AI in Education in China: Applications, Challenges and the Road Ahead

Introduction

Digital transformation of education has become a national-level strategic priority in China. The *Education Modernization 2035* blueprint (2019) explicitly calls for "building an intelligent education system serving lifelong learning for all," while the *Next Generation Artificial Intelligence Development Plan* (State Council, No. 35 of 2017) designates intelligent education as a key focus area. In July 2024, the Ministry of Education and several partner agencies jointly issued the *Guidelines on Strengthening Artificial Intelligence Education in Primary and Secondary Schools*, formally incorporating AI education into China's national education strategy. These policy moves mark a decisive transition: AI in Chinese education has moved from pilot experiments to systematic, large-scale rollout.

The breakout of generative AI has accelerated this shift further. Tools such as ChatGPT, Baidu's ERNIE Bot, Alibaba's Tongyi Qianwen, SenseNova, as well as domestic large language models including GLM and DeepSeek, are being deployed across Chinese education scenarios. From pre-class lesson planning to in-class interaction design, from personalised learning paths to automated homework grading, and from education governance to teacher professional development, AI is producing structural change across every layer of the education system. According to industry estimates, China's AI-education market surpassed the 100-billion-yuan scale by 2025, making it the most commercially viable vertical application of AI in the country.

However, the breadth of policy support and technological penetration does not automatically translate into the depth of educational value. Driven jointly by top-down policy momentum and rapid technological iteration, AI education in China now exhibits a characteristic pattern: *policy moves first, technology follows, practice remains uneven*. Against this backdrop, this article systematically maps the current state of AI in Chinese education, identifies representative use cases, diagnoses the structural bottlenecks that constrain practice, and synthesises both Chinese and international scholarship to propose a path forward — aiming to inform the high-quality development of AI in education.

The Current State of AI in Chinese Education

### (a) The macro picture — three driving forces

From the policy side, the institutional framework for AI education is essentially in place. The Ministry of Education has positioned AI at the centre of China's National Smart Education strategy and has been promoting the development of "Smart Education Demonstration Zones." By the end of 2025, more than fifty cities and districts had been brought into the national pilot programme. At the municipal level, Beijing, Shanghai, Shenzhen, Hangzhou and other leading cities have issued dedicated AI-education policies, forming a three-tier policy linkage running from national down to city level.

From the market side, the commercialisation of AI education has accelerated. iFlytek's "Smart Classroom" suite, TAL Education's "Magic Pen" intelligent learning tools and Yuanfudao's "Xiaoyuan Souti" (Little Ape Homework Scanner) represent distinct technology routes and business models. Industry data indicate that the number of AI-education enterprises in China has exceeded 1,500, spanning K-12, higher education, vocational education and corporate training segments.

From the technology side, multimodal large language models, knowledge graphs and adaptive-learning algorithms now form the technical base for education applications. Professor Huang Ronghuai of Beijing Normal University pointed out at the 2025 International Conference on AI and Education that "the rapid development of artificial intelligence is reshaping the global education landscape," and that open universities — as pioneers of technology application — are "standing at a critical historical moment." This framing exposes the tension between technological momentum and institutional response, while also signalling the strategic recognition that China's education community now accords to AI in education.

### (b) K-12 education — literacy-led curriculum redesign

In basic education, AI education has been endowed with the strategic mission of "cultivating the next generation of talent." Application in K-12 currently centres on two dimensions:

*First*, the formal inclusion of AI in the curriculum. The 2024 Ministry of Education guidelines require primary and secondary schools to integrate AI content into existing subjects such as science and information technology, and to gradually build an AI-education curriculum system. A number of regions began piloting school-based AI courses in the 2025–2026 academic year, organised on a "theory-and-practice-as-one" model covering AI fundamentals, key technologies, multimodal applications (text, image, audio, video), intelligent office automation, code generation, and AI ethics and law.

*Second*, AI-powered classroom enhancement. A large-scale empirical study by Xiao Wen, Pian Yang and Ma Xiaomin, based on a survey of more than 48,000 incumbent primary and secondary school teachers nationwide, shows a clear pattern: teachers have reached a high degree of consensus on the value of using AI, and their willingness to adopt it is strong, but their capacity in technical understanding, content-generation quality and deep integration with pedagogy lags significantly — with pronounced variation across educational stages, subject domains and years of service. The researchers identify the strongest demand for high-quality training support in lesson planning, classroom integration and curriculum research. Their diagnosis of a systemic bottleneck — *"insufficient cognitive updating, mismatched resource allocation and inadequate translation of training into practice"* — captures the core challenge at the K-12 level.

