Beyond the screen: How AI research is redrawing the map of education
ECNU researchers use AI to understand and enhance the human elements of teaching and learning
image: ECNU's Shanghai Institute of AI for Education using AI as a lens to understand learners and empower teachers. View more: https://aiedu.ecnu.edu.cn/
Credit: Faculty of Education, East China Normal University, Shanghai, China
From the micro-dynamics of children to the macro-patterns of thousands of classrooms, these studies represent a cohesive mission. It uses AI as a powerful assistant to reveal the complex realities of human development and interaction, and then to build intelligent systems removing administrative burdens, thereby freeing educators to focus on mentorship, creativity, and ethical guidance.
1. Debunking Myths About Autistic Children
For decades, a cornerstone belief in autism diagnosis has been the apparent lack of eye contact. A groundbreaking study turns this assumption on its head. Using a multimodal behavior observation laboratory, researchers found that during play, both autistic and typically developing children focus primarily on toys (60–80% of the time), sparing only a small fraction (6–14%) for looking at an adult’s face.
“This challenges the longstanding belief that autistic children inherently avoid eye contact,” say Qu and Liu, “Our findings suggest we must reconsider assumptions and interventions. It indicates that focusing on broader communication cues, like gestures, may be more impactful than prioritizing eye contact alone.”
This revelation demonstrates the core approach: using AI-driven, naturalistic observation to achieve a more accurate, empathetic understanding of the learner.
2. LLMs as a Collaborative Partner
With a deeper understanding of learners, the next question is: how can technology sustain and empower those who teach them? Research on Large Language Models (LLMs), like ChatGPT, outlines their immense potential to automate time-consuming tasks: generating personalized quiz questions, creating programming exercises, drafting learning objectives, and providing initial feedback.
“LLMs demonstrate great potential in making large-scale personalized teaching a reality,” state Liu et al. “However, the emphasis should be firmly on collaboration, teachers take the role of collaborator and LLMs as the material generator, teachers need to monitor and supervise the performance of LLMs.” This suggests AI not as an authority, but as a versatile tool under expert guidance.
3. An Integrated System—ELion Intelligent Chinese Composition Tutoring System
The vision of human-AI collaboration is brought to life in the ELion Intelligent Chinese Composition Tutoring System, a joint project with Microsoft Research Asia. ELion’s evolution tells a strategic story of integration. It firstly began with a precise, only BERT-based model for automated scoring. Then, the emergence of ChatGPT presented not a replacement, but a partner. The team pioneered a synergy: BERT provides reliable, accurate assessment, while ChatGPT uses those results to generate friendly, style-adaptive feedback for students.
“We found that even for LLMs, it may be necessary to incorporate a variety of technologies—both old and new,” explain Zheng et al., “Innovative technology cannot solve problems in the actual world on its own.” This synergy has liberated teachers in over 250 schools, allowing them to focus on higher-order instruction.
4. Data-Driven Insights for Equity and Quality
Beyond assisting individual teachers, AI is also transforming our systemic understanding of the classroom itself. Do China’s classes predominantly center around teacher presentation instruction? Researchers from ECNU developed the AI-powered High-Quality Classroom Intelligent Analysis Standard system (CEED) to find out this answer.
Applying CEED to 1,008 classroom videos, the researchers revealed that on average, teacher presentation occupied 51.9% of class time. More crucially, the system uncovered a reverse trend: as students progress to higher grades, teachers tend to ask fewer open-ended questions, opting for “closed” and “safe” types despite the students' increased cognitive ability. It may be a potential mismatch between pedagogy and development.
“Using AI for big data annotation enables systematic statistical analysis of large-scale data, providing unprecedented objective feedback,” claim Gao and Yang. This moves classroom evaluation from record to evidence, offering teachers and administrators a powerful mirror to reflect on and improve practice toward greater interaction and equity.
The journey begins with seeing, moves to supporting, and ultimately leads to understanding. Each step is grounded in empirical research, and each step reinforces the ultimate goal: empowering educators with deeper insights and more effective tools.
***
References
Gao, Y., & Yang, X. (2025). Are China’s Classes Predominantly Centered Around Teacher-Presentation Instruction?—A Large-Scale Data Analysis Based on Classroom Intelligent Analysis Systems. ECNU Review of Education, 8(2), 349–355. https://doi.org/10.1177/20965311251322181 (original work published 2025)
Liu, J., Jiang, B., & Wei, Y. (2025). LLMs as Promising Personalized Teaching Assistants: How Do They Ease Teaching Work? ECNU Review of Education, 8(2), 343–348. https://doi.org/10.1177/20965311241305138 (original work published 2025)
Qu, L., & Liu, Q. (2025). Is a Child Who Doesn’t Look at People Always Autistic?—A Closer Look at Joint Attention. ECNU Review of Education, 8(2), 367–371. https://doi.org/10.1177/20965311251319050 (original work published 2025)
Zheng, C., Xia, W., Mao, S., & Xia, Y. (2025). ChatGPT, BERT, or Both? This Is Not a Question: The Evolution Story of LLMs in ELion Intelligent Chinese Composition Tutoring System. ECNU Review of Education, 8(2), 356–366. https://doi.org/10.1177/20965311251315205 (original work published 2025)