Artificial intelligence is no longer a distant concept in education. It is already in the classroom. What began as a simple aid, much like a calculator or an electronic dictionary, is quickly evolving into something far more influential. Today, AI systems provide feedback on writing, guide reading, and even take part in evaluating learning. As this shift unfolds, classroom interaction itself is being reconfigured. Learning is becoming a dynamic exchange between learners and intelligent systems.
Research at Macao Polytechnic University points to a clear conclusion. The real issue is not how advanced the technology is, but how learners engage with it and how educators interpret and support that engagement. The focus is gradually moving away from the capabilities of tools and toward the quality of the learning process. In other words, the central question is no longer what AI can do, but how people learn with AI.
From Assistant to Learning Partner
Few areas illustrate this shift more clearly than foreign language writing. For decades, providing timely, personalised feedback has been one of the most demanding aspects of teaching. AI, at first glance, seems to offer a solution.
A study led by Associate Professor Teng Feng at the Faculty of Languages and Translation, involving 40 university students, explored how ChatGPT could support writing instruction. The results were encouraging: students who used the tool showed higher levels of motivation and were more actively engaged in class.
Yet the most striking finding lay elsewhere. What truly made a difference was metacognitive awareness, the ability to reflect on and regulate one’s own thinking.
Students with strong metacognitive skills did not simply accept what AI produced. They interacted with it. They questioned feedback, filtered suggestions, and treated ChatGPT less as a machine and more as a conversational partner. Others, however, took a more passive approach, reproducing AI-generated content with little reflection, and learning less as a result.
This distinction is revealing. AI does not automatically improve learning outcomes. Its value depends on whether students know how to engage with it thoughtfully. In this sense, the future of smart education hinges not on smarter machines, but on more reflective learners.
When Students Say “No” to AI Feedback
In human-AI collaborative learning, feedback from AI is not always readily accepted. A study led by Associate Professor Wei Wei at the Faculty of Applied Sciences, in collaboration with The Hong Kong Polytechnic University, examined how second-language learners respond to AI-generated feedback. Among 45 computer science students using ChatGPT for report writing, nearly half of the suggestions were ultimately rejected. Feedback related to content, such as argument structure and use of evidence, was rejected more frequently than feedback focusing on form, such as grammar and spelling.
Why does this happen? Drawing on the Unified Theory of Acceptance and Use of Technology (UTAUT), the study identifies several key factors: mismatches between AI feedback and student expectations, the cognitive effort required to process suggestions, inconsistencies with teacher or peer input, and limitations in students’ own language proficiency or confidence.
Importantly, rejecting AI feedback should not be seen simply as resistance. It often reflects active judgement. Students compare different sources, weigh the costs of revision, and, at times, defend their own writing intentions.
These findings carry important implications. Enhancing AI-assisted learning is not just about improving algorithms. It also requires embedding AI literacy into the curriculum, helping students learn how to ask better questions, evaluate the reliability of AI outputs, and make informed choices between human and machine input.
Understanding Learners through AI
If metacognition and AI literacy are so important, can AI itself help us better understand learners’ cognitive development? A large-scale study led by Associate Professor Patrick Cheong-Iao Pang at the Faculty of Applied Sciences, in collaboration with the Northern Arizona University, University of Science and Technology of China, and Fuzhou University, set out to address this question.
Using five AI models alongside four explainability techniques (including feature importance, Morris sensitivity analysis, SHAP, and LIME), the team analysed data from more than 1,000 first-year university students, aiming to interpret the so-called “black box” of predictive modelling.
The results point consistently to two influential factors: students’ self-perception and parental expectations. For example, students who perceived themselves as weak in mathematics tended to show lower overall cognitive performance, while higher parental expectations were associated with stronger outcomes.
At the same time, different explainability methods highlighted different aspects of the data. Some were more sensitive to health-related factors, such as vision or emotional fluctuation, while others captured more complex interactions among variables. Together, these approaches offer educators a more nuanced and multidimensional perspective.
When educators understand how AI arrives at its conclusions, they are better positioned to refine teaching strategies, design targeted interventions, and identify students in need of additional support.
Towards a Human-Centred Vision of Smart Education
Macao Polytechnic University has long been committed to advancing research and teaching at the intersection of artificial intelligence and education. Together, these studies address a fundamental question: how can we build learning environments that remain genuinely human-centred in an age of AI integration?
As technology continues to evolve, so too do the challenges and opportunities of smart education. The University places strong emphasis on cultivating AI literacy, critical thinking, and independent learning, ensuring that technological development supports the broader goal of holistic human development.
Recent initiatives include the launch of a PhD programme in Educational Technology and Innovation, as well as co-hosting the IEEE TALE 2025 conference, which brings together scholars from around the world to explore the future of AI in education.
In an era of human–AI coexistence, the key concern extends beyond how much students know. It lies in how they think, how they learn, and how they preserve human judgement and creativity while working alongside intelligent systems. Through sustained research and forward-looking vision, MPU continues to contribute to this evolving conversation.
Sustaining Living Heritage in the Digital Age