This article by Huang Ronghuai was originally published in Academic Frontiers, issue 14, 2024. Its analysis and examples retain the context of that publication.
Article contents
Abstract
The iterative evolution of large AI models is prompting education systems to become conscious of the need for systemic change. Education is at a critical stage, moving from passively accepting external change towards actively adapting and driving internal transformation. At the conceptual level, the continuing integration of these models is accelerating changes in views of education and teaching: a new conception of knowledge based on co-creation and sharing; learning based on intelligent connections and construction; integrated, open curricula; and human–AI collaborative teaching.
In practice, their deeper application is fostering new educational forms: diverse, cross-domain collaborative teaching, mutually empowering collaborative learning, and safe, trustworthy collaborative decision-making. Responding to this wave requires continuing to strengthen teachers’ and students’ competencies for the intelligent era, improving mechanisms for admitting intelligent products into schools, conducting human–AI collaborative educational practice in an orderly way, developing standards for educational applications of generative AI, and safeguarding technical security and ethics. These measures can support a new smart education ecosystem.
Clusters of next-generation information technologies, including the internet, big data and AI, are driving structural reorganisation and process redesign throughout production and daily life, while changing how education is organised and delivered. General Secretary Xi Jinping has described AI as an important driving force behind a new round of scientific and technological revolution and industrial transformation, profoundly changing how people produce, live and learn and ushering in an intelligent era of human–AI collaboration, cross-sector integration, co-creation and sharing. He has called for actively promoting the deep integration of AI and education to advance educational transformation and innovation.
Large AI models exhibit scale, generalisation, emergence and versatility. Their powerful capabilities in generating multimodal text, images, audio and video, conducting contextual dialogue, reasoning logically, calculating with data and handling tasks are accelerating intelligent transformation across industries and creating new opportunities for deeper educational digital transformation. Understanding the changes in educational conceptions and forms arising from these models, and enabling their reasonable, orderly and safe integration into the education ecosystem, is a key issue for education in the intelligent era.
The iterative evolution of large AI models is awakening education systems to systemic transformation
Disruptive technologies intensify international competition and social transformation
Global scientific and technological innovation has entered an exceptionally intensive and active period. A new technological revolution and industrial transformation are redrawing the global innovation landscape and reshaping the world economy. Continually evolving disruptive technologies profoundly affect countries’ futures and people’s well-being.
First, rapid technological development has become an irreversible trend. AI, quantum computing, biotechnology and other technologies are entering production and life in clusters, continuously and across many dimensions. Their effects have, to some extent, exceeded expectations and the capacity of social, economic and political systems to adapt. They may even surpass intuitive human perception and prediction, introducing new variables into international competition.
Second, AI is a current focus of international competition. Major countries are seeking breakthroughs in scientific and technological innovation and adopting national AI strategies, treating accelerated AI development as an important means of taking the initiative in global technological competition.
Third, technological progress is compelling comprehensive social transformation. Its pressures are increasingly embedded in industrial structures, labour markets, lifestyles and governance, profoundly changing how society operates and is structured. The World Economic Forum’s white paper Jobs of Tomorrow: Large Language Models and Jobs identifies occupations characterised by repetitive work, such as credit authorisers, cashiers and clerks, as highly exposed to AI, with as much as 81% of their tasks potentially automatable.
Comprehensive social transformation therefore calls not only for keeping pace with technology, but also for adjusting talent development systems, optimising industrial structures and innovating in governance to respond actively to technological challenges and opportunities.
Capital logic drives large AI models into multiple application settings
Capital has evolved from a conventional provider of economic resources into a catalyst for integrating AI and other technologies into diverse settings. Capital flows often indicate the direction of technological development, accelerating iteration and encouraging deeper integration across industries.
First, investment provides continuing support for model development. In little more than a decade, AI has progressed from task-specific models using supervised learning with labelled data, to pretrained models combining unlabelled-data pretraining and labelled-data fine-tuning, and then to large models combining large-scale unlabelled-data pretraining, instruction fine-tuning and human alignment. Development continues to accelerate and iteration cycles are shortening. Different types now include cloud-side and on-device models, general-purpose and industry-specific models, and language and multimodal models. Since 2022, OpenAI has successively introduced ChatGPT, GPT-4, Sora and GPT-4o.
