This Article Overview
8000
Words
Reading Time
15
Minutes
01
What Are the Differences Between Generative AI and Past Information Technologies?
In 2023, generative artificial intelligence (GenAI) struck like a thunderclap, changing my understanding of computers. The first thing that amazed me was AI painting; I only needed to describe the scene I imagined in detail with prompts, and instantly exquisite images would be generated. Even more astonishing was opening the Taobao app on my phone, typing “Taobao Q&A” in the search box, and saying “Design a course plan for XX.” In an instant, a complete teaching plan appeared before my eyes! Like the magic bottle in “One Thousand and One Nights,” it has overturned years of prejudice against computers. Generative AI can not only paint but also compose music, write poetry, create couplets, write articles, make PPTs, code, produce videos, and even chat with humans, becoming an intelligent assistant for every teacher and a mentor and study partner for every student. This is a significant innovative breakthrough in the field of computer applications that I have seen in decades. Microsoft CEO Nadella said, “For knowledge workers, this is equivalent to the Industrial Revolution.” Generative AI will fundamentally change the way you use computers. Currently, based on the information publicly disclosed by OpenAI, generative AI is related to at least five key technologies and architectures: transformer models, basic architectures based on transformer models, reinforcement learning from human feedback (RLHF) technology, instruction tuning technology, and chain of thought technology, resulting in four prominent core capabilities: generative content creation ability, dialogue context understanding ability, sequential task execution ability, and programming language parsing ability. What impressed me is that generative AI cleverly combines human wisdom with the advantages of computers; the RLHF technology iterates through human interaction, ultimately generating content that meets or even exceeds human expectations, leaving human users amazed. On November 6, 2023, OpenAI announced the enhanced version of GPT-4 and the GPT Store version, allowing any non-programmer user to create their own generative AI agent through voice dialogue, once again shocking the world. One can imagine that GPT-5.0 and GPT-6.0 are already in line behind it. What will the world look like in 5 or 10 years?! Faced with this sudden technological change, everyone is unsure how to respond. Suddenly, various salons, summits, articles, reports, and speeches about generative AI surged. After the audience’s excitement like a grand festival, they realized that we cannot use generative AI in teaching. Currently, the National Internet Information Office, the Ministry of Education, and seven other departments have jointly issued the “Interim Measures for the Management of Generative AI Services,” pointing out the development direction of generative AI: “The country insists on balancing development and security, promoting innovation and legal governance. Effective measures should be taken to encourage the innovative development of generative AI. Providing and using generative AI services must comply with laws and regulations, and respect social ethics and morals.” With the guidance of national policies, a large number of Chinese generative AI technologies have rapidly developed and officially provided services to the public, providing technical support for every teacher to use generative AI, making it possible for schools to apply generative AI in teaching, officially opening the door for Chinese education to enter the era of generative AI. 02
Generative AI and the Transformation of Learning Methods: Generative Inquiry Learning
Every teacher being able to open and use generative AI software is the first step in applying generative AI in teaching. I suggest readers try it out themselves by logging in and registering to use iFLYTEK Spark, Baidu Wenxin Yiyan, and other generative AI large models approved for public use by the National Internet Information Office. You can send your ideas to the generative AI using prompts, and the AI will respond instantly. For example, prompt 1: Please design a unit teaching plan for the elementary school math class on “Fraction Addition.” (Readers can replace it with their specific grade, subject, and course name). Prompt 2: Design a unit teaching plan for the middle school biology class on “Heredity and Variation,” following Bloom’s taxonomy of educational objectives’ six levels (remembering, understanding, applying, analyzing, synthesizing, creating), including teaching objectives, syllabus, teaching activities, and providing detailed teaching strategy suggestions. Prompt 3: Please draft a keynote report for an international conference titled XXXX. Readers will find that regardless of the educational issues you are concerned about, generative AI can give you unexpected responses. The more specific and precise your prompt design is, the more specific and precise its response will be. This kind of human-computer dialogue-style communication can broaden your original thinking, but it may also be inaccurate, even “seriously talking nonsense,” with a high degree of uncertainty, requiring you to think and judge independently. This learning process based on generative AI is very similar to playing chess: you as one side of the chessboard give AI prompts, just like you place a piece on the board, and the AI does not know how you will ask; then, it will calculate based on your prompts and respond to you; for you, how the AI responds is also