Abstract: AI is already making its way into Hong Kong’s classrooms, but achieving personalised learning at scale depends on more than adopting increasingly advanced tools. What matters is whether learning data can be preserved, integrated and used over time. At present, schools and AI platforms often rely on different data formats and standards. When students change schools, advance to a new stage of education or move between platforms, their learning records may not follow them, creating “data silos”. This article argues that Hong Kong should establish three guiding principles—data portability, interoperability, and data ownership and control—so that AI can help teachers develop a sustained understanding of each learner and bring genuinely personalised education into the classroom.
Our previous article discussed how teaching students according to their individual needs has long been an enduring ideal in education. In the past, however, constraints on teachers’ time, class sizes and resources meant that only a small proportion of students could benefit from genuinely personalised instruction. By continuously recording the learning process and analysing students’ needs, AI has made personalised learning at scale possible for the first time. Yet turning that possibility into reality depends on whether learning data can be preserved and used over time.
Knowing the Learner Is the Foundation of Personalised Teaching
Imagine a primary school student for whom AI has been part of everyday learning since the first day of school. He uses AI to revise his compositions in Chinese lessons and practise pronunciation in English lessons, among other activities. Over time, AI records not only his scores, but also the mistakes he frequently makes, the concepts he struggles to understand, and even the kinds of prompts most likely to help him grasp a difficult idea.
The notion that AI could know a student this well is not science fiction. It is a transformation that education systems around the world are already confronting. In its Guidance for Generative AI in Education and Research, UNESCO observes that generative AI is forcing us to reconsider why, what and how we learn, as well as how learning should be assessed and validated. As AI reshapes curricula and the capabilities students need, assessment can no longer focus solely on final answers. It must become more process-oriented, examining how students think and learn. In other words, AI is changing not only what and how students learn, but also the criteria by which we judge how well they have learned.
This marks a new starting point for knowing the learner. Research on learning analytics has repeatedly shown that data accumulated throughout the learning process provide a far more complete picture of a student’s actual abilities and needs than a single score. A substantial body of research has also demonstrated that continuous assessment based on such evidence can improve learning outcomes. The value of AI lies in making this comprehensive picture of learning possible for the first time—but only if the underlying data can accumulate over time.
Returning to our hypothetical student, suppose he changes schools in Primary Three. His original school has already discovered that he loses marks in mathematics not because he cannot perform the calculations, but because he misidentifies the key information in word problems. His low composition scores do not reflect a lack of ideas, but difficulty organising paragraphs. His teachers even know that visual breakdowns of mathematical problems and talking through his ideas before writing are particularly effective for him.
His new school, however, uses a different set of AI platforms. None of the data accumulated over the previous two years can be transferred. His new teacher sees only a report card showing a score of 72 in Chinese and 65 in mathematics and must begin the process of understanding him all over again. Three years later, when he enters secondary school, the same thing happens once more. Six years of exercises, revisions, questions and learning records for an entire class cannot be carried forward, leaving the secondary school teacher to get to know more than 30 students from scratch.
Data Silos Are a Major Challenge
This is the challenge that Hong Kong must address in the next stage of AI-enabled education. In June 2026, the Education Bureau published the Blueprint for Digital Education in Primary and Secondary Schools. On the same day, it provided every publicly funded school with a grant of HK$500,000 to procure AI solutions, with all publicly funded schools participating in the initiative. Together with the HK$2 billion earmarked under the Quality Education Fund, these measures mean that AI has now entered the classroom. The next question should be: can these AI tools genuinely help teachers understand their students and improve teaching?
Over the course of a year, the Hong Kong Future Economy Institute’s research team visited more than 80 primary and secondary schools and analysed over 200,000 pieces of de-identified student work completed with AI. Although schools had adopted AI to varying degrees, the effectiveness of its use differed considerably. The decisive factor was not which school had the most advanced AI, but whether learning data could be continuously preserved, integrated and used.
A small number of schools that have established mechanisms for tracking data are already using AI to improve teaching. Some primary schools continuously compare changes in students’ vocabulary, sentence structures, content and organisation before and after AI-assisted interventions. Some secondary schools use AI to identify common learning difficulties, reduce repetitive work and devote more time to individual guidance. AI does not replace teachers in delivering instruction; it helps them understand their students.
Yet a school rarely uses only one AI tool. Different platforms store data from different subjects, while different providers use their own formats and standards. The data are consequently scattered across multiple systems and are difficult to integrate. Teachers are unable to see a complete picture of the student. This is the problem of “data silos”.
Data silos matter not because more data are inherently better, but because valuable data can change how teachers interpret a score. The same score may represent three entirely different students: one who has not mastered basic calculations, another who calculates well but struggles to understand word problems, and a third who understands the concepts but repeatedly misses steps in multi-stage questions. If a teacher sees only a score of 62, the three students appear identical. If the teacher can examine their problem-solving processes over time, it becomes clear that they require entirely different forms of support.
