Meet The Small Team Behind Singapore's Big AI Education Bet
In an unmarked office in the Ministry of Education annex on North Buona Vista Drive, a team of nine educators, two researchers, and a former software engineer have been quietly rewriting what Singapore secondary students learn about artificial intelligence. The team, led by Dr Tan Wei Ming and embedded inside the Curriculum Planning and Development Division, have spent eighteen months turning what started as a stack of optional after-school enrichment modules into a mainstream syllabus that 38,000 students will sit through this year.
Singapore is now the first country in Asia to move AI literacy out of optional and into compulsory secondary school curriculum at scale. This piece looks at how the team got there, what the syllabus actually teaches, and where the early results are landing.
What The Syllabus Actually Teaches
- What is AI doing? A working definition of generative and predictive systems, with hands-on exposure to a curated set of AI tools.
- How does AI fail? Bias, hallucination, brittleness, and the gap between model confidence and accuracy.
- Where does the data come from? A rights and provenance lens on training data, with case studies from Singapore copyright disputes.
- How should I act? Responsible use, including a clear classroom protocol on disclosure and citation.
The most distinctive feature is the third question. Singapore's syllabus is one of the few in the region that pushes adolescent students into thinking about training data and licensing, rather than leaving the topic at "AI is biased." The team brought in IPOS lawyers to co-design the case-study materials.
The Pedagogical Bet
Dr Tan and the curriculum team made a bet early on that the syllabus would not be a prompt-engineering course. That bet has held even as private tuition centres in Singapore now market AI prompting classes to anxious parents. The MOE position is that prompting is a transient skill, while the underlying critical reading of AI output is durable.
The practical effect in classrooms is that students spend more time critiquing model outputs than producing them. A typical lesson begins with a teacher-supplied AI-generated answer to a humanities or science question, and the class is asked to identify what is wrong, what is right, and what cannot be verified without external sources.
We do not want students who can prompt. We want students who can read a model output and tell you why it is wrong.
How The Team Got Here
The original mandate came from the late 2024 review of the National AI Strategy 2.0. MOE was asked to deliver an AI literacy programme that could move from pilot to mainstream within two years. The team initially proposed a three-year ramp, and was given two.
The acceleration has had visible costs. The syllabus is heavier than the team would have ideally written, and several of the case studies are still being piloted live in classrooms because there was no time for full pre-trial. The compensating decision was to over-invest in teacher training. The team built a fellowship programme at the National Institute of Education that runs over six months and includes a written assessment and a classroom observation. 240 teachers have completed the fellowship to date, against a 1,800 target for 2027.
What The Early Data Shows
MOE released a first round of evaluation data this month, drawing on assessments from the 56 pilot schools. The results are encouraging in narrow ways. Students who completed the full Sec 1 module showed a 32% improvement in identifying AI-generated misinformation and an 18% lift in citing sources correctly when using AI tools. Gains in actual subject-matter learning attributable to AI literacy are smaller, around 4% to 6%, and statistically noisy, which the team had expected.
The more interesting finding is teacher confidence. Teachers who completed the NIE fellowship reported a 41% increase in confidence handling AI questions in non-Computing subjects, including English Literature and Social Studies. Roughly 73% of fellowship completers also reported using AI tools in their lesson preparation, up from 22% before the programme. That is the leverage MOE was looking for.
| Programme Element | 2024 | 2025 | 2026 (current) |
|---|---|---|---|
| Schools running curriculum | 0 | 12 (pilot) | 56 |
| Students reached | 0 | 4,800 | 38,000 |
| Trained AI Pedagogy fellows | 0 | 72 | 240 |
| Required curriculum hours | 0 | 6 (optional) | 12 (compulsory) |
| Pre-loaded AI tool licences | 0 | 3 | 9 |
How Singapore Compares Across Asia
Singapore's rollout is the most ambitious mainstream deployment in Asia. South Korea has a similar literacy framework but rolled it out as guidance rather than a compulsory module, and uptake varies sharply by province. Japan's MEXT is piloting an AI literacy course in 240 schools but has not yet committed to nationwide compulsory rollout. India's MoE has prioritised teacher training over a fixed national syllabus, and Hong Kong is moving toward an integrated curriculum at slower pace.
The outlier is China, where AI literacy is treated as part of a broader programme of "AI plus education" and is woven into multiple subjects rather than taught as a discrete module. The two approaches will be worth comparing in 2028 once Singapore's first cohort completes Secondary 4.
For readers tracking the AI tutors push in Asia and Singapore's AI governance, the school curriculum is the public-facing layer of a deeper national strategy that links workforce, model governance, and citizen literacy.
What Could Go Wrong
The rollout is not without risks. The most acute is teacher capacity. Singapore's secondary teachers are already balancing PSLE preparation pressure, mother tongue requirements, and Character & Citizenship redesigns. Adding twelve hours of AI literacy is meaningful even if MOE has been careful to embed it within existing modules rather than as a new subject.
A second risk is curriculum drift. The syllabus needs to update at least annually to keep pace with model capability changes, and MOE's traditional curriculum revision cadence is much slower than that. The team has built a six-monthly review cycle into the programme, but that cycle is unproven.
What Other Asian Education Systems Should Watch
The Singapore approach is exportable, but only partially. The country's small scale, single national system, and well-funded teacher training infrastructure make rollout easier than it would be in a federal system. Indonesia, the Philippines, or India would face significantly different operational challenges. The pieces that travel are the syllabus design, the case study library, and the teacher fellowship structure. The pieces that do not travel are the speed of central decision-making and the budget per student.
Frequently Asked Questions
Are primary school students included?
Not yet. The current rollout covers secondary 1 to 4. MOE has begun scoping a primary-school version, but the team consider primary-age AI literacy a separate design challenge that needs a different pedagogical approach.
Which AI tools do students use in class?
MOE has procured nine AI tool licences for classroom use, including Anthropic Claude, Microsoft Copilot for Education, Google Gemini, and four Singapore-built tools. Students do not use general consumer ChatGPT during the classroom hours.
Is parental opt-out allowed?
No. AI literacy is now compulsory under MOE policy, similar to other Character & Citizenship Education modules.
How is the programme funded?
The USD 9 million programme spend is separate from the wider National AI Strategy 2.0 budget. Schools receive an annual top-up grant for AI tools, and teacher fellowship costs are funded centrally through NIE.