unpredictable and uncertain, and then you can ask follow-up questions based on the opponent’s response and give the opponent your next piece; this is also unpredictable for the AI; it will calculate based on your current piece and respond again… This iterative cycle of interaction full of uncertainty and depth can continuously stimulate the imagination and interest of both parties. This is why chess has maintained its charm throughout human history, and also why generative AI differs from past teaching methods and applied information technologies. When generative AI is applied to the field of education and teaching, regardless of which learning theory and teaching needs you research and practice, it will ultimately come down to the biggest difference between generative AI and other technologies: Artificial Intelligence Generated Content (AIGC). Due to the sudden emergence of generative AI, there are currently no discussions about the application of generative AI in teaching in classic works of pedagogy, teaching theory, curriculum theory, educational psychology, learning science, etc. In various teacher training activities across regions, to explain the new developments of generative AI in teaching to frontline teachers, we call this new type of learning method Generative Inquiry Learning. What is Generative Inquiry Learning? Generative Inquiry Learning refers to a teaching method in which teachers and students strictly adhere to national regulations, ethical standards, and information security requirements while reasonably applying generative AI to assist teaching, and through learners’ independent exploration, critical thinking, and creative thinking, promote optimized learning. Generative Inquiry Learning is a student-centered teaching method under the environment of generative AI, emphasizing student autonomous exploration, practical operation, and building knowledge systems through human-machine interactive dialogue, requiring students to use generative AI to assist teaching under the guidance, support, and control of teachers, aiming to cultivate students’ critical thinking, problem-solving ability, teamwork spirit, and innovative thinking among other comprehensive qualities. This is a way of learning that combines human wisdom with machine intelligence. The English term for Generative Inquiry Learning is Generative Quest Learning, abbreviated as GenQuest or GQL. Generative Inquiry Learning is an inevitable trend in the development of educational informatization and the digital transformation of education (as shown in Table 1). 
In 1946, the University of Pennsylvania successfully developed the world’s first electronic tube computer, ENIAC, which was large and could not be used for teaching. In 1958, IBM launched the personal computer and designed the first teaching program to teach binary arithmetic using computers, initiating program teaching experiments in computer-assisted instruction. In the 1990s, multimedia computers entered schools, where teachers demonstrated multimedia courseware in collective teaching, and multimedia teaching emerged. Since entering the 21st century, the internet has entered school classrooms, where teachers first thought of letting students search for information online, then engage in group discussions, and hands-on create learning research reports. A network inquiry learning method called WebQuest first emerged in the United States and quickly spread globally. The main learning steps of WebQuest are divided into six stages: introduction, task, process, evaluation, conclusion, and credits. In the later development of WebQuest, frontline teachers categorized WebQuest into short-term and long-term formats based on their school’s teaching arrangements: short-term WebQuest learning activities are completed in 1-3 classes, aiming to help students master current knowledge points; long-term WebQuest learning activities generally take a week, a month, or more, helping students deeply explore knowledge content around a thematic project and gain a more comprehensive and profound understanding of the thematic content. Education researchers at home and abroad have found that many students merely search for information online and then copy and paste it into their learning reports, which constitutes low-level, shallow learning; thus, they proposed high-order thinking ability cultivation based on Bloom’s taxonomy to promote students’ understanding and knowledge transfer through “deep learning,” including teachers designing basic questions, unit questions, and content questions in a problem-oriented teaching approach, project-based learning, interdisciplinary unit teaching, problem-solving, and core competency education, enriching and expanding the WebQuest learning method in the internet age. In 2023, schools and teachers at the forefront of educational digital transformation experiments across regions attempted to use generative AI to assist teaching, creating a new teaching method: Generative Inquiry Learning. This is an upgrade and new development of the WebQuest learning method in the generative AI era, further expanding students’ activities of collecting learning resources online into interactive dialogues using general language models (LLMs) and combining them with students’ critical thinking to promote the development of students’ higher-order cognitive structures, innovatively upgrading the learning method represented by WebQuest in the internet age. The Generative Inquiry Learning model includes the following six stages, as shown in Figure 1, where the most attractive and challenging is the dialogue exploration stage.
