Education in an AI Age

Mr Speaker, our system of public education is, undeniably, among the best in the world. Even if we may quibble about details, the statistics speak for themselves: our most recent ranking in the OECD’s Programme for International Student Assessment (PISA) has us ranked first worldwide, and the World Bank’s Human Capital Index (HCI) likewise has us up top.

Yet this success, while not undeserved, belie an important reality: we have done exceedingly well till now, but it is highly uncertain if we will continue to do so, especially in an artificial intelligence (AI) age where traditional educational advantages are being rapidly upended by the advance of technology.

As PM Wong stated in his Budget address, our response to AI cannot be one of fear. This year’s Budget exhibits an explicit tilt toward helping our nation confront the challenges of AI, and—having spoken on the importance of AI in my Budget response two years back—I agree with this stance. Hence, I support the Budget.

But it would also be foolish to be blind to the immense transformative potential of AI to not just disrupt—but completely overturn—the competitive advantages of the Singapore model. We cannot proceed with business as usual. We need change not because our educational system is failing to deliver today, but because it is not adapted to deliver tomorrow. This is the hardest type of change, requiring foresight, wisdom, and courage. After all, if it ain’t broke, why fix it?

Where we were, and how far we’ve come, in public education

Let’s first acknowledge how far the system has come. Public mass education began in earnest a decade before independence, following the establishment of our Ministry of Education (MOE). MOE’s goal at the time was to reconcile the diverse approaches to education, given our multiracial population, and to orient schooling toward the goals of nationbuilding and workforce preparedness.

This effort was immensely successful. The overall number of years of schooling in the population went from less than 4 years in 1960, to more than 12 by 2015. With an emphasis on science and technical education—coupled with a rigorous system of sorting and tracking—public education became the backbone of our nation’s remarkable economic progress. Indeed, if you ask economists what they think was the most important factor underlying our nation’s move from Third World to First, many will likely point to our educational system.

But there are reasons for caution for the future

We must not take this success for granted, however, because the global educational landscape is rapidly changing. 

The advent of AI will mean an erosion of the slight edge in knowledge and ability that our schooling system had previously been able to confer. Emerging research already points to how AI compresses the skills gap, by raising the productivity of inexperienced and lesser-trained workers. This is the case even for college-educated professionals engaged in cognitive tasks. While this may yet fail to translate at the aggregate level, AI is nonetheless rapidly fulfilling its promise to vastly improve the efficiency of individuals.

This was the thrust of my speech in response to the budget statement two years ago. In the interim, AI has advanced at a breathtaking pace. Tasks that may have eluded AI, save for some determined and curated prompting, are now routinely answered with speed and accuracy. In the courses I teach at university, I struggle with how to best obtain separation between students, based on submitted assignments alone.

The undeniable conclusion is that our traditional model of climbing the ladder of human capital development—to become competent, reliable, and trustworthy executors and operators—is no longer likely to be a distinguishing factor.

The AI revolution will upend the way we approach education at all levels.

At the primary level, small classrooms develop curious minds

In primary schools, we still have classrooms that have as many as 40 students. This was never tenable, insofar as large classrooms affect not just the ability of our kids to receive sufficient teacher attention and absorb lessons, but also the ability of our teachers to adequately manage their pupils without excess stress.

It has become even less acceptable today, in an AI age. While some champions of technology have lauded the promise of AI to produce education customized to each learner, research in the cognitive sciences is unambiguous in concluding that children simply respond best to instruction delivered by another human, rather than a device.

Furthermore, while educational technology may confer some benefits—notably in language acquisition and problem-solving—these are frequently offset by drawbacks, such as reduced attention spans, diminished memory, and weakened critical thinking. The saving grace for such applications is when they are applied in a thoughtful and deliberate manner, which occurs when EdTech is treated as a complement, rather than substitute, for the teacher. This, in turn, calls for a better ratio of educators to students in the classroom.

I’m also keenly aware that, at these early ages, it is absolutely crucial that kids build a solid foundation for all that is to follow. This will set them up not just to succeed in their subsequent academic endeavors, but also in all sorts of life outcomes. It is therefore crucial that we do not allow our youngest learners to be left behind in their formal educational journey, reliant on supplementary tuition and parents-turned-teachers to make up for shortfalls in understanding that they would ideally have received directly in the classroom.

Much of what primary school students learn may seem like a boring reprise of known concepts. But these ideas are new, at least to the kids that are being exposed to them for the first time. And if these kids don’t grasp the basic premises they are being taught, don’t get excited about acquiring new knowledge, and resort to memorizing in lieu of true understanding, we set them up to be incurious individuals, ill-prepared in an AI-centric world where the ability to absorb and regurgitate information will no longer be sufficient to set a student apart.