Multimodal interaction and content generation have matured further, with development moving towards multimodal models, video-generation models, embodied intelligence and AI for Research (AI4R).
Second, capital drives deployment across fields. As model technology matures, the potential of extensive data and diverse application settings needs to be realised. Innovation is shifting towards using existing data efficiently and exploring new settings. Examples include supporting more accurate diagnosis in healthcare, risk control and investment decisions in finance, and traffic optimisation and congestion reduction in transport. Overall, capital logic increasingly steers model development towards application needs. Large AI models are driving intelligent transformation across industries and, when combined with the Internet of Things, big data and cloud computing, can provide more intelligent solutions and support industrial upgrading.
Education is actively adapting to the new wave of large AI models
An observable trend in current educational change is the critical transition from passively accepting external developments to actively adapting and driving internal transformation. This represents education’s awakening: the system becomes an agent of change, rather than only responding to technology and capital. Education authorities must consider not just direct applications but the educational logic behind transformation, preparing people for future development.
First, higher education is at the forefront. Institutions are rapidly integrating new technologies. AI teaching assistants and intelligently connected learning environments change content delivery and teaching organisation, while interdisciplinary research and learning use AI to renew scientific research paradigms, integrate knowledge and enable innovative applications.
Second, vocational education faces more urgent transformation needs. As demand for skilled labour rises, it is a key bridge between education and employment. Institutions and governments are increasing investment and strengthening industry partnerships so that curricula connect adaptively with market needs.
Finally, school education is steadily advancing digital transformation under strong policy guidance. Governments at different levels are providing supportive policies and funding for smart education platforms, big-data and AI applications in teaching and management, and digital literacy training for teachers and students.
Education is a key field for accelerating the application of large AI models
International organisations, governments, academia and industry are collectively exploring their integration into education.
First, documents encourage, regulate and guide educational applications. UNESCO’s Guidance for Generative AI in Education and Research aims to support better integration of generative AI into education. The UK Department for Education’s publication on generative AI in education sets out its position on educational uses, including large language models such as ChatGPT.
Second, research and practice explore implementation pathways in an orderly way. Vanderbilt University provides extensive training and workshops for faculty, staff and students to help them use large AI models. Tsinghua University has launched a pilot-course initiative for AI-empowered teaching, developing discipline-specific applications of large language models and diverse teaching scenarios.
Positive responses from different actors show education moving from external drivers towards internal awareness of the need for change. Stakeholders must remain clear-headed and cautious about accelerating technological innovation and avoid being swept along by capital logic. They must also understand the models’ potential educational value, technical risks and limits. A human-centred approach should enable models and educational needs to evolve interactively and realise educational benefits. Through collaboration among multiple actors, large AI models can help change educational conceptions and forms and build a smarter, more flexible and efficient education ecosystem.
The continuing integration of large AI models accelerates changes in conceptions of education and teaching
Breakthroughs in large AI models have profoundly affected education and prompted reflection across society on education in the intelligent era. General Secretary Xi Jinping has stressed a virtuous cycle linking education, science and technology, and talent; coordinated strategies for invigorating the country through science and education, strengthening the nation through talent, and innovation-driven development; and integrated progress in education, technological innovation and talent cultivation. In recent years, China has introduced education policies that systematically adjust national arrangements for educational goals, school structures and curriculum standards.
These are concrete strategic expressions of national intentions for talent development, intended to respond to changing times and prepare people for national strategic needs.
However, examination-oriented thinking means that progression rates, rates of excellent performance and admission to key institutions remain common yardsticks for judging education. Standardised talent development cannot meet society’s diverse needs. Educators, researchers and the wider public urgently need to reconsider the changes and challenges that large AI models bring to educational ideas and build broad, strong social consensus. This can strengthen coherence within the education ecosystem and enable constructive interaction and coordinated progress among national, school and societal efforts to develop people. Driven by policy, technological innovation and social expectations, transformation is urgent.
Its key is to understand and advance fundamental changes in conceptions of knowledge, learning, curricula and teaching: the four new conceptions for the intelligent era.