First, stimulate interest. Teachers can ignite students’ learning interests in various ways, such as creating problem situations to guide students’ learning interests. The setting of problem situations should be closely related to real life, course content, and learning tasks. If the problem is an open-ended question based on real-life scenarios, it can further stimulate students’ initiative to explore. Second, task assignment. Teachers clearly present the tasks for generative inquiry learning based on curriculum standards and teaching objectives, guiding and controlling students’ learning processes in a task-driven manner. Third, dialogue exploration. Under the guidance of teachers, students aim at the learning tasks to be completed and interact with generative AI in dialogue, recording their critical thinking in chain-style dialogues and exploring in various ways. For example, five basic searches (Baidu, Bing, WeChat article search, Bilibili video search, China National Knowledge Infrastructure journal paper search, library book retrieval), peer communication, consulting teachers, parents, and field experts, conducting experiments, and natural and social investigations, etc. They record their learning experiences and fill out learning sheets. Learning sheets are scaffolds prepared by teachers before class to guide students’ learning processes, specifying the specific steps and process records for students to conduct generative inquiry learning under the stipulated teaching objectives, which can be used for process evaluation. Fourth, transfer practice. Students independently think and solve problems in new situational problems, completing assignments and exercises assigned by teachers. Fifth, conclusion sharing. Students summarize their learning gains, integrating their experiential learning into rational cognitive structures, and share their learning gains and research conclusions with group members. Sixth, evaluation feedback. Teachers provide learning evaluations and suggestions for further development. Unlike various teaching methods familiar to everyone in the past, Generative Inquiry Learning allows learners to interactively learn with AI in chain-style dialogues. The thought chain is a strategy used by artificial intelligence to handle complex tasks; this technique breaks down complex large tasks into smaller tasks containing multiple intermediate steps through a series of sequentially related instructions, with each small task guided by relatively simple instructions to assist the model in generating and solving complex logical reasoning tasks. In teaching, the thought chain dialogue between teachers, students, and generative AI means that users ask a series of continuous questions and iterative inquiries, progressing from surface-level to deeper-level questions, like squeezing toothpaste or peeling an onion, gradually forming a human-computer dialogue method for solving complex problems. This game-like thinking training allows teachers and students to gain a learning experience completely different from past computer-assisted teaching and the use of digital educational resource platforms. For example, in generative inquiry learning activities, teachers and students cultivate independent thinking, critical thinking, decision-making thinking, challenging thinking, human-machine collaborative thinking, systemic thinking that considers the big picture, resilience qualities of “not being greedy for victory,” rigorous reasoning thinking, unconventional thinking, innovative thinking, and thinking that observes the development of things in terms of time and space, as well as deep learning thinking of “calculating more wins than less” and philosophical thinking of “playing chess is like life,” gaining insights that other teaching methods cannot provide. An important technological innovation supporting Generative Inquiry Learning is the upgraded development of the course management system, which has matured in the mobile internet era, into the “Generative Inquiry Learning Course Management System,” integrating suitable generative AI large models for educational use with course teaching management to support teachers’ lesson preparation and student learning management. Frontline teachers have reported that the application of generative AI in classroom teaching faces two major challenges: One is that AI large models can “seriously talk nonsense,” making them unusable in teaching; the second is that according to the Ministry of Education’s “Five Management” and UNESCO’s “Guidelines for the Use of Generative AI in Education and Research,” minors under 13 cannot use generative AI independently and must do so under adult supervision, making it impossible for all students in the classroom to register and use generative AI with their mobile phones. Why do all large models have hard problems with information accuracy? Large Language Models (LLMs) are AI models based on deep learning, trained on massive data and large-scale computing resources, capable of understanding and generating natural language. One of the famous representatives of large language models is OpenAI’s Generative Pre-trained Transformer 3 (GPT-3), which has billions of parameters, allowing it to perform exceptionally well in natural language processing tasks. However, due to the general-purpose corpus used for training large models, which aims to handle general content generation, it inherently has deficiencies: training and maintaining large language models is a resource-intensive