Generative AI's Disruption of Future Education: Opportunities and Challenges Coexist
-
"AI in education will shift from focusing on the application of intelligent technologies to human-machine collaborative development, showing a trend from weak AI to strong AI."
Recently, at the 2023 Global Smart Education Conference, Professor Huang Ronghuai, Co-Dean of the Smart Learning Institute at Beijing Normal University, made this statement in his speech. The conference was jointly organized by Beijing Normal University and the UNESCO Institute for Information Technologies in Education (UNESCO IITE).
At the "Generative AI and the Future of Education Forum" co-hosted by NetDragon (00777.HK), experts, scholars, and industry professionals from around the world discussed four key topics: the opportunities and challenges of generative AI in education, the reshaping of educational models by AI, human-machine collaborative teaching, and ethical issues in AI educational applications.
Professor John Shawe-Taylor, Director of the UNESCO International Research Centre on AI and a professor at University College London, stated: "We need fundamental solutions to expand the breadth and depth of education, including cross-cultural and cross-national approaches, enabling learners from diverse backgrounds and educational foundations to learn at all stages of life." In his view, AI can clearly drive these advancements.
He also mentioned the risks associated with AI: "It may make people's lives more monotonous and less imaginative. In AI design, we must ensure that everyone can benefit from AI fairly. Many current systems are not yet sufficient and lack transparency."
What can generative AI do in the field of education?
Professor Huang Hua, Dean of the School of Artificial Intelligence at Beijing Normal University, shared several experimental results on-site. When generating teaching cases for the first-grade Chinese textbook (Ministry of Education edition), ChatGPT provided a general template framework, iFlytek's (002230.SZ) Spark Cognitive Model offered content for the first two units, while Wenxin Yiyan still has some gaps to fill.
To Professor Huang Hua's surprise, when primary school Chinese, math, and English questions were input for three major AI models to solve, the accuracy rates were all relatively low. For junior high school Chinese, math, and English questions, ChatGPT answered two out of five correctly, Wenxin Yiyan (Baidu's model) got three right, and iFlytek Spark answered four correctly. Huang Hua analyzed that iFlytek Spark might have more educational scenario data in its training corpus.
When asked "Who is Bing Xin?", ChatGPT "completely talked nonsense," iFlytek answered half correctly, while Wenxin Yiyan got it all right. Huang Hua believes that as a Baidu product, Wenxin Yiyan benefits from Baidu Encyclopedia's training data, making its answers more accurate.
"Large models have weak reasoning abilities for complex problems, especially in logical reasoning, commonsense reasoning, and numerical reasoning, which will be difficult to improve in the short term," Huang said. "If a vertical large model is specifically developed for primary and secondary education with limited training data, its performance would definitely improve significantly."
Recently, multiple education companies have successively launched large models and their applications in the education sector. For example, NetEase Youdao (NYSE: DAO) introduced the 'Ziyue' education large model, with its first hardware product—the Youdao Dictionary Pen X6 Pro—featuring a virtual oral coach.
iFlytek has also integrated its Spark Cognitive Model into learning devices, offering functions like human-like grading for Chinese and English essays, interactive math tutoring, and English speaking practice. TAL Education Group (NYSE: TAL) is developing a math-focused large model, MathGPT, and plans to release product-level applications based on this self-developed model within the year.
At a forum, Chen Hong, Senior Vice President and CTO of NetDragon Websoft, shared multiple applications of generative AI, such as generating comprehensive post-class review materials from lecture recordings. These materials expand on explanations of key points, incorporating related videos, exercises, and teacher Q&A resources, all presented in a structured format.
Additionally, generative AI can create fairy tales incorporating vocabulary from Chinese lessons, paired with AI-generated picture books, making vocabulary learning more engaging and less rote.
'The rapid development of large models, now increasingly specialized, has indeed opened new directions for education,' said Chen Changjie, Vice President of NetDragon and Deputy Director of Beijing Normal University's Smart Learning Institute, during a post-event interview. 'On a smaller scale, it changes how humans acquire knowledge; on a larger scale, it may revolutionize teaching models, expanding into a three-dimensional structure involving teachers, machines, and students.'
'A major challenge is adapting our educational content to the customs and habits of different countries,' Chen Hong explained. NetDragon's education products are already available in nearly 200 countries and regions, with plans for its overseas business to spin off and list. To address localization, he cited examples like generating textbook illustrations tailored to different countries to ensure content aligns with local customs.
At the event, NetDragon also launched a public smart education platform—EDA (Edmodo Academy)—which leverages AI, the metaverse, 3D models, and micro-animations to provide open access to vast educational resources globally. Users can also modify and share course materials.
Regarding the specific applications of AI in schools, Zhang Zhi, Director of the Baoshan District Education Bureau in Shanghai, shared local practices. Each student's learning process is recorded to form a data trajectory, creating a learning profile. Knowledge graph-based adaptive learning can provide personalized assignments for every student.
Additionally, through data analysis and situational awareness analysis, students' psychological issues can be promptly identified, such as indicators of hesitation or happiness levels in facial expressions, thereby establishing an early warning system based on an 'educational brain.'
Fu Minjia, an information technology teacher at Lixian Primary School in Kecheng District, Quzhou City, Zhejiang Province, also praised the assistance of generative AI in her work. For example, it helps with writing papers, drafting plans, and particularly with composing student evaluations. 'It's very challenging to capture children's attention, and teachers often struggle with this,' she noted.
For example, by combining children's personality traits, plants are used to describe the entire class of children. "It not only tells you what plant they are but also describes them very accurately, saving us a lot of time and improving work efficiency."
"China's long-standing education model, which has been primarily focused on knowledge imparting, is about to change." Huang Hua believes that teaching should shift from knowledge transmission to learning guidance, emphasizing the cultivation of students' critical thinking. Evaluation should also transition from outcome-based, coarse-grained assessments to process-oriented, fine-grained comprehensive evaluations. Management should follow suit, moving from large-scale teaching data to more refined and convenient management.