The Role and Function of AIGC in Computer Education

Since OpenAI released ChatGPT on November 30, 2022, AIGC has rapidly attracted widespread attention, followed by explorations and discussions on its applications in various fields, including higher education. On July 10, 2023, the National Internet Information Office and seven other departments jointly announced the “Interim Measures for the Management of Generative Artificial Intelligence Services” (referred to as the “Measures”)^[1]. The Measures encourage the innovative application of AIGC technologies across industries and fields to generate positive, healthy, uplifting, and high-quality content, explore optimized application scenarios, and build an application ecosystem.

The research on applying AIGC in education is very active, involving various aspects such as the evolutionary logic of applications in the education sector^[2], human-intelligence interaction processes and models^[3], academic ethics, and risk prevention^[4]. The research methods are mostly theoretical analyses, while empirical studies are relatively scarce, and research that combines teaching practice to carry out targeted teaching innovations is even rarer.

As an important part of higher education, computer foundational education has developed for over 40 years. The application of AIGC brings unprecedented opportunities to computer foundational education, along with various challenges. In various application explorations, the role positioning and functional extension of AIGC in computer foundational education, especially within specific teaching scenarios, remain unclear. This is mainly reflected in whether AIGC is only used as a new type of tool. If not, then what is AIGC’s role in building the teaching ecology? Furthermore, how does the functional extension of AIGC manifest under different role positions?

If these questions cannot be effectively resolved, not only will the effectiveness of AIGC’s use be affected, but it may even lead to safety and ethical risks. In the context of developing new productive forces and cultivating new talents, it is particularly necessary and urgent to explore AIGC empowerment solutions aimed at new demands and solving new problems that can be used in teaching practice.

Compared to traditional university foundational courses such as higher mathematics and university physics, the computer foundational courses in China started relatively late. Nevertheless, after more than 40 years of development, computer foundational courses have played an important role in promoting student development and have become an essential part of higher education. Literature^[5] divides university computer general education into three phases from the perspective of the evolution of course content: the popularization phase, the foundational education phase, and the general education phase. Following this division, from the perspective of empowering student development, computer foundational education can be roughly divided into the following stages.

(1) Approximately 1984-1995, marked by the establishment of the National Higher Education Computer Foundational Education Research Association, which promoted the inclusion of computer foundational courses in higher education science and engineering curricula. The main task of computer foundational education during this stage was to guide students to recognize and use computers, with BASIC becoming the first programming language for many university students. The characteristic of this stage is the empowerment of new application skills through computer tools.

(2) Approximately 1996-2012, this stage focused on courses such as computer culture basics, computer technology basics, computer application basics, information management technology basics, and computer-aided design technology basics. Operating systems, application software, programming languages, and databases became the main skill tools for learning. This stage can be summarized as the empowerment of literacy development through computer technology.

(3) Approximately 2013-2022, during which computational thinking began to receive attention, and cultivating students’ computational thinking became a consensus in the field of computer foundational education. Compared to the grammar-focused programming teaching, this stage’s computer foundational education is more inclined to guide students to use computational thinking and computational methods to solve practical problems. This stage can be summarized as the empowerment of thinking development through computation.

(4) Approximately 2023 to present, during which computational thinking remains a focus of computer foundational education, and artificial intelligence technology has begun to comprehensively impact computer foundational education. On one hand, artificial intelligence has become part of the computer foundational curriculum; on the other hand, unlike previous tool-based applications, AI technology itself is gradually becoming an important force integrated into the entire teaching process. The emergence of AIGC has undoubtedly significantly promoted this process. The expectation of the industry for computer foundational education at this stage is to comprehensively empower student development through artificial intelligence.

Looking at the above four stages, from guiding students to recognize computers, to teaching students to use computer tools, and then to teaching students to use computational thinking and computational methods to solve problems, the first three stages are interconnected. By the fourth stage, AIGC’s function is no longer limited to being a new type of tool, but rather transitions from the role of a tool in the “human-machine” interaction system to a multi-identity collaborator in the “teacher-student-machine” collaborative system, thus extending its functions.

Therefore, first, AIGC is not just a new type of tool for assisting teaching; it should have a role positioning and functional extension that transcends tool attributes. Secondly, as an important component of the “teacher-student-machine” collaborative system, AIGC needs to be further subdivided in its role, undertaking various roles such as information collector and supplier, thinking assistant constructor, and task collaborative creator, depending on different application scenarios and the problems that need to be solved, and it may undergo role transformation based on interactions with teachers and students as needed. Moreover, the functions realized behind each role are also different, and in addition to tool-based functionalities, the functions attached to scene requirements have also been greatly enriched and extended. In this context, it is particularly necessary to design teaching plans that integrate AIGC with teacher-student collaboration and apply them in teaching practice.