A new conception of knowledge based on co-creation and sharing
This conception emphasises acquiring diverse, high-dimensional metacognitive knowledge. It encompasses relational knowledge generated through ‘federated learning’ across neural, social and conceptual networks; contextual knowledge arising from embodied cognition through interactions between bodily capabilities and technological environments; and rule-based knowledge formed through machine learning, including deep and reinforcement learning. Such knowledge displays collective intelligence, openness, fragmentation and ambiguity. By modelling, analysing and predicting from large datasets, large AI models uncover implicit relationships and patterns. Combined with human instructions and context, they creatively generate knowledge content and simulate human thinking to enable multidimensional analysis of complex problems, changing how knowledge is produced.
A new conception of learning based on intelligent connections and construction
Intelligent technologies create extensive possibilities for learning situations. Learners should move from passive reception to self-direction, acquiring knowledge, skills and attitudes through active collaborative construction with peers, teachers and intelligent agents. In environments where humans and machines coexist, agents based on large AI models can become learning partners. Interactive environments and roles such as learning assistants and intelligent companions provide support for essay correction, spoken-language practice, tutoring, programming, writing, psychological guidance and mock interviews, improving the learning experience.
A new conception of integrated, open curricula
Educational uses of intelligent technologies open previously closed curriculum boundaries. Content moves from fragments towards networks and structures, and curricula from separate subjects towards integration and contextualisation. Learning shifts from knowledge alone towards an organic combination of scientific understanding, technological experience, social participation, cultural awareness and understanding of life. Large AI models systematically renew curriculum development and management, enabling diverse, integrated individual curricula and personalised, interactive experiences.
Shanghai Jiao Tong University’s Siyuan AI Learning Support Platform, for example, allows students to draw knowledge graphs to consolidate learning and develop subject knowledge maps that identify overlaps and clear differences, helping optimise curricula.
A new conception of human–AI collaborative teaching
Large AI models and related technologies supplement, enhance and extend teachers’ capabilities, encouraging a shift from transmitting knowledge to constructing learning situations and organising and guiding activities. Collaborative teaching emphasises multidimensional teacher–student–machine communication and safe, trustworthy, ethical interaction with intelligent agents. Educational robots, avatars and digital twin teachers can interact deeply with teachers and students, serving as administrative assistants, teaching assistants or co-teachers. In Tsinghua University’s Mind, Individual and Culture course, for example, an AI teaching assistant provides accurate, efficient feedback on students’ writing to improve their writing skills.
Deeper applications of large AI models foster new forms of human–AI collaborative education
Large AI models mark a historic point of innovation in machine intelligence. They accelerate systemic changes in conceptions of knowledge, learning, curricula and teaching, while creating momentum to reshape educational forms. Human–AI collaboration in education means people and machines interacting and working as an integrated whole across educational environments to achieve shared goals and improve outcomes. Its practical logic is to combine complementary strengths and collaborate effectively to meet diverse needs. Compared with earlier assistive tools with limited intelligence and adaptability, large models represent a leap in machine intelligence and expand interaction modalities.
From this perspective, they offer opportunities for systemic educational change through collaborative teaching, collaborative learning and collaborative decision-making.
Diverse, cross-domain human–AI collaborative teaching
Large AI models further accelerate the transition from teacher–student to teacher–student–machine structures. Relationships become more complex across practical settings, featuring human–machine coexistence and multidimensional interaction, and supporting co-teaching.
First, teaching scenarios expand. Large models can make machine intelligence more effective in specific settings and overcome application bottlenecks. Combining humanoid robots and large models produces embodied intelligence with stronger multimodal perception, understanding, adaptation, autonomous planning and decision-making in the physical world. Intelligent interaction with real environments can support human–AI dual-teacher classrooms. Model-driven learning platforms, intelligent tutoring systems and other teaching agents can overcome time, space and teaching barriers to provide efficient, immediate and adaptive support for diverse online learning needs.
Second, collaboration moves to a deeper level. As technologies handle a wider range of tasks, teaching enters a field jointly shaped, led and controlled by teacher and machine intelligence. This changes teacher roles, knowledge authority and instructional design. Effective collaboration goes beyond dividing tasks or substituting one party for another. It requires stronger technological literacy, classroom management, responsiveness and flexible coordination. Teachers need to apply professional experience and practical wisdom to diverse, deeply collaborative task relationships with intelligent agents.
