection with classes, a technical understanding of a system that can utilize various functions of the AI system more efficiently is required. In addition, instructional design is needed to expand thinking and strengthen capabilities through the process of structuring and understanding the contextual relationship between concepts based on the learned knowledge of students using AI-based problem learning systems.

1. Introduction

Edutech is used in various ways to enhance the quantitative and qualitative effectiveness of the teaching and learning process based on the rapid development of the hardware of information devices and the software technology necessary for the operation and management of these devices. In particular, the use of artificial intelligence as edutech, which has recently attracted attention, is being used in various forms depending on the level of technology, such as customized education tailored to the characteristics of learners, interactive systems, learning and inquiry support, student writing analysis, and intelligent agents [1, 2].

These changes also affect the university curriculum, expanding to the use of customized learning and adaptive teaching methods in consideration of students’ learning situations [3, 4, 5]. This is being attempted as an alternative to teaching and learning to meet the needs of the teaching and learning process, such as bridging the learning gap and improving basic academic ability, in the lecture field, where students who take these subjects have various majors and the difference in their prior learning and learning competencies is based on the same subject content.

2. AI-based edutech for adaptive teaching and learning

2.1 Characteristics of AI-based edutech

From Alan Turing’s Turning test in 1950, the Expert system used in Dendral and MYCIN in the 1970s, and deep learning, a type of machine learning in the 2000s, to Generative AI, the rapid development of artificial intelligence, which can be seen at several symbolic stages, is expanding its role to its use as edutech. However, algorithms in which artificial intelligence (AI) learned through deep learning analyzes data on the amount of individual students and learning outcomes to present individual learning paths are one of the important features of AI-based edutech, thanks to the improved performance of parallel computers capable of processing big data at high speed [2, 3]. These algorithms are related to the Rule Base Algorithm, which sets priorities based on rules, organizes, proceeds, and verifies decision trees, and the Item Response Theory, which analyzes data in response to questions to calculate the degree of learning completion of respondents and the difficulty of individual questions [6, 7]. The flowchart in Figure 1 briefly illustrates the principle of problem generation that plays an important role in customized or adaptive teaching and learning based on these algorithms.

Figure 1.

Problem generation process.

First, taking Concept I as an example, if a student presents the correct answer to a given problem (Concept I-1) dealing with Concept I in the learning path, the learning of Concept I-2 proceeds, followed by the learning of Concept I-3 and Concept I-4, and learning related to Concept II can also begin. If the student gives an incorrect answer to the problem (Concept II-1), as shown in the schematic diagram related to Concept II, the algorithm will allow further learning of the content with the accompanying supplementary explanation and then make the student solve the problem of Concept II-1’ similar to Concept II-1. If, as in concept III, a similar problem-solving is retried, the correct answer is not presented immediately, and the wrong answer is repeatedly submitted, the learning proceeds by continuously solving the similar problem until it is recognized as complete learning. As shown in concept III-1 in Figure 1, if learning completion is not achieved even through repeated attempts, the system recognizes the level of learning by the ‘Learning’ step, which is a lower level than ‘Learning Completion’, then the next learning proceeds. In the learning path shown in Figure 1, this problem generation method proceeds up to the main concepts I, II, and III, and learning-related information is collected by checking the number of consecutive correct answers, the number of correct answers and incorrect answers. This is used as direct data for AI algorithms to determine whether individual learners’ understanding of concepts is completed or not. For example, in a system that has set three consecutive correct answers to a single concept as ‘Learning Completed’, if a student submits three consecutive correct answers to a question about Concept I, the system recognizes it as ‘Learning Completed’ and proceeds the learning path with the next concept, concept I-2. Whereas, if the incorrect answer is submitted between the submitted answers even though the correct answers have been submitted three times in this process, the learning path is conducted to solve additional problems, determine ‘complete learning’ using a separate scoring method, and solve the next step, concept I-2. For other topics selected in consideration of teaching and learning objectives, this principle is also basically used to generate problems that reflect priorities in the rule-based algorithm to proceed with learning. These AI problem learning systems were initially provided as independent adaptive problem learning, but recently, their scope of use has been expanded in a way that is used to implement adaptive teaching methods using problem learning in classes [8, 9].

