As AI enters APAC classrooms, what skills should students still practise?

Sir Ken Robinson said that in a talk that has circulated for years. His point was about education's limits: a school cannot determine what a young person becomes. It can only create the conditions in which they develop.

A student now has another kind of 'teacher' in the classroom: AI that can produce a plausible first answer before the student has worked out what they think. The challenge for education is not simply teaching students how to use that tool. It is deciding what they still need to struggle through themselves.

Aaron Thean, provost of the National University of Singapore, put the distinction more directly. He told The Straits Times in August that “the core business of a university is really to develop the human, not to develop the AI.”

AI access for students in Singapore, Vietnam, and Malaysia

From 31 August 2026, NUS students, faculty and staff receive ChatGPT Edu in a university-managed workspace. The university says conversations in that workspace are not used to train OpenAI’s systems. From academic year 2026/27, first-year undergraduates take THE1008 Applied Generative AI: From Prompting to Evaluation. Graduates are meant to complete at least two AI courses, and every major is to include a course on how AI is changing that profession. Students, NUS says, should learn to “critically interpret AI-generated results.”

The 2026/2027 NUS first-year guide describes fluency in AI, the standard all students are meant to reach through compulsory courses, as: “I can use AI in my major, and I know how to evaluate it.”

In the NUS announcement dated 11 August, Thean called AI tools “powerful amplifiers of human intellect, creativity and judgement,” and said the NUS–OpenAI collaboration, through ChatGPT Edu, should develop “critical thinking and ethical discernment.” The Straits Times also described NUS introducing in-house assistants such as a Socratic coach that prompts with questions rather than answers, and features that ask students to spot gaps in AI-generated answers.

In May, in remarks reported by Channel NewsAsia, Desmond Lee, Minister for Education, and Minister-in-Charge of Social Services Integration, said that from 2027, students in universities, polytechnics, the Institute of Technical Education and continuing education would get baseline AI competencies inside their discipline. The skills include what AI can and cannot do, and “learning with AI in ways that deepen understanding rather than replace thinking.”

The Straits Times reported on 30 August that Singapore Management University now grades some writing-and-reasoning students on their ability to evaluate AI outputs. Michelle Lee, who teaches sustainable marketing there, releases an essay question two weeks early and allows AI for research, then a 15-minute in-class write with no materials. “AI will not be able to tackle it or answer in a perfect way,” she said. She also wants “a problem scenario where there are no easy answers.” Pitch Perfect, a five-minute live pitch followed by peer questions, is 30% of a management-communication grade.

Vietnam is also introducing AI in schools on a wide scale. On 20 August, Vietnam News Agency reporting carried by VnExpress said the Ministry of Education and Training had issued an AI-education framework. Each class gets 12 periods of core AI content a year, with optional clubs and projects on top. The framework is organised around four competency areas: human-centred thinking, AI ethics, techniques and applications, and system design. Human-centred thinking and ethics are described as the guiding principles.

Teaching, the report said, will emphasise experience, practice, problem-solving and age-appropriate project-based learning. Ethics may be taught through discussion, debate, role-play and case studies. Primary students meet simple applications, daily-life uses, personal data and copyright. Lower secondary students use tools to make digital products and solve learning problems. Upper secondary students explore, design and improve simple tools through projects, and move on to analysis, critical evaluation and solution development. VietnamNet reported on 18 August that a draft National AI Transformation Strategy aims at 10 million workers with basic AI skills in Vietnam by 2030. The school framework is one part of that wider push to build AI capability.

On 28 July, at the AI Malaysia launch in Cyberjaya, Prime Minister Anwar Ibrahim said 100,000 youths aged 18 to 30 would complete modules on the Rakyat Digital platform, including AI Security, Cyber SAFE, Agentic AI, Generative AI and Cloud Untuk Rakyat, and then receive a free three-month subscription to popular AI apps from 31 August. He called it the “democratisation of access to AI,” and said the transformation should not widen urban-rural or rich-poor gaps. The announcement names a literacy sequence and a subscription.

Those are three different bets. Singapore has included evaluation in a compulsory university course and has started to keep live hours in which a student must think. Vietnam has put practice and critical evaluation on the school timetable. Malaysia has put generative AI in 100,000 young hands. The development problem sits underneath all three.

You cannot check what you have never practised

AI can produce a plausible essay, a summary, a block of code or a set of slides before a student has worked out what they think. While it is useful, it is also where the problem starts.