Equally important is a caution emerging from the scholarship: the risk of *cognitive offloading* as AI tools proliferate. Scholars Shang Junjie and colleagues identify three structural contradictions. At the *goal* level, a misalignment between "learning taking place" and "tasks being completed," as cognitive offloading marginalises the learner's own agency. At the *mechanism* level, a rupture between observable behaviour and internal learning processes, as data-driven methods struggle to represent cognitive and metacognitive states. At the *paradigm* level, an imbalance between technology-driven and learning-science-driven design. These tensions remind us that technological efficiency must not come at the expense of what education is fundamentally for.

### (c) Higher education — governance, ethics and the trusted-AI agenda

In higher education, AI application is concentrated in three areas: teaching support, governance decision-making, and ethical regulation.

In teaching support, universities are beginning to deploy AI-driven adaptive learning platforms and personalised learning-path recommendation systems. Some institutions have started integrating AI skills across entire professional clusters — for example in materials science, AI-technology applications and similar fields — through models that贯通贯通 (integrate) science, education and industry in talent development.

In governance and ethics, Chinese higher-education institutions are drawing on the risk-tiered management framework of the EU *Artificial Intelligence Act* to begin constructing a homegrown AI governance system. Scholars have used textual analysis of the EU AI Act's key clauses to propose a governance pathway for Chinese universities built around academic integrity, ethical safety and educational value as core principles. At the same time, the construction of *trusted educational AI* has attracted academic attention. Lu Yu, Lei Yixin and Chen Penghe proposed a three-layer architecture — *foundation layer, explanation layer, service layer* — respectively addressing secure and fair sharing of education data, dual explainability (both in pedagogical and technical terms), and transparency and accountability for education users. This framework offers an operationally grounded theoretical reference for the ethical safety of AI in education.

### (d) Vocational education and lifelong learning — industry alignment and skill renewal

Vocational education in China exhibits a distinctly *industry-aligned* flavour in its AI adoption. As AI increasingly empowers knowledge workers and intelligent creators, the "learn AI, use AI, create AI" model for developing next-generation talent is being promoted across the vocational system. The 2025 project catalogue published by the Jilin Provincial Vocational Education Research and Management Platform includes entries such as *"Research on the Dilemmas and Pathways for High-Quality Vocational Education Development under the Background of AI-Driven 'Technological Unemployment'"* and *"Research on the Impact of AI ChatGPT on the Cultivation of Vocational Education Talent"* — reflecting vocational educators' deep engagement with AI-driven labour-structure change.

In lifelong learning, intelligent learning-support systems based on open universities and online education platforms are under construction. Professor Huang Ronghuai has emphasised that AI's permeative impact on education is wide-ranging and requires coordinated planning across basic, higher and vocational education, building an integrated talent-development ecosystem spanning the full AI-education ladder.

Representative Use Cases

### Case 1 — iFlytek "Smart Classroom": end-to-end intelligent teaching

iFlytek's "Smart Classroom" is the benchmark deployment in AI education, now in use in primary and secondary schools across multiple provinces. Its architecture spans the full teaching cycle: *before class*, an intelligent lesson-planning system assists teachers in generating teaching content and activity designs; *in class*, speech recognition and affective-computing technologies analyse classroom interaction data in real time and generate instructional behaviour reports; *after class*, automated homework-grading and diagnostic learning-analytics modules deliver precision personalised tutoring recommendations. The case exemplifies the "full-chain embedding" model of AI-enabled teaching, but also exposes the adaptation gap between available tooling and teachers' technical capacity.

### Case 2 — National Smart Education Public Service Platform

The Ministry of Education-led National Smart Education Public Service Platform integrates AI-driven personalised learning-resource recommendation, covering basic, vocational and higher education. Built on big-data and knowledge-graph technology, the platform provides cross-disciplinary personalised learning paths for students, and has introduced learning-behaviour analytics and quality-assessment mechanisms. The 2024 National Social Science Fund for Education key project *"Research on Application-Demonstration Standards for the National Smart Education Public Service Platform"* (project No. ACA240027) takes this platform as its subject, advancing standardisation of AI-education deployment from a standards-design perspective.

### Case 3 — TAL's "Magic Pen" and Yuanfudao's "Xiaoyuan Souti": AI-powered after-school tutoring

TAL's "Magic Pen" and Yuanfudao's "Xiaoyuan Souti" represent AI after-school tutoring tools aimed at the K-12 segment. The former combines OCR and knowledge graphs to recognise questions and link multiple knowledge points; the latter uses image recognition and intelligent matching to deliver rapid question-answering. However, with regulators tightening oversight of "photo-search-homework" applications, this category of product is under pressure to pivot from *tool-based tutoring* toward *capability-based learning* — raising a deeper debate about the relationship between AI tools and learner autonomy.