Mutually empowering human–AI collaborative learning
Educational applications gradually change how learners learn. Alongside their own cognitive processing, learners can access machine intelligence through cognitive offloading and improve that intelligence through iterative feedback, producing human–AI co-learning.
First, learning experiences improve. More intelligent interaction allows machines to act as tutors, companions and assistants, naturally collecting and analysing multimodal learning-process data to explore implicit emotional changes and cognitive states. They can offer immediate emotional companionship, targeted learning support and personalised learning pathways.
Second, learning mechanisms change. Collaborative learning alters information processing: instead of interaction only between a learner and the external environment, a collaborative intelligent structure forms between learner and machine, affecting how learning occurs. When they complete activities or tasks together, machines share cognitive work previously undertaken entirely by the human brain. Large models become an external cognitive resource supporting joint understanding and processing, allowing human and machine intelligence to complement and strengthen one another.
Safe, trustworthy human–AI collaborative decision-making
Successive model breakthroughs can upgrade educational governance towards evidence-based decisions, data-driven processes and human–AI collaboration. Governance becomes more scientific, precise and intelligent.
First, governance methods transform. Educational data increasingly grow rapidly in volume and speed and are fragmented and diverse, posing governance challenges. Large models can extract and process extensive information across modalities and uncover implicit patterns. Dynamic monitoring, evaluation and feedback can provide decision-makers with timely, comprehensive and objective analysis. This can reduce decision pressure from information overload and address distortion, cognitive bias and delayed adjustment across multiple decision-making levels, enabling a shift from broad-brush, experience-based governance towards data-driven, scientifically precise, evidence-based governance.
Second, governance thinking changes. Collaborative governance can generate new ideas and insights as well as new methods. Training and analysing large datasets and extracting features of physical objects can help create digital-twin mapping environments for education systems. Simulated experiments can reveal patterns and predict educational elements and processes to inform real practice. This supports a shift from reactive governance towards proactive, anticipatory governance.
Key initiatives for using large AI models to build a smart education ecosystem
Continuously strengthen teachers’ and students’ competencies for the intelligent era
Responding to technological effects on labour markets and career prospects requires prioritising competencies and improving digital creativity, adaptability and capacity to thrive.
For students, first improve mechanisms for developing digital literacy and skills. Teach information technology effectively and establish assessment and monitoring systems that track progression and accurately map development. Second, strengthen independent learning. Encourage active-learning approaches that cultivate self-planning, self-monitoring and self-evaluation. Third, improve learning with intelligent technologies. Large models can support knowledge acquisition, independent learning and companionship, helping students review and learn course content, enjoy better experiences and manage learning time, tasks and processes.
For teachers, first encourage active exploration and use of intelligent technologies. Their engagement can move through awareness, experience, practice and dissemination. Teachers should learn AI knowledge and principles, use technology to support their own learning and teaching, and exchange experience. Second, cultivate awareness of changing roles. Schools should provide digital literacy and skills training based on individual needs, improve human–AI collaborative teaching, and encourage teachers to provide emotional support and develop students’ character. Third, use targeted teaching research to support professional development.
Intelligent technologies can collect and analyse data of different types, sources and dimensions to identify gaps in pedagogy, subject knowledge and technology use. Summaries and analytical reports can support precise, collaborative teaching research, helping teachers understand classroom processes, reflect on design and practice, and improve teaching.
Improve mechanisms for admitting intelligent products into schools
Integrating these products into education while preventing privacy and security problems and harmful content requires safeguards for a healthy, safe learning environment.
First, establish comprehensive, differentiated admission review. Classify products and grade their risks, examine educational intelligent products in detail by subject, and analyse the chain of associated times, places, people, resources, functions and disciplines. Rigorous review and adaptation should support integration into diverse settings and match personalised needs at different educational stages.
Second, explore comprehensive, continuous dynamic oversight. Evaluate safety and educational suitability, conduct spot checks and routine inspections, dynamically manage black, grey and white lists, and develop scenario-based risk warnings and emergency mechanisms. Assess both safety and effectiveness.