2.2 Learning diagnostics

The evaluation method conducted by the instructor for learning diagnosis and analysis can be classified based on the factors of problem generation, scoring, and scoring result analysis as expressed in Table 1.

	Question creation	Question distribution	Scoring	Result analysis
Paper based test system	Professor	Professor/Assistant	Professor/Assistant	Professor/Assistant
Computer-based test system	Professor	System	System	System
AI-based test system	System	System	System	System

Table 1.

Evaluation for learning diagnosis and analysis.

First, In the case of the traditional paper test conducted using paper prints, a professor directly creates, distributes, scores, and analyzes problems. Although the assistant supports it, the professor is involved and participates in a series of processes related to the evaluation of learning. Next, In the CBT system using screens and keyboards on computers connected to the Internet, a professor develops, creates, and enters a problem into the system through a professor or other inputter. Problem Distribution, scoring, and analysis are a method of receiving help from a computer connected to the Internet of systems. Recently, in conjunction with the comprehensive learning system, it also supports overall matters related to learning, such as learning materials, information, and announcements provided to students. Finally, in an AI-based test system, an AI algorithm is involved in the distribution of the generated problem and management of information about learning in an integrated manner in computer-based test systems. A subsequent learning path through analysis of the automatically scored result is determined and proposed by the AI algorithm.

The AI system identifies topics and concepts that each student needs to learn through initial diagnostic tests, generates related problems to be initially distributed based on the percentage of correct answers, and starts individual learning paths. The student’s problem-solving scoring results in the learning process according to this learning path are stored as data. Basic results such as the completion rate of problem learning for each topic and concept and data such as the time required to solve problems by concept, access time, and end time are stored as diagnostic information about the learning progress of individual students and then are sometimes shared to professors and students. The information is provided to the professor as a sum and average information about all students. Additionally, diagnostic information related to students’ conceptual understanding is provided by providing information such as questions with high incorrect answer rates, questions with many correct answers, and time required in order to analyze questions. These materials are converted into a relative understanding of specific topics among individual learners’ entire subjects and used as basic data to determine the learner’s level of prior learning, helping to determine the content and quantity of instructors’ teaching and learning activities. These learning result analyses and diagnostic data are organized and shared on a dashboard for easy use by instructors. This is also possible when the CBT evaluation system itself or the CBT system is operated in conjunction with the comprehensive learning management system, but the functions are much more diverse and rich in AI-based testing systems.

In particular, it provides both instructors and individual students with comprehensive and integrated information on learning based on individual learning paths. Based on this, question-and-answer required for learning can be made more effective between students and professors, which seems to contribute to enhancing interaction. In addition, in the case of a problem learning system that provides the ability for instructors to view the student’s learning screen the same, it allows one to understand the more specific request for learning.

For example, in the case of ALEKS, which is frequently used for AI-based adaptive problem learning, ‘ALEKS Pie’, which allows you to intuitively check the level and level of learning of each subject, and ‘Progress Report’, which provides information that details the learning situations are provided to professor and individual students. This information can be used to identify the level of learning through relative comparison with other students by providing average information for all students.

Also, since the AI-based problem learning system creates and distributes similar types of different problems on the same subject to students, it helps to manage and supervise large-scale or non-face-to-face test takers and to secure and strengthen fairness.

However, the questions loaded on the AI-based platform mainly deal with multiple-choice question types, and in the case of subjective questions, they mainly deal with short-answer questions that involve writing calculation results using expressions with implied concepts or writing down simple words, so professors are responsible for solving concept-related problems. To check the interpretation and understanding process or to evaluate comprehensive thinking during the problem-solving process, the professor may need to conduct an additional separate evaluation.

2.3 Self-directed learning

In addition to being used for evaluation to diagnose the learning level of students from the instructor’s perspective, AI-based adaptive learning appears to play an important role as self-directed, individualized problem-solving learning from the learner’s perspective. In this context, the paper-and-pencil tests, computer-based tests, and artificial intelligence-based tests as examined in Table 1 are classified according to the scoring and analysis results that can help students learn, and the method of providing supplementary explanations can be shown in Table 2.

	Scoring results announcement	Analysis results announcement	Review
Paper based test system	Professor/Assistant	Professor/Assistant	Correct answer
Computer-based test system	System	System	Solution
AI-based test system	System	System	Solution/supplementary explanation adapted problem

Table 2.

