Revisiting The Dialectical Relationship Between AI And Education

This article is authored by: Jia Jiyou, Professor, Director of the International Research Center for Educational Informatization, Peking University.

Below is a summary of Professor Jia Jiyou’s report content

Currently, the field of artificial intelligence is experiencing a new development, with large language models represented by GPT deeply integrated into various aspects of daily life. These large language models are based on vast corpora and powerful deep learning networks, possessing functions such as language learning, understanding, and output. Typical models include ChatGPT, iFlytek Spark, Wenxin Yiyan, Zhipu Qingyan, Tongyi Qianwen, and Kimi. This report discusses three issues: the latest developments in artificial intelligence and its educational applications, challenges posed by artificial intelligence to education, and educational response strategies.

Artificial intelligence refers to the intelligence implemented on machines (including computers) using artificial methods, or in other words, it is the simulation of human and other biological intelligence by machines, including perception, memory, thinking, behavior, and language. The main guiding theories of artificial intelligence are three: symbolism (logical reasoning through symbolic logic operations), connectionism (mimicking the parallel computation of human brain neurons to identify and learn certain patterns or behaviors), and behaviorism (reinforcing or weakening certain behaviors through rewards or punishments). The main research areas of artificial intelligence include eight: knowledge engineering, machine learning, pattern learning, natural language processing, intelligent robotics, expert systems, artistic creation, and automated program design.

The main task of knowledge engineering is to store and represent common sense and specialized knowledge in computers, quickly search for certain knowledge based on user needs, and reason to solve new problems based on existing and new knowledge. Representative applications of knowledge engineering include educational resource platforms across the country and various provinces and cities (National Smart Education Platform, Sunshine GaoKao), various types of encyclopedias on the internet (Wikipedia, Baidu Encyclopedia), dictionaries (Wordnet, HowNet), and search engines.

The goal of data mining is to study how computers extract information based on data and induce knowledge based on information. Specific work includes statistical description, classification, clustering, rule association, prediction, combinatorial optimization, and visualization. Under the guidance of cognitive, behaviorist, and connectionist theories, various algorithms have emerged in the field of data mining, such as production rule reasoning, decision trees, regression analysis, artificial neural networks, genetic algorithms, Bayesian algorithms, nearest neighbor algorithms, support vector machines, fuzzy logic, rough sets, etc. Common commercial data mining software includes Excel, Access, SPSS; free open-source data mining software includes Weka, R language, MYSQL, Deeplearning4j, DMTK, OPENN, TensorFlow, etc. There are many practical application cases of data mining, such as the online learning activity analysis project by Professor Jia’s team, which constructs a student online activity index model (OLAI) from three dimensions: speed, quality, and quantity, clustering students’ online learning activities from strong to weak into seven categories, and conducting more precise teaching design based on students’ online learning activity index status.

The main task of pattern recognition is to identify faces, eye irises, voices, fingerprints, gaits, printed and handwritten characters, emotions, etc., using the research results of data mining. Currently, artificial intelligence has surpassed humans in accuracy and speed in tasks such as facial and image recognition.

Professor Jia Jiyou cited three educational application cases of pattern recognition. The first case is the online live learning companion system and testing system at Peking University during the pandemic, which can monitor the examination process of students in real-time through cameras to identify whether the student taking the exam is indeed the student; the second case is the automatic evaluation system for physical education class actions at Peking University, which can automatically identify students’ body movements and provide scoring and feedback, offering personalized posture evaluation and guidance; the third case is the behavior analysis system for teachers and students in excellent classes, which uses artificial intelligence to automatically recognize and manually calibrate to label teacher-student behaviors, summarize verbal behaviors for statistical analysis, and produce visualizations such as pie charts, time sequence graphs, and transition matrices of classroom video verbal behaviors, along with textual explanations, summarizing the characteristics of classroom videos for certain subjects or categories.

Natural language processing utilizes the research results of knowledge engineering and pattern recognition to understand human speech and text, generate human text and speech, and complete tasks such as speech recognition, speaker recognition, speech synthesis, text understanding, text generation, translation, and chat dialogue. Large language models can be used for document generation, PPT design, learning resource retrieval, long text interpretation, question design, answer evaluation, and essay correction. They can also create intelligent agents to complete specific tasks. Many language learning systems have now entered language learning classrooms to help students learn pronunciation, listening, and reading.

Intelligent robotics have been widely applied in various fields, such as the use of robotic dogs to transport equipment at the 2023 Asian Games and the use of robots for lunar sampling in the Chang’e 6 mission in 2024; many primary and secondary schools have also begun to use robots to conduct maker education, cultivating students’ interests and enhancing programming skills through the combination of software and hardware.

Expert systems use the research results from the aforementioned multiple fields to solve problems that only human experts in certain special fields can address, such as in medicine, education, and military fields.

In the field of education, expert systems include intelligent teaching systems, also known as intelligent agents, intelligent tutor systems, intelligent assistant systems, etc. Intelligent teaching systems simulate excellent teachers or teaching assistants and interact with students to diagnose their learning characteristics, then provide personalized tutoring under certain teaching theory guidance, helping students learn a particular course, subject area, or knowledge point. For example, engaging in Socratic dialogue with students or cognitive scaffolding. The positive impact of intelligent teaching systems on subject teaching has been fully validated in numerous empirical research papers and meta-analyses based on these empirical studies.