Third, develop education-specific large AI models with greater accuracy, logical coherence and educational suitability. Strengthen ethical testing, reduce errors, bias and inappropriate outputs, and build safe, compliant and ethical digital learning environments.
Fourth, improve practical guidance for campus applications. Education authorities, practitioners and researchers should jointly develop guidance for student growth, teacher development and school transformation. Starting from differentiated, augmented and collaborative teaching, knowledge acquisition, independent learning, campus empowerment and stronger home–school connections, guidance should support reasonable, compliant use.
Conduct human–AI collaborative educational practice in an orderly way
As intelligent technologies evolve and enter education, collaborative teaching is becoming an important direction for future practice. Teachers should be encouraged to explore and use these technologies effectively in collaborative environments.
First, use digital pedagogy as the theoretical foundation. Anchor practice in effective teaching and deep learning; optimise digital environments; use resources and new technologies as means; and pursue effective learning and teaching in trusted digital environments centred on learners and technological empowerment.
Second, deepen research into collaborative teaching mechanisms. Multidisciplinary and interdisciplinary methods should investigate teaching behaviours and complex teacher–machine–student interactions, enriching theory. Evidence-oriented research should systematically evaluate overall effects and iteratively improve teaching structures and procedures.
Third, issue practice guidance and teaching standards. Encourage generative AI in analysing learning needs, retrieving resources, generating test items, instructional design and assessment feedback, while guiding students’ reasonable use.
Fourth, disseminate experience. Identify examples that improve educational quality and draw on exemplary teachers’ leadership. Systematically organise training around real teacher needs and spread advanced, transferable, high-quality practice. Build professional communities and online platforms for teacher exchange, collaboration and resource sharing, jointly exploring new directions.
Develop standards for educational applications of generative AI
Preventing unpredictable risks and protecting learners whose judgement is still developing requires ethical principles, service norms and educational service standards.
First, technical standards provide foundational, general and forward-looking requirements. These include, without being limited to, stable and scalable platform architectures; convenient, secure access; strict data encryption and privacy safeguards; common data-exchange protocols for interoperability; and clear ownership and use rights for content.
Second, quality standards should measure platforms’ value, efficiency and effectiveness. They should cover regular testing of model accuracy and stability, reliable and diverse data sources, monitoring generated content for accuracy, relevance and suitability, and evaluating whether applications effectively support teaching objectives.
Third, service standards should cover technical services, content rights, operating guidance and feedback. They should define application boundaries, users and scenarios; provide guidance for teachers and students; and include detailed procedures and good practice. Technical services should include round-the-clock support and regular training. Content rights should clarify intellectual property ownership to prevent infringement. Operating guidance should explain system procedures and usage norms. Accessible feedback channels should enable timely responses to questions and suggestions.
Build a framework for large-model technical security and ethics
Large-scale use may create problems including false or incorrect information, privacy and data breaches, and algorithmic bias and discrimination. Multiple actors must coordinate to ensure orderly, compliant design, development and application.
First, strengthen regulation and external oversight. Relevant policies should guide and manage development and use. Protect data and privacy against leakage, unauthorised access and other violations. Implement data governance and trusted computing, adopt strict encryption standards, and conduct routine security audits and risk assessments.
Second, strengthen the responsibility of research and development organisations and personnel. Technically, improve traceability in system design and application and explainability in decision procedures. Design transparent algorithmic frameworks, record and track decisions, and provide detailed explanatory documentation. Socially, strengthen industry self-regulation and embed trustworthy principles to prevent, monitor and eliminate discrimination and bias. Establish industry standards and self-regulatory agreements and publish regular transparency reports.
Third, strengthen users’ self-restraint and application norms. Use should comply with data and cybersecurity laws and regulations and remain reasonable and responsible. Avoid unlawful or unethical conduct, invasions of privacy and data misuse. Internal training and communication should strengthen legal awareness and ethical responsibility, alongside oversight and complaint mechanisms.
Author: Huang Ronghuai
Source: Academic Frontiers, issue 14, 2024
Original publication DOI：10.16619/j.cnki.rmltxsqy.2024.14.003
Original link: https://aiedchair.bnu.edu.cn/news-9689-en/