Rather, children need to be excited about their educational journey, and approach school in a holistic way: as a place to not just to be fed dry academic material, but a safe space of ask all sorts of questions, to challenge prevailing wisdom and assumptions, and have fun picking up pursuits as varied as art and music and sport. And of course, if they are so inclined, languages, along with math and science. 

At the secondary level, continual assessment returns the focus to learning, not exams

Major, high-stakes exams, long a mainstay of gauging student understanding and evaluating student performance, will have less relevance in an AI age. This requires us to—pardon the pun—reevaluate how we evaluate our kids, especially in secondary school. We need a mindset shift: away from testing as a signal of competence and a means of streaming students, to a way to check progress and identify if any particular student may have fallen behind.

This need to change mindsets isn’t lost on our educators or learners themselves. MOE officials at the highest levels have acknowledged that exams aren’t the be-all and end-all. And students that embrace freedom from an exam-oriented curriculum often come to see that studying can be a joyful and meaningful affair.

Now, continual assessment has been a longstanding component of our educational system. It certainly was since I was in school (which was a long, long time ago). But our weightings reveal lip service paid to the notion of “continual.” Until recently, midyear and final exams could comprise as much as four-fifths of the total grade, which places an inordinate amount of emphasis on these markers.

So even though the Ministry formally abolished midyear exams three years ago, without an accompanying rebalancing of the weight placed on exams and tests versus other forms of assessment, the effort will remain halfhearted. Indeed, the continued presence of a relatively high-stakes year-end exam can leave students feeling unprepared when they do not have a midterm gauge. Anecdotal evidence even suggests that students and parents are making up for this change, by simply having tuition centers administer mock midterms in lieu. 

Furthermore, the PSLE and “O” Levels remain major, consequential tools for tracking students. I, along with others on both sides of the aisle of this House, have repeatedly called for a reconsideration of these examination milestones, to grant more students the chance to pursue through-train education, if they so choose. In an AI age, opportunities for late bloomers to catch up will be more abundant, which means that such flexibility becomes even more important.

The fact is, until the incentives for performing well in exams are truly changed, it is hard to expect anyone to relinquish their single-minded focus on exams.

Yet here’s the nub: there is substantial evidence that genuine continual assessment is far better in bringing about learning and understanding. Frequent but low-stakes quizzes—as much as one-a-day—can improve conceptual retention and understanding. Nor should we fear that this approach—which admittedly is commonplace across East Asia—need undermine ultimate student performance. Systems like those in Estonia and Finland, as well as the U.S. states of New Hampshire and Vermont, have relied on formative assessment and continuous feedback for years, and have produced a steady stream of excellent students.

Similarly, the evidence on tracking too early—by early, the literature suggests between the ages of 10 and 12, which is precisely when we start subject-based banding and institute the PSLE—increases educational inequality, without any clear achievement gains. While Ministry officials have ensured that subject-based banding retain mixed-form classes—consistent with research that supports how within-school tracking can help students with heterogeneous ability cope better and promote positive peer influence—there remains significant sorting between schools, which can undermine the benefits of tracking in the first place, while continuing to place undue importance on exams as a sorting tool.

To be clear, this call to shift attention away from exams and toward continual assessment decidedly does not mean a relaxation of standards. Rather, it calls for the sort of steady engagement that can keep standards consistently high. This changed emphasis is also better-aligned with fixing one of the principal challenges educators face in the modern AI-enabled classroom: students who cannot sustain enough attention and interest to complete basic tasks expected of educated individuals, like reading an entire book cover-to-cover, grasp what they have read accurately, write with accuracy and originality, and internalize mathematical skills.

At the tertiary level, we must train next-generation workers and researchers

AI will probably have the greatest impact on the way we go about tertiary education (I’ll pause here to declare that I’m a college professor). Until now, university education has had a twofold aim: to impart the workforce with modern, value-added skills—via high-quality teaching—and to infuse our economy with the latest innovation and technology, through cutting-edge research.

Both will now be upended by AI. What does it mean to transmit knowledge to undergraduates, when such knowledge can be easily accessed and repurposed at the touch of a button, is rapidly devalued within years of graduation, and can usually be obtained at a cheaper cost? What does it mean when answers to assignments—usually arrived at after extensive reflection, discussion, and struggle—may be obtained in minutes with a simple prompt?