Professor Jia Jiyou also introduced the self-developed XisaiKe system for English teaching and the MIATS system for mathematics teaching. The XisaiKe system features multi-role dialogue, listening practice, grammar exercises, reading comprehension, and English reading, supporting communication between teachers and students as well as peer-to-peer interaction through discussion areas, significantly improving middle school students’ English learning outcomes, enhancing learning interest, and reducing teachers’ teaching burdens while promoting professional growth. The mathematics teaching system MIATS V2.0 can provide adaptive testing based on item response theory and extensive data analysis, offer intelligent tutoring based on the theory of the zone of proximal development and Polya’s “How to Solve It” framework, and provide learning and testing companionship through multiple channels.

The artistic creation function has already been applied in various fields, such as the snowflake effects that followed the movements during the opening ceremony of the Winter Olympics, and poster design for e-commerce websites.

Automated program design involves allowing computers to design new programs based on human requirements. Language model systems like iFlytek and Kimi already possess strong program design and modification capabilities, enabling them to design and write simple computer programs.

The ability of artificial intelligence to replace humans has evolved from physical labor to mental labor. As generative artificial intelligence technology matures, users can use AI to quickly complete multilingual text translation, various types of paper writing, graphic creation, program writing and modification, statistical analysis, etc., for a small fee, significantly reducing the economic costs compared to hiring professionals. Previously, only graduates with higher education and relevant professional training could perform mental labor. Now, these graduates face employment threats from the development of artificial intelligence technology, which will further impact the enrollment, teaching, and management of related majors, exacerbating the increasingly severe employment issues for graduates in the era of mass higher education.

Closed evaluation refers to tasks assigned by teachers that students complete under the organization and supervision of teachers within a certain time frame, including phase exams, graduation, and entrance exams that are highly related to students’ interests. Closed evaluations assess students’ memory, thinking, and language expression abilities within a specified time. AI systems may perform better than ordinary students in certain subjects, making them more likely to be selected for jobs or further education. According to tests conducted by Microsoft researchers, GPT scored higher than the average human examinee in objective questions of China’s college entrance examination in English, Chinese, geography, history, biology, and American English, and scored higher than the highest human score in English, while still lagging behind humans in mathematics. This indicates that AI has a clear advantage in mastering memorized content, although it is still relatively weak in logical thinking abilities.

The emergence and maturity of generative artificial intelligence technology allow students to quickly find answers and solutions to specific problems using these technologies, generating a paper, program, or artwork based on themes or keywords, which can lead to laziness in students’ thinking and hinder their cognitive development.

The emergence of any new technology will lead to the disappearance of some jobs and the creation of new ones. For example, large models have already given rise to the new position of prompt engineer. Artificial intelligence technology itself requires high-quality talents to continuously develop, not only computer programmers but also researchers from other disciplines to contribute wisdom and strength from different perspectives. The “Adjustment and Optimization Reform Plan for the Setting of Disciplines and Majors in General Higher Education” issued by five departments, including the Ministry of Education in March 2023, pointed out the need to actively adapt to national and regional economic and social development, knowledge innovation, technological progress, and industrial upgrading, and to optimize, adjust, upgrade, replace, and establish disciplines and majors.

Therefore, it is necessary to plan comprehensively, layout thoroughly, and develop evenly, ensuring that while promoting the development of artificial intelligence, all educational systems can produce outputs, allowing graduates of various forms to find suitable jobs.

AI-supported adaptive assessments and open assessments can provide personalized assignments and tests for each student while ensuring fairness and quality in evaluations, representing a direction for reforming teaching assessments. In terms of evaluation content, there should be a reduction in purely memorized knowledge and a focus on assessing students’ abilities to flexibly apply existing knowledge to solve problems, critical thinking, and creative thinking. This reform direction aligns with the requirements of the “Overall Plan for Deepening the Reform of Educational Evaluation in the New Era” issued by the Central Committee in October 2020, which emphasizes changing relatively rigid test formats, enhancing the openness of test items, and reducing rote memorization and mechanical practice of exercises.

Educational management departments should formulate policies and guidelines, leveraging central and local finances to encourage universities, research institutions, and high-tech enterprises to develop intelligent teaching systems and other AI systems across various disciplines, integrating these system resources on public platforms like the National Education Resource Platform, and deploying these systems in educational institutions at all levels.

Measures such as pre-service education for teacher trainees and in-service teacher training should be implemented to help teachers understand the functions and usage of AI systems.

The history of human development is one of continuously increasing the use of auxiliary tools. In the face of developing artificial intelligence technology, schools should guide students to use it according to educational stages: preschool and primary school students should attempt to use it under the guidance of parents and teachers, rather than using it independently; secondary school students may use it appropriately; and university students should use it extensively and deeply.

Finally, Professor Jia Jiyou emphasized that artificial intelligence has achieved remarkable accomplishments, and related technologies such as data mining, knowledge engineering, natural language processing, pattern recognition, and emotional computing are maturing, which can help educators design intelligent educational systems across various subjects to meet students’ personalized needs and promote their overall development. However, education must always emphasize a human-centered approach, where technology serves as an auxiliary means, and no technology should adversely affect students. Wishing all teachers and students to make good use of intelligent systems to empower educational development.