AI Did My Kid's Homework. Now What?
Your Grade 8 child has a math assignment due tomorrow. They open ChatGPT, type in the problem, and copy the answer into their notebook. Homework: done. Learning: none.
This is happening in classrooms across Ontario right now. Teachers know it. Parents are starting to notice. And the conversation around AI in schools has gotten loud enough that some boards are banning it outright while others are scrambling to figure out what "appropriate use" even means.
But here's the thing: the problem isn't AI. The problem is what kind of AI, and how it's being used.
There's a real difference between a tool that does the work for a student and one that helps a student do the work themselves. Understanding that difference is the most important thing a parent can know about AI and their child's education right now.
Why getting the answer is actually the problem
When a student looks up an answer — whether in a textbook answer key, from a classmate, or from an AI — and copies it down, something specific happens: the problem disappears, but the confusion doesn't.
The assignment is done. The gap in understanding is still there.
This matters more in some subjects than others. In math especially, concepts build directly on each other. A student who doesn't understand why you need a common denominator when adding fractions — but copies the correct method — will hit a wall the moment fractions appear in a new context. The formula worked for the homework. It won't work for the test, or for the next unit that assumes the concept is solid.
Researchers call this "surface learning" — the appearance of understanding without the underlying structure. AI tools that give instant answers accelerate surface learning dramatically, because the friction is gone. The friction — the moment of not knowing, of trying and failing and trying again — is actually where learning happens.
When you remove all the friction, you remove the learning.
What students think they're doing vs. what's actually happening
Most students who use AI to get answers aren't trying to cheat in any meaningful sense. They're doing what feels efficient. They have six subjects, two after-school activities, and a test tomorrow. The path of least resistance is to get the answer and move on.
The problem is that they often genuinely believe they've learned something. They read the AI's explanation. They understood the words. They copied the steps. It felt like learning.
But reading an explanation and being able to produce one yourself are completely different cognitive processes. Understanding a worked example and being able to work a new problem from scratch are completely different skills.
This gap — between feeling like you understand and actually being able to apply — is one of the most well-documented problems in education research. It has a name: the illusion of knowing. AI tools that provide complete, fluent, confident answers make the illusion stronger than ever. The explanation sounds so clear. Of course you understand it.
Then the test arrives, the context changes slightly, and the understanding that felt so solid turns out to be a borrowed one.
The difference between a bad AI tutor and a good one
Think about the best teacher or tutor you ever had. Did they just tell you the answer when you were stuck?
Almost certainly not. They probably did something like this: asked you what you'd tried so far. Pointed at the part of the problem you hadn't looked at carefully. Asked you a question that made you think. Waited — sometimes uncomfortably — while you worked it out. Only gave you a hint when you were genuinely stuck, and even then, the hint pointed you toward the answer rather than providing it.
Good tutors guide. They don't do.
The reason this works is cognitive: when you produce an answer yourself — even if you needed guidance to get there — you own it in a way you don't when you copy it. The mental effort of working something out is what makes it stick.
This is what makes most AI tools genuinely problematic for learning: they're designed to be helpful, and "helpful" in most AI contexts means giving you what you asked for as completely and clearly as possible. A student who asks "how do I solve this equation?" gets a full, clear, step-by-step answer. The AI is doing its job perfectly. And the student is learning almost nothing.
What good AI tutoring actually looks like
The alternative isn't to remove AI from the equation — it's to use AI the way a good tutor would.
A good AI tutor, given the same equation problem, would respond differently. Something like:
"Before I help you with this, tell me what you've tried so far."
Or:
"What do you think the first step should be here?"
Or, if the student is genuinely stuck:
"Look at the left side of the equation. What do you notice about it? What operation might help you isolate x?"
None of those responses give the answer. All of them move the student toward it by engaging their own thinking. The student still has to do the cognitive work. The AI is the scaffold, not the crane.
The difference in outcome is significant. A student who works through a problem with guidance — who has to produce each step themselves, even when prompted — retains the concept. A student who reads a worked example retains the format but not the understanding.
This is also more like what Ontario teachers are trying to do in the classroom. The 2020 Ontario math curriculum explicitly emphasises reasoning and problem-solving — not just procedure. A student who has only ever been given the steps hasn't developed the reasoning the curriculum is building toward.
What parents can do at home
You don't need to police your child's AI use minute by minute. But a few habits make a real difference:
Ask them to explain it to you. After homework is done, pick one question and ask your child to walk you through how they solved it — not what the answer is, but why they did each step. If they can explain it, they understood it. If they stumble, they probably borrowed the answer.
Make the rule about process, not tools. "Don't use AI" is an instruction that's easy to get around and hard to enforce. "You can use any tool you want, but you have to be able to explain your answer to me" is a standard your child internalises — and it's actually the right standard for learning.
Notice the pattern, not the instance. One copied homework assignment is fine. A habit of getting answers without building understanding will show up in tests, in later grades, and eventually in a loss of confidence when the scaffolding disappears.
Talk about it openly. The most useful conversation isn't "are you using AI to cheat?" It's "do you feel like you actually understand this stuff?" Many students have a quiet sense that they're not really learning, even when their homework looks fine. Creating space for that conversation is more useful than surveillance.
Why this matters more in some grades than others
The compounding effect of surface learning tends to hit hardest at particular transition points in the Ontario curriculum:
Grade 4–5: The shift from concrete arithmetic to abstract reasoning. Students who haven't fully consolidated multiplication, fractions, and place value start struggling with everything built on top.
Grade 7–8: Algebra, proportional reasoning, and multi-step problems all require genuine conceptual understanding. Surface-learned procedures from earlier grades fail here.
Grade 9: The jump to secondary school exposes every gap that got papered over in elementary. This is when parents first hear "my child was always good at math" followed by a troubling report card.
At each of these points, a student who has been getting answers without building understanding hits a wall. The wall feels sudden. It's usually been coming for a long time.
The design question every AI education tool should answer
When evaluating any AI tool for your child's learning — whether it's a tutoring app, a homework helper, or a subject-specific tool — there's one question worth asking:
When my child gets stuck, does this tool give them the answer, or does it help them find it themselves?
The answer to that question predicts more about whether your child will learn from it than any other feature, subject coverage, or price point.
This was the design question at the centre of building Polo, the AI tutor inside Patharo. The goal was never to be the most helpful AI in the sense of giving the clearest answers fastest. It was to build something closer to a patient tutor — one that asks questions before giving hints, gives hints before giving steps, and gives steps before giving answers. One that treats the student's confusion as the starting point for learning, not as a problem to be eliminated.
The research on tutoring consistently shows that the most effective tutors spend most of their time asking questions, not giving answers. Building an AI that behaves that way is harder than building one that answers everything clearly. But it's the only version that actually helps a student learn.
The bottom line
AI is not going away. Ontario students are going to use it — for homework, for studying, for writing, for math. The question isn't whether they use it but whether they use it in a way that builds their understanding or replaces it.
The tools that give answers are easy to use and feel helpful in the moment. The tools that guide — that ask questions, that wait, that point rather than tell — are more demanding. They're also the ones that leave a student more capable after using them than before.
That's what a tutor does. That's what good AI should do too.
Polo, Patharo's AI guide, is designed to help Ontario students work through problems — not work through them for them. If you're curious what that looks like in practice, you can try a free plan with your child.