There are already inklings of the tsunami of change that is to come. Generative AI can easily synthesize and deliver information that all but the very best undergraduates struggle to absorb, relieving the need for entry-level positions. Agentic AI can already accurately execute a wide range of tasks that new graduates used to fill, which means that firms will have a weaker incentive to hire them. 

Indeed, I believe it will not be foundational models—which have captured so much of public imagination and excitement with their clever answers to questions and seemingly human interactions—but agent-based ones that will be at the forefront of this disruption. Cognitive jobs are all vulnerable to replacement by such tools, which produce faster, often more accurate, and increasingly higher-quality output than humans. The best models are even able to do so with minimal guidance and instruction.

The superior capabilities of AI isn’t limited to intellectually intensive activities, like research or analysis, alone. AI can perform better in just about any profession where intellect and ingenuity is deployed, from vacation planning to financial planning, from video game design to building design, from data analysis to legal analysis, from diagnosing bugs in software to diagnosing cancers in humans. The aspirational careers of Asian societies—doctors, lawyers, engineers, bankers—are now all potentially replaceable by AI.

Moreover, we must not be lulled into a false sense of security by our nation’s current unemployment rate, which has yet to capture the potential displacement of jobs that could result when AI and robotics become more entrenched in the economy. The future is better inferred from the unemployment rate of new workers. Notably, while our youth unemployment rate remains considerably below those of other advanced economies, this has steadily trended up since the middle of 2024. Reports of fresh graduates struggling to secure employment are piling up.

What this means, then, is that the skills we teach in lecture halls and labs need to evolve away from rote learning and solving known problems, to leveraging what makes humans unique: soft skills like empathy, networking, judgment, creativity, and vision.

I already see it in my courses: the way to separate the weaker students from the stronger ones is frequently no longer what they submit in written assignments—something Singaporean undergrads often excel at—but by their inquisitiveness, their willingness to question prevailing wisdom, their ability to think outside the proverbial box, and their strength in bringing others together on a common task. Students must want to understand what they’re taught, not just cram yet-more information without care for the fundamentals.

It also means that our outdated approach to university entry—which is conditional on relative performance in an academically-oriented, high-stakes exam—has to be revisited. In last year’s COS, I suggested that we move toward granting entry based on satisfying absolute entry criteria. This is now even more urgent, since relative academic performance will be ever-less relevant to university success.

Reform must also percolate up to more advanced levels of tertiary education. It is natural to ask what is left for research, when—even at the PhD level—only a tiny, talented minority may still credibly lay claim to intellectual superiority and methodological sophistication?

PM Wong’s $37 billion funding for the Research, Innovation, and Enterprise (RIE2030) plan is undeniably a welcome step. And if resource constraints were the only limitation to research success, we would be well-positioned, even though—as I had previously mentioned—our national research and development (R&D) spending still lags those of comparable technologically-forward economies. 

But the reality is that scholarship is not born out of funding alone—as important as it is—and especially in an AI age, when differences in the capacity to produce research are likely to collapse. Instead, what is relevant is the sophistication of the questions we ask, and the way we approach problem-solving. For this, we must still appeal to what remains unescapably human.

Our most groundbreaking innovations have always relied on humans affording their imaginations the absolute freedom to follow wherever their ideas lead. This occurs best in open, liberal, democratic societies, which offer the best conditions for the accumulation of productive human capital and the generation of world-class scientific output. These conditions then contribute to productivity and long-run economic performance.

Our autonomous universities must stop chasing global rankings, for their own sake, but to instead focus on the genuine nature of academic enterprise. In doing so, I am certain that recognition and reward will nonetheless find their way to them.

Leaving no student behind

Sir, no matter how much we revamp our educational system—even if we were to adopt the seemingly radical steps proposed here—the reality of real-world distributions of human ability is that there will always be a group that falls on the left tail, and risks being left behind.

Yet this AI moment offers tantalizing opportunities. Teachers can tap on AI-generated resources to offer education that is a more interesting and more customized, a luxury that would have previously taken hours of work. Weaker students now have more realistic chance, than ever before, to shine in the classroom. Not only will AI compress the talent gap, it will devalue the traditional degree, which opens the door for those endowed with hitherto undervalued skills to finally stake their place in the modern economy.

This includes those that may not be as book-smart, but who are gifted in myriad other ways: those blessed with superior hand-eye coordination, or are more innately social, or exude practical intelligence, or possess a tinkerer’s mentality.

Singapore has, since its birth, been an ethically, culturally, and religiously diverse nation. Let us also embrace neurological and intellectual diversity, starting from the ground up, with our educational system. Let us allow diversity to be the core strength of our educational system in an AI age.

‍