industry, leading to high costs and low efficiency; large language models can introduce biases from their pre-training data into the generated text, resulting in inaccurate information, which cannot fully meet the strict regulatory requirements in the educational field; the data collected and trained by different large model companies come from various social and cultural backgrounds, inherently containing information biases and unpredictable information security issues. The accuracy problem of large models is a double-edged sword; their “largeness” prevents them from deeply penetrating specialized academic fields. Just like the principle of pressure in physics: if the same computational power and funds are used to train a large model, the larger the area, the shallower the pressure; the smaller the area (like the tip of a needle), the deeper it can penetrate. If we think in reverse, developing small models focused on specialized subjects and specific teaching scenarios is a new way to solve the first problem. Small Language Models (SLMs) are the future direction for applying generative AI in the education sector, with more advantages in application development compared to large models: small language models are highly customizable, can be tailored to user needs, and focus on specific goals. For example, they can be customized for a subject, a textbook, or even specific teaching application scenarios. They have high efficiency and low cost; since small language models are smaller in scale, they can be trained with less data, operate more efficiently, and run on less powerful hardware, saving costs and being more practical, yielding higher returns. They also have high accuracy; small language models can effectively control the quality and completeness of training data through targeted training on specific professional goal datasets, providing high-quality accurate results more reliably, which is especially important for education. They offer high security; compared to large language models, the smaller codebase and fewer parameters of small models maximize security by minimizing vulnerabilities, giving them an advantage in safety. They also provide high transparency and controllability; small models have a more transparent and interpretable operation, allowing school users to ensure compliance with security protocols and regulatory requirements. Small models can process data locally or in controlled environments, helping protect sensitive information and data privacy, thus preventing risks of data leakage or unauthorized access, which is crucial for educational regulatory compliance and the local deployment of small models in sensitive institutions. It should also be noted that the current level of technology in semantic recognition and content generation in artificial intelligence cannot achieve 100% accuracy and control, but we cannot abandon the effort due to this limitation. We can improve the digital literacy of teachers and students comprehensively, accelerate the improvement of large models’ content generation control capabilities, and explore the collaborative development of multiple elements such as technology, products, regulations, and teacher training to achieve a balance between development and safety, ensuring the content safety of generative AI and promoting the healthy development of its educational applications. Teachers suggest two methods to allow all students in the class to access generative AI: The first method is for students to form groups, and the class teacher registers a large model account with their mobile phone, allowing the whole class to use a few teachers’ accounts, which helps teachers guide and control all students’ online activities; the second method is for large model providers to directly offer generative AI services to local education departments, allowing school teachers and students to access generative AI large models through the intelligent education systems developed by the companies. The development of Generative Inquiry Learning presents new expectations for the construction of learning resources. Since teaching and learning are personalized activities of teacher-student interaction, current educational resources cannot provide targeted teaching services based on students’ different situations, and this “last mile” problem has troubled everyone for a long time. Now, it is necessary to encourage and support large model companies to collaborate with education departments, and the national and provincial smart education public service platforms should provide teachers with educational-specific small model editing training tools and a co-creation and sharing ecological platform, allowing every teacher to fully utilize the massive educational resources on the resource platform, edit and process them into data resources suitable for training small models in their subject areas through human thought and judgment. Just as endless content resources emerge daily on new media, relying on the collective power of large model companies and millions of users, fixed educational resources will transform into millions of small model assistant resources, revolutionizing the construction of intelligent learning resources in China and achieving a revolutionary change in the transformation of learning resources. 03
The Impact of Generative AI on Teaching
Mr. Li Bingde summarized teaching activities into seven elements, as shown in Figure 2. What changes will occur if each element of teaching combines with generative AI? 
