Raising an AI-Ready Generation Might Start With Less AI

Why “back to basics” may be Singapore’s most strategic AI move

A student told me something that should stop any educator mid-sentence.

I asked if they used AI. They said yes.I asked what for. “School work.”I warned that AI can be wrong.They shrugged: “I don’t care if it’s right or wrong… the homework was done.”

That one line captures the real educational risk of this decade.

It isn’t that students will use AI. They will.The risk is that we quietly redefine “learning” as producing an answer, instead of building understanding.

And once a system trains students that output matters more than reasoning, we should not be surprised when they stop caring whether the output is true.

The question Singapore should ask

Singapore is moving to integrate AI and deepen EdTech use in schools. The official intent is thoughtful: age-appropriate use, pedagogy first, and guardrails to avoid dependence.

But intent doesn’t automatically translate into outcomes—especially when the technology is designed to be frictionless.

So the real question is not “Should schools use AI?”

It is: What kind of mind must a student develop before AI becomes a helpful tool rather than a cognitive crutch?

What the evidence is whispering (and sometimes shouting)

We already have decades of evidence that friction is not always bad for learning.

Handwritten note-taking, for example, has been shown to outperform laptop note-taking on conceptual understanding—partly because handwriting forces selection and summarisation rather than transcription.

On labs and experiments, the evidence is more nuanced than most people think: virtual tools can be as effective as physical ones for conceptual learning overall, but physical experience matters when sensory interaction is part of the concept. That’s a quiet warning against replacing “real-world interaction” with screens by default.

And on generative AI, early research is raising a pointed concern: when the tool does the thinking, the brain may simply do less of it. A widely discussed MIT Media Lab preprint reports lower neural engagement in ChatGPT-assisted writing tasks (though it is not yet peer reviewed). Other research, including survey-based work, suggests people may reduce critical thinking effort when they trust AI outputs more.

None of this argues that AI is “bad.”

It argues that timing and incentives matter.

The workforce paradox: the future needs validators, but the pipeline is training acceptors

In the labour market, employers consistently emphasise analytical thinking and rising demand for AI-related skills.

But there is a paradox hidden in that trend:

• The economy needs more people who can validate AI outputs.

• Yet AI makes it easy to become someone who simply accepts outputs.

If a student grows up using AI as a shortcut before they develop strong internal models of reasoning, checking, and explanation, then the “AI user” we produce may not be an “AI supervisor.” They may become an “AI consumer.”

That’s the broken reinforcement loop: outsourcing thinking early reduces the capacity to supervise thinking later.

A different strategy: “AI later, literacy now”

If Singapore wanted a real competitive advantage, it might be this:

Raise a generation that can use AI—without needing AI to think.

That doesn’t require rejecting technology. It requires sequencing it.

Here’s one practical way to frame the policy and pedagogy:

Stage 1 (Primary): Build the human operating system

Optimise for:

• reading depth and comprehension

• handwriting and summarisation

• mental maths and estimation

• scientific play and physical experimentation

• attention, patience, and “productive struggle”

In this stage, the goal is not efficiency.The goal is cognitive infrastructure.

Interestingly, some countries are moving in this direction by reducing screens, strengthening access to textbooks, and restricting distractions like phones—less as nostalgia, more as a literacy and attention strategy.

Stage 2 (Lower Secondary): Add tools, but teach verification as a skill

Here, introduce AI in controlled, explicit ways:

• “AI as a drafting partner, not an answer machine”

• students must check claims, cite sources, and explain reasoning

• assignments require a process trail: what they asked, what AI returned, what they accepted/rejected and why

This is where “critical thinking” stops being a slogan and becomes an assessed behaviour.

Stage 3 (Upper Secondary/JC/Poly): Treat AI like a powerful but fallible instrument

At this stage, students can handle:

• prompt discipline

• model limitations and failure modes

• hallucination detection

• bias and incentives

• triangulation and evaluation

• decision-making under uncertainty

This is where you produce people who can lead AI, not follow it.

The uncomfortable truth: education becomes what we measure

That student who said “I don’t care if it’s right or wrong” is not just a tech story.

It’s an incentives story.

If school rewards the appearance of completion over the quality of reasoning, students will optimise for completion. AI simply makes that optimisation easier.

So the question becomes:

Are we prepared to redesign assessments so that students are rewarded for thinking, not just producing?

Because if we don’t, we may successfully create an AI-enabled school system… that produces less AI-ready adults.

A final thought

In FYT’s work, we often say that good analytics starts with the question, not the tool—and value is created end-to-end, with humans accountable for judgement.

Education is the same.

The strategic goal is not “more AI in classrooms.”The strategic goal is more capable humans in a world with AI.

And paradoxically, the fastest way there may involve something that feels slower:

More books. More handwriting. More labs. More explanation.More time spent learning how to be sure—before learning how to be fast.