Reflective learning in evaluation.

In the case of the paper test system, students usually solve test questions printed on paper, and the answers written on this questionnaire are scored by the instructor or assistant and the analysis of the scoring results. After that, the solution learning is conducted through the process of additional support such as announcing the correct answer, providing face-to-face or non-face questions and answers to necessary parts, and providing related recorded videos. In the case of CBT, if the problem generated by the instructor is loaded into the system with the correct answer, it is automatically scored, and the scoring and analysis results can be shared and announced. However, it is difficult for students to comprehensively examine information related to learning, and solutions provided in the problem-learning process are mainly presented to the extent that the solutions correspond to simple explanations. Also, Because the information that students can see is limited, in conjunction with the Learning Management System (LMS), through additional data editing steps, the problems mounted by the instructor are scored, the results are shared, and the analysis is provided through additional data editing steps. In this respect, in the case of an AI-based problem learning system, compared to a Computer Adaptive Test, which is known as an evaluation that considers learners’ learning levels among CBTs, AI-based tests provide more sophisticated learning diagnostics due to deep learning of artificial intelligence and provide a customized set of problems that further reflect learners’ learning levels and situations, allowing students to learn self-directed concepts using “solvable” and “to solve.” Additionally, individual students can obtain and use their learning-related information in a relatively diverse and sophisticated manner through the use of dashboards in a similar manner to instructors, and they can judge and use their learning situations and ask for help, such as question-and-answer that fits their learning needs. The supplementary explanation supported in the problem-solving process deals with a relatively detailed and varied amount of content, not just feedback. However, in the case of learning effects obtained from explanations, the degree of understanding and acquisition of knowledge is related to the student’s level of understanding of prior learning, so there seems to be a limit to some extent for students to independently grasp the concept’s concreteness and connection between concepts [10]. Additionally, for concept problems where incorrect answers are submitted, solving similar problems is repeated to provide additional learning.

3. AI-based adaptive learning system

AI-based adaptive problem learning can be used for personalized assignments and regular evaluations within the course operation. To this end, the basic teaching design for the operation of the semester is established based on the characteristics of the subject classification, the characteristics of the subject, the major of the student, and the grade. In addition, after reviewing the content structure of the curriculum to be covered, necessary topics must be selected from the overall content system of the AI-based system and set in the system so that they can be dealt with effectively.

3.1 Content analysis

In order to use the AI learning system as problem learning, it is necessary to grasp the hierarchical structure that deals with the essential knowledge required in the curriculum and the lower level according to the knowledge within the AI system. For example, the hierarchical structure of each piece of knowledge can be dealt with based on the major subject, the sub-subject, and the concept, which is the lower layer of each sub-topic, as shown in Table 3. In this regard, most programs of AI-based problem learning systems are introduced mainly by educational companies, and their programs already create problems dealing with the concepts within the system at the launch stage and provide them in input and mounted form. Therefore, the most common method of problem generation is to select and use the entire topic that is automatically presented and mounted in the system, aimed at complete learning proposed by AI. Also, based on the instructional design of the subject, the professor may select and create specific topics in consideration of the amount of problem-solving learning of students that increases as the number of topics selected increases and the completion rate of learning that may be affected accordingly. In the course of learning, the learning deadline for each topic-related problem may be set to provide a learning environment so that problems related to all concepts can be solved freely throughout the semester, or to provide an environment in which concepts related to the topics covered in the lecture can be learned according to the progress of the semester.

Major subject	Sub subject	Concept	Summary description	Difficulty	Question number
Measurement	Scientific notation	Decimal	use of an exponent to represent a large number	easy	I-1

Table 3.

An example of a hierarchical analysis of knowledge.

This content analysis strengthens the connection with the content of the knowledge system that must be dealt with to achieve the educational goals and objectives of the subject, helping to more effectively utilize the content provided by the AI-based system. The figures related to the learning progress and learning completion rates of the concepts covered in the content hierarchy of Table 3 and the results of the formative evaluation conducted to confirm the educational effectiveness through teaching and learning performance are used to confirm whether the content provided by the system matches the educational goals.