Whether in traditional teaching modes or in AI collaborative teaching modes, the fundamental goal is cultivating people, and the high-level development of students’ thinking at the cognitive level is a concrete manifestation of the effectiveness of education. In traditional teaching modes, the learning process often involves teachers imparting knowledge, and students receiving knowledge, leading to the formation of problem-solving ideas, and then solving problems. In this process, thinking is gradually cultivated and internalized into problem-solving abilities, but the cultivation of thinking does not always proceed smoothly, and often requires multiple iterations of interaction between teachers and students during the stages of encountering problems, raising questions, and solving problems. The teacher’s preaching, teaching, and answering questions accompany the students’ learning growth. With the progress of society, large-scale, personalized, and high-efficiency learning has become new requirements, and traditional educational models that solely rely on teachers’ increased manpower are becoming increasingly difficult to sustain.

Literature[6] proposes a method addressing the poor performance of large language models (LLM) in reasoning tasks such as mathematical problems, symbolic operations, and common-sense reasoning, suggesting that before providing an answer, LLM should generate a series of short sentences to mimic the human reasoning process when answering questions. Experiments show that guiding the thinking chain through prompts can improve LLM’s performance in reasoning tasks.

In fact, LLM’s ability to tackle challenging reasoning problems draws on the human learning process, and the breakthrough progress of AIGC technology, represented by LLM, in human-machine interactive Q&A has made the construction of a “teacher-student-machine” collaborative system feasible. However, scattered and fragmented questioning cannot effectively address the problems students encounter in learning, and improper questioning may even lead to AIGC generating irrelevant or erroneous content. To avoid misleading outcomes, it is still necessary for teachers and students to play a subjective initiative role in the refined design of the interaction process. Therefore, a dual chain iterative teaching design scheme based on AIGC’s thinking and problem chains is proposed, as shown in Figure 1.

First Stage: Teacher-led, iteratively forming a problem-solving thinking chain through interaction with AIGC and mapping it to a problem chain. Beginners often lack a holistic approach to problem-solving, and relatively scattered knowledge fragments are insufficient to construct a complete thinking chain, which is also why most beginners struggle to raise questions. During this stage, teachers need to reconstruct content structure based on the learning path, design the initial thinking chain, and gradually modify and improve the thinking chain through repeated interactive Q&A with AIGC. Based on the thinking chain, a problem chain that aligns with the process of thinking development is formed through Q&A iteration, which is the output of the first stage. During this stage, AIGC plays the role of a teaching assistant and lesson preparation assistant, no longer limited to functions similar to a search engine, but can fully empower teachers’ lesson preparation and teaching design.

Second Stage: Student-centered, constructing their own problem chain based on the initial problem chain provided by the teacher (the output from the previous stage) through interactive Q&A with AIGC, gradually satisfying students’ personalized learning needs. Through iterative Q&A with AIGC, knowledge-based questions are gradually resolved; on the other hand, knowledge content is gradually connected to form a system, which can be mapped to a problem-solving thinking chain. The thinking chain constructed by students is gradually built up through iterative Q&A with AIGC, subtly enhancing students’ thinking abilities. During this stage, AIGC will take on the roles of a tutoring teacher and learning companion, not merely limited to Q&A tool functions, but can fully empower students’ knowledge acquisition and thinking construction process.

In the iterations of the two stages, AIGC is deeply involved in the teaching process and assists in the students’ thinking construction, which is the most significant feature distinguishing it from the previous “human-machine” interaction systems. In practical implementation, teachers need to make more precise types and granularity divisions of the thinking chain and problem chain according to the characteristics of the course and the actual learning situation and iteratively adjust them, thus making how to ask questions a key issue. Table 1 summarizes the types and attributes of questions directed at AIGC. If a precise match of applicable scenarios can be achieved based on learning needs, and questions can be subdivided by type and granularity, then the specificity and efficiency of AIGC collaboration will be effectively enhanced.

3.1 Course and Learning Situation Overview

The teaching practice relied on the elective course “Deep Learning Project Practice” in the general education series of artificial intelligence. As an elective course, there are no restrictions on the major or grade of the students, and there are no prerequisites. However, the characteristics of this course, such as no restrictions on major, grade, or prerequisites, pose significant challenges to teaching, mainly reflected in the following two points.

(1) Programming fundamentals are essential, but most students lack the programming skills required for the course. How can students master the programming techniques needed for the course in a short time and understand the composition and operation of deep learning projects, without getting bogged down in syntax learning or feeling frustrated to the point of giving up?

(2) A systematic understanding is essential, but most students have not formed a holistic view. How can the course help students develop a comprehensive understanding and use computational thinking to construct problem-solving solutions and implement them using deep learning techniques, rather than getting stuck in fragmented technical details without a grasp of the overall picture?

Since the “Deep Learning Project Practice” elective course was launched in the spring semester of 2021, the learning situation has the following characteristics.

(1) Most elective students are from science and engineering backgrounds, but there is a large grade span, resulting in varying foundational skills.

(2) Students have diverse purposes for selecting the course, with most aiming to gain a preliminary understanding of new technologies and empower their professional development, and very few seeking to earn credits.