Every teacher has an intelligent teaching assistant. Teachers master the skills of prompt design, issuing appropriate prompts to AI, which can help teachers do many things, just like a wise and capable teaching assistant. For example, grading homework, writing lesson plans, designing assignments, creating PPT presentations, writing research articles, designing micro-videos, drafting various work documents, coding, and designing and managing experiments, thus improving teachers’ work efficiency and reducing their burden while enhancing quality. In the near future, any teacher can have a personal assistant powered by artificial intelligence, far exceeding today’s information technology. Every student has an intelligent mentor and study partner. Students using generative AI can ask knowledge-related questions one-on-one and receive help; they can receive suggestions for writing ideas, polish their essays, enhance their imagination in writing; provide varied practice assignments, enrich problem-solving ideas; offer project-based learning suggestions; provide examples of programming code; serve as partners and mentors for English conversational learning; participate in large unit and interdisciplinary learning activities, with AI becoming a smart partner in learning, answering students’ questions, providing learning reference resources, and promoting personalized teaching and tailored education. Recently, I saw reports from schools in remote areas of Guizhou and Sichuan, where teachers organized students to converse with iFLYTEK Spark, allowing children in the mountains to see a different world. Generative AI brings personalized mentors and study partners to students, genuinely helping to narrow educational imbalances and urban-rural gaps. Every principal has an intelligent assistant. Generative AI helps principals design various educational management documents; provides reference strategies for educational governance; searches for educational resource websites; builds campus culture; assists in writing educational research papers; offers suggestions for specific work at the school; provides reference plans for school environment design; and even assists schools in drafting management measures for the application of generative AI in education. Various assistants in schools can share both large and small model data, connect with each other, forming a group of educational assistants that can provide intelligent services around the clock. Every textbook is paired with a generative AI small model assistant. Reshaping educational publishing with generative AI is a new track for the digital transformation of various textbook publishers. Based on previous multimedia textbooks and cloud-based digital textbooks, generative AI small models targeting different textbook versions can be developed and trained based on the printed textbook version, combined with digital human technology, metaverse virtual learning space technology, speech synthesis technology, etc., to provide intelligent learning assistants for students. This requires publishers to collaborate with large model developers to provide “small model editing training systems” for textbook publishing editors, assisting users in customizing scenarios; utilizing proprietary data to train small models, allowing users to optimize/evaluate prompts, relying on large models to provide algorithms and computing power, enhancing the efficiency of users training their own small models. Similar to how Microsoft Office has become a foundational productivity tool for knowledge workers worldwide, the small model editing training system will become a productivity tool for editors engaged in textbook publishing to train textbook small models, transforming publishers from traditional print book publishing institutions into new publishing organizations that produce multimedia textbooks and intelligent learning assistants. This will be a disruptive revolution in textbook construction. The curriculum system is evolving towards generative courses. In classic domestic works on curriculum teaching theory, it is mentioned that ancient Chinese literature already had the concept of “curriculum.” A curriculum is the totality of educational content that learners gain through interaction with the educational context under the planned and organized guidance of educators. The curriculum encompasses the content and processes of teaching, extracurricular learning, and self-study activities, serving as the overall plan and process of teaching and various learning activities of students. It is evident that a common characteristic of curricula is that the learning content is pre-designed, planned, and compiled by people. Now, the sudden emergence of generative AI has broken the convention of strictly planned curricula for hundreds of years, creating a new curriculum form through generation, which we call generative courses. The generative course in the AI era refers to the dynamic generation of new learning content, including text, images, audio, video, code, etc., through chain-style dialogues between teachers and students with generative AI, changing the way teaching and learning activities occur and reshaping curriculum resources and teaching structures. The learning content and curriculum resources of generative courses are generated during the process of human-machine interactive dialogue, which cannot be predicted by the pre-planned curriculum compilation and print textbook publishing process. This intelligent generation method subverts people’s cognitive models of traditional curricula and textbooks. Table 2 compares the differences between traditional curricula and generative courses. 
Generative AI integrates into every aspect of the curriculum, and the course learning content resources generated through AI and teacher-student interactive dialogue possess characteristics of generativity, uncertainty, and richness, allowing teachers to associate educational theories, teaching strategies, and subject-specific training datasets with the design of courses, the generation of curriculum resources, and the management of teaching activities, enhancing AI’s understanding of teaching content and adaptive learning resource generation capabilities, forming a complementary relationship with traditional curriculum resources. This is a new track for future curriculum construction. 04
Suggestions for the Application of Generative AI in Teaching
