At the 29th Harvard China Forum held on April 11, 2026, Peng Zhuangzhuang, President of TAL Education Group, shared insights on the evolving role of artificial intelligence in education. He noted that as AI technology advances rapidly, learning methods and product forms within educational settings are undergoing significant transformation.
Peng highlighted that education is transitioning from a "knowledge-oriented" approach to a "competency-oriented" one, moving toward a new paradigm centered on the learner and driven by computing power. This shift aligns with recent reforms in China's "New Curriculum Standards," which emphasize the cultivation of subject literacy and skills over mere knowledge acquisition. He clarified that this change is not driven by any single company but represents a broader direction being pursued across the entire education system. The New Curriculum Standards advocate for competency-focused requirements across multiple subjects. For instance, mathematics education now places greater emphasis on using mathematical language to describe real-world problems, while language arts instruction prioritizes expressive and aesthetic abilities.
This transformation is also reflected in learning processes and assessment methods. Peng pointed out that contemporary exams increasingly feature greater reading volumes and comprehensiveness, which are seen as evaluations of interdisciplinary competencies.
When asked how AI can address the long-standing challenge of balancing high quality, personalization, and scalability in education—often termed the "impossible triangle"—Peng stated that AI offers new possibilities. Digital content and intelligent tools, for example, can provide personalized learning experiences to a larger number of students. However, he cautioned that increased resource availability does not automatically lead to reduced disparities. "In practice, students with stronger learning motivation are more likely to proactively utilize these resources," he noted, emphasizing that student motivation remains a critical factor influencing outcomes. Consequently, educational product design must continue to focus on stimulating active student participation.
Regarding product and industry competition, Peng observed that while some companies emphasize "large model integration" or the speed of model updates, the crucial question is whether these capabilities genuinely translate into improved learning outcomes, which requires validation through practical application. "Model capabilities are continuously improving, but the key lies in their ability to solve problems within specific learning scenarios," he remarked.
On the topic of quantifying AI's impact on learning behaviors, Peng mentioned that besides outcome metrics, behavioral data—such as whether students proactively ask questions or consistently engage in the learning process—are now considered important indicators. "Learning outcomes are the result, while these behaviors are closer to the process that generates those results," he explained.
From a technological perspective, Peng believes the industry's focus is shifting from standalone model capabilities to applications involving "Agent" systems. These systems coordinate different models, tools, and knowledge bases to accomplish specific tasks. He indicated that such systems rely not only on model performance but also on integrating contextual information from the learning process, such as a student's progress, textbook content, and classroom dynamics.
"Such contextual information is often missing in general-purpose models but is crucial in educational settings," he added.
When questioned about where the core assets of education companies will reside as general-purpose models become increasingly powerful, Peng suggested that differentiation will increasingly manifest at the application layer. Key differentiators include the ability to evaluate and orchestrate models based on accumulated real-world problem data, the establishment of learning contexts tailored to student progress and curricular content, and the skill of transforming content into formats easily understood by students. "While model capabilities themselves are constantly advancing, the critical factor is how we integrate models, data, and content to create learning experiences that students can effectively use," he concluded.
Comments