Therefore, the characteristics of students are that they are unfamiliar with the course content, have narrow perspectives, possess limited or no programming skills, but have good learning motivation; while the course itself features new technologies, a wide range of topics, and some content that is quite advanced, which may induce anxiety, yet the advanced nature of deep learning technology is appealing to students. Relying solely on the teacher’s efforts may struggle to meet the highly differentiated and personalized learning needs of students.

3.2 Python Programming Basics Teaching Unit Design and Practice

Practicing deep learning projects requires a certain level of Python programming foundation. The ability to read and debug code is a basic skill necessary for completing deep learning projects, which is also the primary goal of traditional programming courses. However, more importantly, students need to understand the thought processes and methods for completing deep learning projects using Python, which requires them to be familiar with Python and be able to apply it comprehensively to construct complete projects. Therefore, a 2-hour Python programming basics teaching unit was set up in the course. Since this is not a simple compressed version of a programming design course, and such a short duration is also insufficient to meet the comprehensive requirements of programming teaching, teaching innovation is needed to improve teaching efficiency, with specific designs as follows.

1) Application-oriented content reconstruction.

Based on the requirements of deep learning project practice for Python programming skills, the content of the Python programming basics unit is reconstructed starting from solving problems rather than focusing on the characteristics of the programming language itself (focusing on the question of why to learn). It can be divided into four modules: data structures, object-oriented programming, JSON, and exception handling, each addressing a specific aspect of how Python solves problems in deep learning, as shown in Figure 2.

2) Project-based thinking chain design.

This unit uses a project to scrape specified data from a webpage using Python, which is a relatively simple web scraping program with over 150 lines of Python code, posing a challenge for beginners. In the teaching design, by analyzing the calling relationships between functions, an overall structure oriented towards the project is established for students, extracting relevant knowledge concepts of Python for each function’s intended functionality, allowing students to become familiar with the Python language. Instances are designed for practice, deepening understanding and mastery of knowledge points, and each step can also be fed back to the previous step for iterative improvement, as shown in Figure 3.

3) Problem chain design oriented towards project tasks based on the thinking chain.

Based on the thinking chain, a series of corresponding problems can be designed, as shown in Figure 4, but Figure 4 only shows a coarse-grained problem chain. In actual teaching, it is not necessary to rigidly correspond the thinking chain with the problem chain; teachers still need to play a leading role, demonstrating problem chain Q&A for certain steps while guiding students to establish their own problem chains, thus forming a problem-driven, application-oriented learning process with clear learning objectives. The assistance of AIGC will help continuously stimulate learning motivation.

3.3 Q&A Examples

This article provides a Q&A example to assist students in completing the Python programming basics unit. Considering the learning habits and knowledge loads of beginners, the construction of this problem chain adopts a from-point-to-line task formation approach, rather than a direct task-oriented approach. The problem chain is shown in Table 2.

Teachers can use the problem chain as an example and encourage students to design personalized problem chains based on their learning progress and iterate continuously. Teaching practices in the spring semester of 2024 indicated that 95.5% of students recognized the learning method guided by the problem chain, while 86.8% believed that the use of AIGC with the problem chain improved their learning outcomes.

3.4 Teaching Reflections and Improvements

After teaching practices in the spring semesters of 2023 and 2024, the dual chain iterative teaching design has gradually improved, but it also has some issues.

(1) The quality of the problem chains constructed by students based on their learning progress still needs to be enhanced, as the high degree of fragmentation in the questions posed leads to suboptimal AIGC assistance.

(2) Illusion issues may cause AIGC to generate erroneous content, and without identification and review, the negative effects of misleading information will be difficult to avoid.

In the future, more teaching cases based on dual chain iterative design will be further developed, problem chains will be refined, multiple iterations will be conducted to reduce the occurrence of illusions, and parallel instances will be designed to promote learning transfer and help teachers and students achieve high-quality teaching.

Learning to ask questions is the source of motivation for learning. In traditional modes, students’ questions often require teachers to answer them. Coupled with changes in students’ preferred communication modes, relying solely on student-initiated questions and teacher responses can no longer meet the needs of large-scale personalized learning. Integrating AIGC into the teaching process also supports high-quality and large-scale personalized learning. AIGC is not omnipotent; teaching still requires teachers as the main body to organize and promote it, while learning still requires students to complete it autonomously. AIGC serves more as an assistant teacher, lesson preparation assistant, personal tutor, and learning companion, among other roles. Therefore, AIGC is not intended to replace the thinking of teachers and students; reasonable design of teaching plans and effective application of AIGC provide new opportunities for high-quality education and offer new pathways to further enrich the empowerment of computer foundational education through artificial intelligence.

References:

First Author Profile: Wang Ruobin, Male, Northern Industrial University, Professor, Research Direction: Intelligent Computing, Computer Education, [email protected].

Citation Format: Wang Ruobin, Li Meihui, Song Wei, et al. The Role and Function of AIGC in Computer Education: A Dual Chain Iterative Teaching Design and Practice[J]. Computer Education, 2024(10):159-163, 168.