(1) Top-level Design Every school should formulate management measures for the application of generative AI in education. This is the prerequisite that schools must first consider in light of the rapid development of generative AI. Due to the powerful content generation capabilities of generative AI, accompanied by unpredictable hidden dangers and risks, UNESCO’s “Guidelines for the Use of Generative AI in Education and Research” repeatedly emphasize that researchers, teachers, and learners need to be aware that generative AI does not understand the text it generates, and it can and often does produce incorrect statements. A critical approach must be taken towards everything it produces, strengthening the regulation of generative AI applications in education. Governments, educational institutions, generative AI providers, school administrators, and teachers should carefully assess and regulate the potential risks of artificial intelligence, formulating basic principles, procedures, measures, and regulations for the use of generative AI in education and teaching, ensuring information security, assessing, and strictly controlling the potential impacts of AI-generated content on the development of human abilities such as critical thinking and creativity, and implementing specific provisions related to the ethics of artificial intelligence. Readers can try to let AI assistants help draft school management measures for the application of generative AI in education. Use the prompt, “You are an AI education expert, please formulate management measures for XXXX school regarding the application of generative AI in education, drafting eight management measures.” You will see that the draft AI provides is indeed worth referencing. (2) Teacher Training Improve the generative AI literacy of all teachers, including basic skills for using generative AI, prompt design skills, management capabilities for using generative AI in teaching, changes in teaching design, development of assignments and evaluations, and even how teachers speak in class and changes in classroom teaching language, all need to be relearned. Readers can try asking generative AI to demonstrate how teachers should guide students in using generative AI in classroom language. Prompt 1: If students use generative AI to assist learning in class, what changes will occur in teachers’ language during teaching activities? Please provide ten examples of classroom language for teachers. Prompt 2: When teachers let students interact with generative AI during learning activities, how should students ask questions to AI? Please provide twenty examples of prompts students can use when asking generative AI questions. Prompt 3: How to improve the skills of teachers and students in using generative AI in teaching? Please provide ten suggestions. Follow-up: How to avoid the negative risks of students using generative AI during learning? Please provide ten suggestions for teachers. (3) Teaching Research I suggest readers experience having generative AI as your educational research assistant by providing prompts to iFLYTEK Spark and Wenxin Yiyan: If you are a high school Chinese teacher, how to apply generative AI to assist teaching in the thematic teaching activities of the People’s Education Press’s high school Chinese curriculum (readers can replace it with their course name), please provide ten innovative ideas for educational research topics, listing the research titles, significance, and specific implementation plans in a table format. When you see the desired research topic ideas appearing line by line in the table, what are your thoughts?! (4) Overall Reform Education must be prepared to meet the challenges of the knowledge industrial revolution in the age of artificial intelligence, adapting to the rapid development of AI and the future talent cultivation methods. The application of generative AI in teaching represents a profound transformation in teaching and learning methods, requiring a rethinking of everything from curriculum materials construction to learning resource transformation for both teachers and students. This is a systematic reform that requires comprehensive reform and development from educational concepts, institutional construction, organizational development, teacher training, curriculum teaching reform, and educational evaluation. Today, AI is creating a whole new world through generative means, and this revolution is forcing all knowledge economy sectors to join the ranks of generative AI, and all social services must face the digital and intelligent transformation. At this important juncture of transitioning from a knowledge economy to an intelligent economy, the Ministry of Education has proposed to accelerate the strategic development of educational digital transformation. We must fully utilize digital technology to seek new developments in education according to the national principle of “balancing development and safety, promoting innovation and legal governance”: First, change students’ learning through digital education, sparking a learning revolution; Second, empower teachers’ teaching through digital means, pushing for a teaching revolution; Third, drive school governance reform through data, accelerating precise educational governance changes; Fourth, reshape the new ecology of education and teaching led by educational digitalization. Chinese education needs to comprehensively analyze the impact and changes brought by generative AI on society and education, conduct transformation research on curriculum materials and teaching methods, and carry out new teaching practices of Generative Inquiry Learning to actively adapt to the arrival of the generative AI era.
(Author Li Jiahou is a professor at the Department of Educational Technology, Shanghai Normal University. This article is selected from “Curriculum, Materials, and Teaching Methods” 2024, Issue 2, pages 14-21, references omitted.)
“100 People in Educational Informatization” is a think tank media initiated by industry, academia, research, and media, focusing on educational informatization, educational digitalization, smart education, internet + education, artificial intelligence education, educational technology, etc. We aim to “make quality information and knowledge seen faster!”
Note Position + Name, Join the Reader Exchange Group
Contact UsJiao Xiaodi: Lead Writer& Business Cooperation WeChat: jiaox1008
Professor Huang Ronghuai from Beijing Normal University: How Digital Technology Empowers Current Educational Reform
Professor Wang Qiong from Peking University: Suggestions for Schools and Teachers in the Age of Artificial Intelligence
Ni Minjing: Humanity Will Enter a Super Learning Stage
Kevin Kelly’s Latest Speech in 2024: Understanding AI from Four Directions
Li Yongzhi, Director of the Chinese Academy of Educational Sciences: How Education Should Face the Challenges of Artificial Intelligence