### Case 4 — School-based AI courses in leading cities

In the 2025–2026 academic year, a number of schools in Beijing, Shenzhen and Shanghai launched school-based AI pilot courses. Curriculum design emphasises a "literacy-led" orientation, placing AI awareness, innovative thinking, AI application ability and social responsibility at its core. Content progresses from AI fundamentals to machine learning and generative AI, anchored in project-based learning where students apply AI tools to real-world creative and inquiry tasks. This direction aligns closely with the framework proposed by Zhang Yu and Long Yun: education must preserve "desirable difficulty" in thinking, ensuring students grow into subjects with comprehensive innovation capacity — able to *wield* AI rather than be *replaced* by it.

Challenges in Practice

### (a) The tension between technology logic and educational logic

The primary challenge in AI education today is a structural tension between technology logic and education logic. GAI integration into teaching and learning exhibits a "top-wide, bottom-narrow" pattern — heavy on system design, light on subject-specific practice. In terms of *who* does the research, subject-matter specialists are underrepresented; in terms of *method*, technology-functionalism tends to dominate; in terms of *scope*, the complexity of discipline-specific education is often overlooked. The root of this tension lies in the mismatch between the velocity of technological iteration and the rhythm of educational reform, leaving applications often stuck at the level of "technology demonstration" rather than genuine pedagogical transformation.

### (b) The teacher-capacity bottleneck

Large-scale survey data show a clear capability gap among primary and secondary school teachers. While willingness to use AI and recognition of its value are high, actual application ability is limited, particularly in high-quality content generation and deep integration of AI with subject pedagogy. The variation across educational stages, subject domains and years of service further deepens the uneven landscape of AI-education adoption. A training infrastructure gap has become the systemic bottleneck constraining the quality of practice.

### (c) Curriculum fragmentation and the missing integration ladder

A 2026 article in *China Educational Technology* identifies seven concrete challenges facing AI education today: fragmented curriculum goals, content repetition, lagging assessment systems, scarce curriculum resources, outdated teaching models, a weak faculty and incomplete institutional support. Integrated progression of AI education from primary through university is both a strategic route for co-ordinating education, science and talent development, and a strategic pillar of national rejuvenation. Yet in practice, effective linkage mechanisms between stages remain underdeveloped; the coherence and progression of AI-education content is hard to secure; and the gap between theoretical blueprints and grassroots practice remains large.

### (d) Ethical safety and governance

As AI penetrates deeper into education, issues of data privacy, algorithmic bias, academic integrity and cognitive sovereignty are becoming increasingly prominent. Zhang Yu and Long Yun offer a three-tier framing — the *logic of survival* (education must safeguard human knowledge sovereignty), the *logic of subjectivity* (defending human rational sovereignty in establishing the legitimacy of knowledge), and the *logic of education* (guarding against the hollowing-out of students' cognitive schemas). These perspectives provide valuable value-coordinates for building an AI-education governance system.

### (e) Regional disparity and the digital divide

Although policy momentum and market investment are accelerating AI-education adoption, significant disparities persist between regions in terms of technical infrastructure, faculty capacity and resource availability. The AI-education "digital divide" is widening between eastern and inland areas, urban and rural settings, and key schools and ordinary ones. Building an inclusive AI-education ecosystem — one in which technology truly serves all learners — remains a key unresolved issue.

A Path Forward — Towards a New Paradigm of AI in Education

### (a) Re-ground AI in learning science

The "content–process–support" triad proposed by Shang Junjie and colleagues offers a valuable anchor. On the *what* dimension, reconstruct learning content around cognitive structure and metacognitive development. On the *how* dimension, optimise the learning process through multi-mechanism coordination. On the *how-to-do-it-better* dimension, reshape learning support through process-regulation systems. The core idea is to build AI education on the mechanisms of learning itself — coupling technology logic with learning science, rather than letting the pace of technology overtake the rhythm of education.

### (b) A tiered teacher-training system

Addressing the capacity gap among K-12 teachers requires a multi-level training system covering the entire teaching workforce: *cognitive* training — universal AI-literacy upskilling to reshape teachers' understanding of AI; *differentiated* training — stage-specific, subject-specific and experience-specific content tailored to different cohorts; and *embedded* training — practical workshops focused on high-frequency scenarios such as lesson planning, classroom integration and curriculum research. The three-dimensional framework of "cognitive reframing, differentiated supply and scenario-based deepening" proposed by Xiao Wen et al. provides both empirical grounding and an operational blueprint.

### (c) Integrated progression across educational stages

Within the integrated framework of "ideological guidance, vertical continuity, horizontal integration and multi-party coordination," build an AI-education progression ladder: *vertical continuity* grounded in cognitive-development theory, ensuring effective linkage between stages; *horizontal integration* across the eight teaching elements — goals, content, models, assessment, resources, platforms, faculty and safeguards; and *multi-party coordination* through a "school–government–enterprise–family" ecosystem. This is the route to ensuring fair and equitable distribution of AI-education resources.

