Hong Kong, Singapore and Taiwan Take Three Different Routes to AI in Schools

Three systems, three meanings of AI in schools

The comparison, distributed through ACN Newswire, says most public argument about AI in education is between people describing different machines. One is a tutor that writes. Another locates a pupil's weak point. A third drafts lesson plans so teachers finish work earlier. According to the piece, all three exist and none does the others' job. It compares the systems using only what their operators have published.

Singapore's entry is the AI Learning Assistant inside the Student Learning Space (SLS), the national platform most schools already use. Because the feature sits in a space the ministry controls, the ministry can set who sees it and when. In 2026 the ministry named artificial intelligence one of three defining uncertainties facing the education system and set out a four-part literacy frame: learn about AI, learn to use AI, learn with AI, and learn beyond AI. The fourth limb is meant to guard against cognitive offloading.

Hong Kong's entry is a funding scheme rather than a tool. The Education Bureau reserved HK$2 billion from the Quality Education Fund for digital education. About HK$500 million funds a three-year scheme from the 2025/26 school year, giving each successful publicly funded primary and secondary school a one-off HK$500,000. By 25 February 2026, more than 960 schools had applied, around 99 per cent of the city's eligible schools.

Taiwan's entry is an adaptive learning platform that predates the current AI cycle. It performs learning diagnosis and uses conversational agents for teaching. The Executive Yuan approved a programme for the 2026 to 2029 academic years with a budget of NT$11.755 billion, organised around three dimensions and six sub-programmes.

The composition test: who refuses to write it for the pupil?

The first of three classroom tasks is an upper primary pupil stuck on how to show rather than tell, and inclined to ask the tool to write the piece. Here the systems differ most.

Singapore's ministry published a worked example from a Primary 4 English class. Pupils use the assistant with teacher-provided prompts and refine drafts against personalised feedback. The assistant will redirect students if they veer off topic, or if they want to be spoon-fed with direct answers. The comparison calls this a designed refusal and notes it is the intended behaviour, published as such.

Access rules are specific. From Primary 1 to Primary 3, schools will not assign work that requires pupils to use AI directly. From Primary 4, pupils may use educational AI tools under teacher supervision, and only tools designed for education. They first complete a basic module on AI and the SLS features. They also begin a mandatory ten-hour coding and computational thinking programme with an introduction to AI, plus two optional five-hour modules on generative AI and computer vision.

In Hong Kong, the HK$500,000 buys devices, tools, subscriptions or services at each school's discretion, within guidelines. Hong Kong Education City is building a large language model for primary and secondary teaching from 2026, with subject assistants. The Science assistant is being trialled in 2026 for teaching plans, lesson preparation, assessment items and homework. A composition tutor for upper primary is not among the published uses, so the task rests on 960 separate school purchasing decisions made between March and August 2026.

Taiwan's platform is not primarily a generative writing coach. At an October 2026 ministry forum, the president of National Taichung University of Education said general-purpose AI focuses on broad task handling and output efficiency, while the core value of educational AI lies in clear learning objectives, pedagogical principles, and safe and trustworthy design.

Finding the gap: where Taiwan leads and others stay silent

The second task is a pupil who has failed the same mathematics topic twice, where the system must find the specific gap and route practice to it. The comparison says the ranking inverts here.

Taiwan's platform has the longest documented history on this task. It was built to diagnose learning weaknesses quickly and precisely and to serve personalised, adaptive learning paths. The ministry says it improves learning efficiency and supports differentiated instruction. The new programme adds an AI tutor for students, an AI teaching assistant for teachers, an AI decision assistant for principals and an AI helper for parents, plus a learning analytics system that tracks learning behaviour and outcomes.

Singapore handles the task through supervised, education-specific tools in SLS with built-in safety guardrails, available from Primary 4. The ministry has said theoretical knowledge alone is not sufficient and that pupils need well-designed supervised hands-on experience. According to the comparison, what is not published is the diagnostic granularity: how narrow a gap the assistant will identify, and whether a pupil who fails fractions twice is routed to a different practice set.

Hong Kong's published position is the thinnest of the three, which the piece calls a structural consequence of the grant design. Schools must implement AI-assisted teaching in at least three subjects or curriculum areas, each spanning at least two levels.

What the comparison means for schools and policymakers

The main takeaway of the piece is that the same phrase hides different policy choices. Singapore controls access by age and platform. Hong Kong distributes money and lets schools choose, which produced near-universal uptake but leaves outcomes unknown until schools report. Taiwan extends a diagnostic system it has run for years.

The comparison also treats gaps in disclosure as informative. For Singapore, diagnostic detail is not public. For Hong Kong, nobody at the centre will be able to say what happened on the composition task until school reports arrive. The source text provided ends before its coverage of the third task, the teacher's lesson plan and assessment items, so that part is not covered here.