To know whether the answer is any good, a student needs something to compare it against. They need to know the subject, recognise when something does not make sense, and be confident enough to question an answer that sounds convincing. That confidence does not come from a prompt. It comes from doing the work.

A course still has to leave a student with the knowledge, thinking and capabilities it promised. At SMU, Xavier Goh, 21, sees the trade-off in the shift towards live presentations. “Shifting from written final exams to live presentations,” he said, “changes how we use AI rather than reduce our reliance on it.”

That is probably right. A student can use AI to prepare a pitch, just as they can use it to prepare an essay. But when the slides are closed and someone asks a question they did not expect, the student has to think, draw on what they know and work out an answer.

The three things students still have to practise

The issue is not whether students will use AI. It is whether they will still get enough chances to do the thinking themselves. Three skills stand out: knowing when to question an answer, deciding what matters, and working out what to do when things go wrong.

1. Discernment

Knowing when an answer is wrong, incomplete or simply not good enough.

The NUS first-year guide puts the responsibility on the student: AI outputs can be inaccurate, biased or superficial, and students remain responsible for the quality of what they submit. The tools should be treated as collaborators in learning, it says, “not shortcuts around it.” SMU is taking that idea into assessment, with some students graded on their ability to evaluate AI outputs. Vietnam's upper-secondary framework also names critical evaluation as a skill students should develop.

The same problem is beginning to appear in the workplace. The Digital Education Council, in a 21 August note with Google.org, argues that generative AI is now producing analysis that executives once received from layers of human workers. When the machine can produce an answer almost instantly, the valuable skill becomes knowing what to question.

2. Judgement

There is another skill that comes before checking an answer: knowing whether you are asking the right question.

Speaking at the Singapore Computer Society's Tech3 Forum on 20 August, Josephine Teo, Minister for Digital Development and Information, said that as AI gets better at producing “code, analysis and content”, the premium will rise on “the judgement to know what needs solving”, why some solutions are more likely to succeed or fail, and how to take the work forward.

That is harder to teach because there is often no single correct answer. Michelle Lee's “problem scenario where there are no easy answers” is a good example. AI can help a student think through the options, but it cannot decide which trade-off matters most. That decision still belongs to the person who has to live with the consequences.

3. Recovery

Knowing what to do when the first answer fails.

Real work rarely ends with the first answer. Something does not work, a customer asks an unexpected question, a colleague spots a flaw, or the solution turns out to be wrong. Someone still has to work out what to do next.

That is what SMU's Pitch Perfect is trying to preserve. A student gives a five-minute live pitch for a grieving-process app, then has to respond when a peer asks whether it really caters to a user's unique emotional needs. Vietnam's upper-secondary projects are designed around a similar idea, giving students opportunities to experiment, solve problems and improve their work.

While generative AI can produce the first version, it cannot stand in the room when someone asks, “But what if this doesn't work?” The ability to question the answer, decide what matters and keep going when the answer fails may be the part of learning that becomes more valuable, not less, as AI gets better.

After school, the same design choice

The gap does not close at graduation. Teo told the same forum that fresh graduates already worry about entry-level jobs. Judgement of what needs solving, she said, often comes with experience; “there is just no shortcut.” How to broach a subject and rally people is part of that adult work.

The measure of an AI-enabled education may therefore be less about how often students use AI than what they are still asked to do without it. A student still needs to recognise a bad answer, decide what matters, defend a judgement and recover when the first attempt fails. These are not simply skills for surviving an AI-enabled workplace. They are part of becoming an adult who can think for themselves.

That brings the debate back to Robinson’s farmer. Education cannot determine what a young person will become. It can decide the conditions in which they grow. As AI takes over more of the work of producing answers, perhaps the responsibility of schools and universities is to be more deliberate about protecting the experiences through which students learn how to question, judge and think.

What to watch

• How schools teach with AI, not just about AI. Watch what students are asked to do with the tool, and what they are still expected to do themselves, as well as whether institutions assess evaluation, judgement, problem-solving and independent thinking as AI becomes part of everyday study.

• How AI companies work with educators. Watch whether collaborations between AI companies and educational institutions help educators build the human skills that AI cannot provide, including critical thinking, judgement and the ability to question an answer.

• What happens when today's AI-enabled students enter work. Whether employers start asking graduates to work with the assistant closed and not only to show they can use one.

---