### (d) A trusted-AI governance framework

Drawing on the risk-tiered governance of the EU AI Act and adapted to China's education governance context, a three-layer trusted-AI model is proposed: the *foundation layer* focusing on education data security and algorithmic fairness; the *explanation layer* building dual explainability — both pedagogical and technical; and the *service layer* enhancing transparency and accountability toward education users. On this basis, clear rules of use must be established, with academic integrity, ethical safety and educational value as the core of governance — drawing the boundary within which AI education can develop healthily.

### (e) An inclusive AI-education ecosystem

Narrowing the AI-education digital divide — between regions, between urban and rural areas, between schools — requires coordinated action across infrastructure investment, resource-sharing mechanisms, teacher-mobility support and policy倾斜 (tilting). The National Smart Education Public Service Platform should further leverage its role as a national digital-education backbone, using AI-driven personalised resource distribution to lower the cost of AI-education access for remote and under-resourced schools, extending AI-education services to all learners.

Conclusion

AI in Chinese education has evolved from a technology topic into a systemic question of educational transformation. From policy deployment to market practice, from K-12 to higher education and lifelong learning, from the classroom to education governance, AI's penetration is reshaping the education ecosystem in every dimension. Yet the breadth of technology-enabled reach does not automatically equal the depth of educational value realised. The fundamental challenge facing practice today is how to strike the right balance between technology-driven momentum and education's own logic — ensuring that AI in education genuinely serves the all-round development of the learner, rather than substituting technology for education itself.

Looking ahead, the high-quality development of AI in education requires coordinated progress across three levels: *technologically*, rebuilding AI-education products and systems on the foundations of learning science; *institutionally*, completing the AI-education governance system and clarifying ethical boundaries; and *ecologically*, building an inclusive AI-education ecosystem that spans every educational stage, every region and every learner. Only then — in a process in which AI reshapes the landscape of education — can we embrace what technology makes possible, while holding fast to what education is for, and arrive at a genuine unity of technology-enabled potential and educational value realised.

---

References

[1] Xiao Wen, Pian Yang, Ma Xiaomin. "The Current State of Primary and Secondary School Teachers' AI Education Application and Training Optimisation Suggestions — An Empirical Analysis Based on Large-Scale Survey Data." *China Educational Technology*, 2025, 10, No. 465: 114–122.

[2] Shang Junjie, Liu Qian, Xia Qi. "The Theoretical Logic and Practical Pathways of AI in Education — A Perspective from Learning Science." *Frontiers in Educational Intelligence*, 2025.

[3] Zhang Yu, Long Yun. "Reconstructing and Holding Fast to the Logic of Education in the AI Era." *Frontiers in Educational Intelligence*, 2025.

[4] Lu Yu, Lei Yixin, Chen Penghe. "Research on the Construction and Application of a Trusted Educational AI Model." *China Educational Technology*, 2026 (01), No. 468.

[5] Guo Qing, Qiao Cuilan, Li Junling, Cui Hong. "The Logical Progression and Practical Pathways of Generative AI Empowering Subject Teaching." *China Educational Technology*, 2026 (01), No. 468.

[6] Ministry of Education et al. *Guidelines on Strengthening Artificial Intelligence Education in Primary and Secondary Schools*. 2024.

[7] Huang Ronghuai. *How Artificial Intelligence Changes Education*. Beijing Normal University, Smart Learning Research Institute.

[8] *Research on Application-Demonstration Standards for the National Smart Education Public Service Platform* (ACA240027). National Social Science Fund for Education Key Project, 2024.

[9] Ministry of Education. *Education Informatisation "14th Five-Year Plan".* 2022.

[10] Ministry of Education. *National Education Digitalisation Strategy Action.* 2022.

[11] *Research on AI Governance Pathways for Chinese Higher Education — A Study Informed by the EU AI Act.*

[12] China Academy of Information and Communications Technology, JD Explore Research Institute. *White Paper on Trusted Artificial Intelligence.* Beijing, 2021.

[13] State Council. *Next Generation Artificial Intelligence Development Plan.* State Council Document No. 35, 2017.

[14] CPC Central Committee, State Council. *Education Modernization 2035.* 2019.

[15] UNESCO. *AI and Education: Guidance for Policy-makers.* Paris: UNESCO, 2021.

[16] Jilin Provincial Vocational Education Research and Management Platform. *2025 Vocational Education Research Project Announcements.* 2025.

[17] Zhang Bingrui. *Research on Dilemmas and Pathways for High-Quality Vocational Education Development under the Background of AI-Driven 'Technological Unemployment'.* Changchun University of Traditional Chinese Medicine.

[18] Yang Shuang. *Research on the Impact of AI ChatGPT on the Cultivation of Vocational Education Talent.* Jilin Engineering Normal University.

Limits and notes

收尾：适用边界、注意事项。