On 8 September, the headline went around Singapore the way it always does.
First in reading.
Second in mathematics.
Second in science.
Second in the new computational problem-solving assessment.
Out of ninety-one education systems, our fifteen-year-olds placed top-two in every single domain they sat.
We are, by the numbers, among the best-educated populations on earth.
I read the headline. Then I read the report.
| Domain
| 2025
| Rank
| 2022
| Change
| OECD avg
|
| Reading
| 535
| 1st
| 543
| -8
| 461
|
| Mathematics
| 563
| 2nd
| 575
| -12
| 463
|
| Science
| 560
| 2nd
| 561
| -1
| 482
|
| Computational Problem Solving
| 563
| 2nd
| -
| -
| 500
|
The second paragraph is where it gets uncomfortable.
Between 2022 and 2025, the gap between Singapore's strongest students and our weakest widened. Not in one subject. In all three.
Our mathematics score fell twelve points.
Reading fell eight.
And the share of students who cannot clear the basic proficiency floor, Level 2, has grown by four percentage points in mathematics since 2015.
Then there is the number I cannot stop thinking about. 91 score points.
That is the science gap between Singapore students in the top and bottom quarters of socio-economic status. PISA counts twenty points as roughly one year of learning.
Four and a half years of learning. Between two fifteen-year-olds. In the same country.
I still want to be fair to the system. Our disadvantaged students still outperform disadvantaged students almost everywhere else, and that gap of ninety-one points is not far off the OECD average of eighty-five. Singapore is not uniquely unequal.
But direction matters more than position. We are not closing. We are opening. And we are opening at the exact moment a new tool arrived in every child's pocket.
01. The tool is not neutral
Here is what makes 2025 different from every PISA before it.
This is the first cohort to grow up with a machine that will do the thinking for them.
And the report caught it. Across the systems assessed, the rate at which students read in haste, rushing a text and producing a fast wrong answer, nearly doubled between 2018 and 2025. The systems with the sharpest rises in digital distraction recorded the sharpest falls in reading.
Not because the technology is evil. Because speed is seductive, and comprehension is slow.
Give the same tool to two children and you get two different children. The tool does not decide. The child's relationship to difficulty decides.
02. Two homes
I want to be careful here, because the easy version of this story is wrong.
The easy version says wealthy families do it right and struggling families do it wrong. That is not what I see, and it is not what the data says either. There are affluent homes where the iPad raised the child. There are single-income flats where the discipline would humble you.
What I see is not a difference in character. It is a difference in capacity.
In the homes where technology works, the device was never the prize. It arrived early, it was unremarkable, and the rules around it were quiet. The child was given room to self-govern because someone had the bandwidth, years earlier, to teach them how. When AI arrived, these families looked at it and concluded, correctly, that it cannot yet teach a child mathematics. So they use it lightly. The old habits survive.
In the homes where technology fights back, the device is a luxury. It is controlled, rationed, withheld, returned. And so it becomes the most valuable currency in the house. The child does not learn self-regulation. The child learns how to negotiate. Ten more minutes. One more video.
Trust is not built. It is traded.
You know the marshmallow experiment. One now, two if you wait. The children who waited did better in life, and for forty years we read that as a story about willpower.
The follow-up studies said something else. The children who grabbed were not weak. They were rational. They came from environments where the promised second marshmallow did not reliably arrive. Grabbing was the correct response to an unreliable world.
Now, look again at the child who was told all week she could have the tablet later, and later never came. She is not undisciplined. She is adapting.
03. The part nobody says out loud
It is not only the students using AI to skip the thinking. It is the parents.
I have sat across from mothers and fathers who ran their child's homework through ChatGPT before the lesson. Not to check it. To answer it. I have seen a model's working copied into an exercise book in a parent's handwriting and presented to me as the child's own.
I understand why. They are tired. They do not remember how to draw a model. The machine is fast, free and it sounds confident.
But something is transferred in that moment, and it is not mathematics. The child watches the adult they trust most meet a hard problem and route around it. That is the lesson. Not the sum. The posture. They learnt: " I can go around things when it gets difficult."
04. Why "just limit screen time" will not save us
Every well-meaning adult in Singapore gives the same two pieces of advice. Limit the screen time. Don't let them use AI.
I have given that advice myself, for years. I will tell you honestly that I have almost never seen it work.
"Limit screen time" is the parenting version of "just study harder." Everyone agrees with it. Almost nobody sustains it. The parent finishing a twelve-hour shift does not have another fight in them at nine at night, and should not have to.
And prohibition is worse than useless with AI, because prohibition requires enforcement, and enforcement requires knowing. You do not know. The child has a phone. That war is already over.
Forbidding a tool a child can reach in private does not remove the tool. It removes you from the conversation about it.
05. Define the tool before you control it
So here is what I would offer instead. It is not a rule. It is a posture.
A child told "no iPad after eight" learns that the iPad is something to win. A child told "in this house we use this to build, not to escape" learns something about who they are. One is a boundary. The other is an identity. Only one of them survives your absence.
And for AI specifically: stop banning it. Make it useless as a shortcut instead.
Let them use it. Then ask them to explain the answer back to you, line by line. Not the result, the reasoning. Where did this step come from? Why this method and not another? What happens if I change this number?
The child who used the machine to think will pass this easily. The child who used it to avoid thinking will not survive the second question. And they will discover, on their own, that the shortcut cost them something.
You have not policed them. You have let the consequence teach them. That is the only kind of discipline that holds when you are not in the room.
06. One hour against one hundred and sixty-seven
I should tell you where my own limits are.
I meet most of my students once a week. One hour. Sometimes two.
In that hour I can teach a child to think. I can show them what it looks like to sit inside a problem without panicking. I can model the belief that difficulty is survivable.
Then they leave, and there are one hundred and sixty-seven hours before I see them again.
I have watched a child leave my table on Tuesday believing in herself, and come back the following Tuesday with the light switched off. Something happened in the week between. Someone handed her an answer instead of letting her find it. Someone decided she was not the clever one. Someone made her small in a moment I will never see.
This is the honest limitation of everything people like me do. One intentional hour does not survive a week of unchecked shortcuts. Not because the hour is weak, but because the architecture is wrong.
Temptation is available twenty-four hours a day. Guidance is available on Tuesday at four. Of course the screens are winning - we built it that way.
07. What we actually owe them
The child is the only person in this equation with no power.
She did not choose her home, the device, the rules, the rationing, or the adult who decided that thinking was optional. She inherits every decision made above her head, and then she is graded on the result.
When we celebrate a national ranking, we are celebrating her. When the gap widens, we quietly stop talking about her.
I do not think the answer is another app, another tracker, another set of rules imposed from outside. Anything imposed from outside disappears the moment the outside leaves the room.
The answer has to be something she carries.
An identity as someone who can sit with a hard thing. A belief that not-knowing is temporary. The instinct to reach for her own mind before she reaches for a machine.
That is not measured by PISA. It is not on any report card in Singapore. And it is the only thing that will still be working in twenty years.
我们排名第一。但排名，不等于韧性。
We came first. First is not the same as ready.
Keep the standards high. Look past the headline. And read the second paragraph.
Sources: OECD, PISA 2025 Results (Volume I): Future-Ready Students, 8 September 2026 · OECD, PISA 2025 Country Note: Singapore · OECD, PISA 2025 Insights and Interpretations.