3.2 Adapted learning system implementation

AI diagnoses the level of prior learning for the core concepts to be covered in the course. To this end, an initial diagnostic test as shown in Figure 2 is conducted during the orientation or the first lecture for the semester. As an initial diagnostic test, the AI system distributes questions about the subject’s essential concepts, analyzes them based on the results of scoring the answers submitted by the student, identifies the topics and concepts that each student needs to learn, decides related problems to be initially distributed for subsequent learning, and initiates individual learning paths. Based on this, AI assigns individual problems to individual students to learn sub-concepts related to the diagnosed subject. After that, students are assigned different concepts and a numbers of questions according to the students’ individual learning situations. Students’ learning time varies depending on the stage of prior learning or their understanding of learning. In addition, unlike the initial diagnostic evaluation, additional diagnosis, which is another data that determines the student’s level of completion of learning for each topic, is conducted within the AI system, and the timing and evaluation of this diagnosis may be controlled by the instructor. The method of reflecting the grades for the assignment after the end of one semester can be given in various ways using the learning-related data provided, and in the case of based on the learning completion rate, the grades for the assignment can be calculated by assigning a grade such as A/B/C according to the ratio.

Figure 2.

A schematic diagram of system implementation.

As a result of comparing the learning completion rate according to the deadline-setting method in relation to adaptive problem learning operation, the learning completion rate was higher when the deadline was set to induce learning than when the learning of all topics could be carried out without a limit on the learning period for one semester. In addition, in most cases, students tended to learn problems on related topics after the instructor’s lecture [2].

AI-based adaptive problem-solving learning contributes to diagnosing the understanding of students with different levels of prior learning and strengthening the learning of basic and core concepts in the subject, but the connection and expansion process between each learned concept seems to be lacking. In particular, additional learning can be achieved by utilizing problem-related concept explanations provided during the problem-learning process, but such learning is mainly limited to fragmentary concepts, so students without much prior knowledge may be more vulnerable to structuring and solving problems. Considering this aspect, it seems that customized problem learning should be supplemented with interactions that allow sharing of various perspectives and interpretations with instructors and other students during the learning process so that it is not biased toward answering repetitive problems or acquiring simple concepts.

4. AI-based adaptive class

4.1 Adaptive instruction

Adaptive instruction can be understood as identifying students’ learning situations based on their learning data and providing appropriate teaching and learning [9, 11]. However, adaptive measures for students’ learning situations in the teaching and learning field for a large number of students mainly use group learning by learning level in class, but they do not implement completely personalized class management as shown in the operation of supplementary classes to support learning for certain groups of students with low basic learning ability [12]. However, with the development of edutech, adaptive classes so far have recently been linked to AI-based adaptive systems with classes that analyzed the learning status reached in the students’ conceptual learning process according to personalized learning paths, set appropriate teaching and learning goals that most students can reach, and construct classes that can deal with the use of content and concepts, and are expanding learner-centered countermeasures using teaching methods such as flipped learning and blended learning methods [3, 13, 14].

4.2 Instructional strategy

To apply AI-based problem learning in connection with adaptive instruction, it is necessary to select and set the topics to be covered in the class after further elaborating on the content systems to be used by comparing and reviewing the results of the content system analysis conducted in the previous problem learning implementation stage with the curriculum content system. In other words, rather than simply acquiring specific concepts from the content system mounted on the AI platform, it is necessary to design a content composition that creates an efficient connection relationship and structure of content systems that can effectively use each concept so that the instructor can create a mutually cooperative synergistic effect with the content system that has been previously dealt with.

It also requires a system understanding that can flexibly and selectively adjust the various features of AI-based systems needed to implement these designs. In other words, when using the problem learning system in connection with the class, first select whether to use a video or e-textbook that has already been installed in the system, whether to present and submit assignments, the duration of participation in the assignment, the timing of evaluation, and the place of evaluation. The artificial intelligence-based adaptive learning system itself also provides the use of related videos for learning basic concepts, but the use of videos produced by the participation of instructors helped in terms of class composition applying the instructor’s perspective. In addition, it seems necessary to grasp the system utilization technology in order to more diverse and efficiently utilize the data on students’ learning outcomes provided by the AI system. First, students autonomously learn about the subject through customized problem learning. The level of understanding through such prior learning is confirmed through a diagnostic evaluation of the learning system and the instructor checks through a report to see how much the students have learned the topics necessary for learning before the lecture and conducts the lecture. This diagnostic evaluation data was combined with the evaluation data of other individual learners and used to understand the average learning level and learning orientation of all students and reflected in the composition of the class. It is also possible to check teaching and learning outcomes by using an artificial intelligence-based evaluation system for formative evaluation.