This is the essence of teaching according to individual needs. The first student needs practice in basic skills; the second needs help reading and breaking down word problems; and the third needs to develop the habit of checking each step. The true educational value of AI does not lie in automatically deciding what kind of education each student should receive. Rather, AI can organise the information scattered across assignments, questions, revisions and mistakes, making it easier for teachers to identify students’ needs and use their professional judgement to determine how best to intervene.
AI makes personalised learning at scale possible, provided that it helps teachers develop a sustained and in-depth understanding of their students.
Such understanding, however, takes time. Knowing where a student made a mistake today provides only one data point. Seeing the same problem recur over time allows a teacher to form a judgement. Observing how the student responds to different teaching interventions then reveals what works. If these records are interrupted whenever a student changes school, moves to the next stage of education or switches platforms, AI’s most valuable capability cannot be realised.
Three Principles for Using AI to Support Personalised Learning
Enabling AI to support personalised learning does not mean that every school must use the same AI system. Nor does it mean that the government must establish a central platform containing all student data. The real issue is whether students’ learning histories can continue to accumulate across different platforms, schools and stages of education.
Achieving this requires at least three principles: data portability, interoperability, and data ownership and control. These principles answer three practical questions: Can students take their data with them? Can different platforms understand one another’s data? Most importantly, who ultimately decides how the data may be used?
The first principle is data portability: data must be transferable. When students change schools or advance to a new stage of education, when schools switch AI platforms, or even when a service provider ceases operation, valuable learning records accumulated in the past should not disappear. They should be exportable and transferable in a commonly used, readable format.
This does not mean that every conversation between a student and an AI system, or every exercise the student completes, must be preserved permanently. Rather, parts of the learning history that have educational value should be able to follow the student. The value of portability is that students can change schools and platforms can be replaced without forcing teachers’ understanding of those students to return to zero.
The second principle is interoperability: data must be able to flow between different AI platforms. One of the greatest challenges today is that, even when data can technically be transferred, different platforms may define and record the same learning ability in entirely different ways. For example, one English-learning platform may categorise all of a student’s mistakes as “grammar errors”, while another may distinguish between errors involving tense, articles and subject–verb agreement.
Even if the two systems are technically connected, their records may still be difficult to compare directly. Genuine interoperability therefore requires the gradual establishment of common data formats, field definitions and recording practices, as well as a common level of data granularity. Different platforms need a basic, shared understanding of what the data mean.
The third principle is data ownership and control: students and schools should retain an appropriate degree of control over learning data. The information accumulated daily by AI education platforms is not merely a record of general usage. It constitutes a multi-year account of a child’s learning journey, including strengths, weaknesses and changes in ability. In some respects, it resembles a continuously updated “learning profile”.
Such data should not automatically become an asset that a provider may use at will simply because they are stored on an AI company’s servers. Clear rules should determine who may access the data, whether they can be exported when a student changes schools, and whether providers may use them to train AI models or for other secondary purposes. These rights and boundaries should not be determined unilaterally by individual platforms.
This does not mean that secondary uses with educational or research value should be prohibited altogether. Subject to clear rules, appropriate authorisation and adequate safeguards, anonymised or de-identified data may still be used to analyse teaching effectiveness or improve AI models.
What matters is that permission to use the data does not amount to ownership by the platform. Students’ learning data can create new educational value, but the purposes, methods and permissions governing their use should be clearly defined. Students, parents and schools should also retain an appropriate degree of control.
The three principles are closely interconnected. Portability ensures that a child’s learning history is not interrupted when the child changes schools or platforms. Interoperability ensures that information recorded by different platforms can be exchanged, understood and compared. Data ownership and control ensure that students and schools do not lose control over the use of these valuable learning records simply because they were collected by private platforms.
Only when all three principles are in place can the data accumulated by AI each day be transformed from fragmented records into a continuous learning trajectory—one that helps teachers understand their students over time and provide instruction suited to their individual needs.
From Bringing AI into Schools to Achieving Personalised Learning
Hong Kong does not lack AI platforms or learning data. What it lacks is a framework that allows data to be transferred and understood, while giving students and schools a meaningful say in how those data are used.
The question, then, is how such a framework should be established. Does the government need to create a unified platform? Or can common rules be introduced progressively while preserving school autonomy and market innovation?
In our next article, we will propose a practical, step-by-step pathway at three levels—the individual school, the school sponsoring body and the education system as a whole. The objective is to ensure that learning data are not merely collected, but used well, so that the promise of personalised learning can be brought into the classroom one step at a time.