4.3 AI-based adaptive system implementation in class

These class operations are reported to be applied in connection with classes based on the flipped learning teaching method, which is a part of blended learning that has been introduced as a teaching method that can use various edutech as a learner-centered teaching method [3, 15]. An example of the use of this adaptive problem-learning system can be represented in the same way as shown in Figure 3.

Figure 3.

Schematic class utilization of adaptive problem learning system.

The course of the class is as follows. First, students learn beforehand through basic concept videos for each chapter produced by instructors and customized problem learning. In the case of a diagnostic evaluation linked to the progress of the class, a diagnostic report on the pre-learning activities on the day before class is prepared. The class began by identifying problems with high incorrect answer rates with students, and the topics related to these problems were intensively explained in the lecture. The main concepts of the content system provided by the artificial intelligence-based adaptive learning system were identified in the first class, and a summary lecture was conducted in connection with the instructor’s content composition using the topics provided by the courseware. If most students have not yet progressed on a specific topic, the lecture was used in a way that lectures were conducted from the basic contents of the related topic, not just a review. In the second class, information on the progress of students’ personalized tasks is calculated in the form of a weekly report and used to encourage students to learn AI problems. In addition, since it is the second half of the chapter, each topic was viewed from a macro perspective by expanding through the system and causal relationships between the basic concepts learned by students, and presentations and group discussions on the approach in problem-solving learning was also addressed to complement the interaction between peer learners. In addition, the formative evaluation was conducted 15 min before the end of the second class, and it was completed at the end of the class.

The implementation of AI problem learning as a teaching tool linked to lectures at universities will require continuous research on teaching design that presents learning outcomes that experience opportunities for the application and expansion of concepts and foster in-depth analysis, thinking, and reasoning skills by utilizing concepts learned in terms of content. In addition, in terms of methodology, the use of learning diagnosis data will become more diverse and the ability to utilize it will increase compared to operating adaptive problem learning individually. In addition, it is common to use the hierarchical structure of concepts related to the content system to be considered and the understanding of the use of the system for individual problem learning, but its strength and importance seem to be higher.

5. Conclusions

Problem learning using AI-based systems in the teaching and learning process at university seems to be one of the good ways to understand and reinforce concepts in terms of knowledge transfer. However, the goal of basic university education is that conceptual learning should not be limited to simple knowledge understanding and problem-solving. The correlation and contextual connection of each concept should be made, and it should be utilized to play a role related to competency cultivation while expanding in various ways. Therefore, it is thought that the direction of active use of AI-based systems in university education is not to fully implement the various functions and learning completeness provided by the system, but to properly implement the knowledge learned through the system to be effectively linked to various educational activities. Finally, the points to consider for the actual operation of the AI-based system are summarized as follows. First, it can be viewed as content analysis (conceptual structure research) for system utilization. In other words, it is necessary to compare and review the content system established by the instructor, focusing on topics deemed necessary to achieve the educational goals of the subject, in connection with the content system of the system. Next, it seems that there is a need for a system adaptation process that can implement basic functions related to system operation. The costs required for purchasing and using the system should also be considered for long-term use. In the early stages of system operation, it seems necessary to cooperate with edutech experts supported at the university level to utilize the system effectively and efficiently reflecting the effort required from the instructor’s side and the amount of learning required from the learner’s side. In particular, when the amount of problems that must be repeatedly solved to reach the level of learning completeness required by the system is determined, there are only a few areas where the instructor can arbitrarily control the amount of learning the learner can control, so it is necessary to review the system utilization plan in this regard before the start of the semester, which should consider both learning maintenance and learning effects. In addition, since most of the languages currently used in AI-based systems are English, it is necessary to consider them at the teaching and learning design stage to reduce the additional learning burden related to language that may be felt if English is not the native language or does not use English textbooks.