What Should Kids Still Learn When AI Can Give Them the Answer?
The true test of understanding is what a child can explain, teach, present, perform, or discuss live.
When a tool can hand a child the answer in seconds, what is left to learn?
A kid can photograph a worksheet. A student can paste a chapter and get a summary. Another can request an essay outline or a lab write-up before the thinking has started. The question is older than any chatbot. If the answer is easy to get, what should a child still be able to do?
Start by separating two things we often blur. Performance is the finished page, the fluent draft, the easy practice set. Learning is a lasting growth in what a student can do later, without the same help, in a new situation. Cognitive psychologists drew that line long before generative AI. Soderstrom and Bjork (2015) showed that short-term performance is a poor guide to durable learning. What feels smoother in the moment can leave students less prepared later. What feels slower can do the opposite.
Two Things We Often Blur
Performance
Now, with help
- The finished page
- The fluent draft
- The easy practice set
Learning
Later, on their own
- Recall
- Application
- Ability to Explain
A kid photographs a math worksheet, and the answers appear in seconds. The page is finished. That is not the same as learning. Put the device away. Ask them to explain one story problem in their own words.
AI is unusually good at the first of those things. It can make homework look done, a draft look polished, practice look easy. None of that means the knowledge is now the student’s.
A finished page is not proof of learning. Put the device away and ask your student to explain one problem in their own words. That is where you see what they know.
Thinking Depends on Knowledge
So what should kids still learn? Not a list of future-proof job skills. Children still need knowledge and skill in long-term memory. That is what makes thinking possible.
In a widely cited essay for American Educator, Daniel Willingham (2006) argued that factual knowledge is not just raw material for critical thinking. It is what lets students understand language, make inferences, hold ideas in mind, and see the deep structure of a problem. Working memory is limited. Knowledge lets us chunk information and free up space to reason. Without it, attention is spent just holding the pieces.
Working Memory Has Limited Room
Still working out 7 × 8
- 7, 14, 21…
- 28, 35, 42…
- 49, 56
- Which count was I on?
No room left for the story problem.
Knows 7 × 8 = 56
- 56
- What the story asks
- Why it’s multiplication
- Does the answer make sense?
Room to explain.
If a child still has to reconstruct 7 times 8, that fact eats the attention needed to explain the story problem. The finished answer can hide the gap. Practice the fact, then ask them to explain. That is not empty drilling. It leaves room to think.
A classic demonstration is the baseball study by Recht and Leslie (1988), which Willingham discusses in that essay. Junior high students read a passage about a half inning of a baseball game. Students who knew a lot about baseball understood and remembered it better than stronger readers who knew little about the game. Background knowledge was not a nice extra. It was part of reading itself.
The Baseball Study
The same logic shows up in math, science, and writing. Thinking is not a generic muscle. Students think well about what they know well.
A student asks sharper questions about a novel they actually read than about a book they only saw summarized. A summary might spark a question. It rarely supplies the scenes and language that make the best questions possible. If you want better questions, start with the book, not the summary.
Students ask sharper questions about a book they read than one they only saw summarized. Thinking is not a generic muscle: students think well about what they know well.
If They Can Get the Answer, Can They Teach It?
The cost of an answer has dropped. The value of being able to explain one, notice when it is wrong, and do the work without help has not. It has risen. Students will meet fluent, confident, and sometimes incorrect output.
Richard Feynman is often misquoted with this important nugget: if you cannot explain something simply, you do not understand it. While the quote is incorrect, the learning nugget is spot on.
That is the test we need now. If AI can produce the answer, the proof of understanding shifts to what a child can explain, teach, present, perform, or discuss live. Explaining is not a nicety at the end of the assignment. It is how we see whether the thinking happened.
When AI can produce the answer, the proof of understanding is what a student can explain, teach, present, or discuss live. Ask for that more often, not less.
Learning science points the same way. When students are asked to explain ideas to themselves as they study, their understanding improves. That is the self-explanation effect, shown by Chi and colleagues (1994) with eighth graders reading a text about the circulatory system. It was a lab study, not a classroom one. The practical implication still holds: putting an idea into your own words is part of learning it.
A student pastes a chapter and gets a summary in seconds. That does not show they can teach the claim. Close the book. Have them tell you the claim in a few sentences, then check.
The same test holds when the product looks more sophisticated. A student can get an essay outline or a lab write-up in seconds. Ask them to present the argument, or discuss the result, before any tool writes it up. The outline is not the thinking. The thinking is whether they can make the case.
Students may use these tools, just as they use calculators and search. After a tool helps, have them explain a source or a solution in their own words, with the chat closed. Can they perform the idea live?
Feynman saw the gap between fluent recall and real understanding in his own teaching. In Surely You’re Joking, Mr. Feynman!, he wrote that students he met “had memorized everything, but they didn’t know what anything meant.” A finished answer, then and now, can hide that gap.
The Same Tool, Two Ways
Used well
The practice still happens
- A hint
- A counterexample
- A practice question
- A second version to compare
Used poorly
The practice gets skipped
- A finished product
- Nothing the child can explain next
Used well, a tool can offer a hint, a counterexample, a practice question, or a second version to compare. Used poorly, it skips the practice that produces learning. If the goal was a finished product, the tool may have helped. If the goal was a more capable thinker, look at what the child can explain next, on their own.
How Parents and Educators Can Approach This
If AI can produce the answer, the proof of understanding is live: what a child can explain, teach, present, perform, or discuss without the tool. The finished page is not the learning.
As a parent, or working one-to-one
Pick one task this week and ask for the live version. Put the device away. Have them explain a story problem, teach you the claim of a chapter, or talk through a source. Start with the student’s voice, not the tool’s output. Try: Teach this to me before we look anything up. After any hint, have them present a similar item with the tool closed. You do not have to police every tool. You need one regular moment when the thinking has to be performed at the table.
In the classroom
You cannot see every homework shortcut, and you should not build the week around catching them. Build in one live look at the skill: a short oral explanation, a discussion, a presentation, or a performance of the idea in class. That is how a whole class still gets evidence of understanding.
The Research
- Soderstrom, N. C., & Bjork, R. A. (2015). Learning versus performance: An integrative review. Perspectives on Psychological Science, 10(2), 176–199. doi.org/10.1177/1745691615569000
- Willingham, D. T. (2006). How knowledge helps. American Educator, 30(1), 30–37. aft.org/periodical/american-educator/spring-2006/how-knowledge-helps
- Recht, D. R., & Leslie, L. (1988). Effect of prior knowledge on good and poor readers’ memory of text. Journal of Educational Psychology, 80(1), 16–20. doi.org/10.1037/0022-0663.80.1.16
- Chi, M. T. H., de Leeuw, N., Chiu, M.-H., & LaVancher, C. (1994). Eliciting self-explanations improves understanding. Cognitive Science, 18(3), 439–477. doi.org/10.1207/s15516709cog1803_3
- Feynman, R. P. (1985). “Surely You’re Joking, Mr. Feynman!”: Adventures of a Curious Character (R. Leighton, Ed.). W. W. Norton. wwnorton.com/books/Surely-Youre-Joking-Mr-Feynman
More Posts
When Help Becomes a Crutch: AI Tutoring and Overreliance
An AI tutor can raise a child’s practice scores and still leave them weaker without it. Effective help leaves the student equipped to do the next problem unassisted.
Nov 18, 2026 · 4 min readAI and LearningThe Foundations That Still Matter: Literacy and Numeracy
Most young children do not yet have their own AI tools. The work now is to build reading, number sense, and the habit of explaining what they know, so they are literate and discerning before that access arrives.
Nov 17, 2026 · 6 min readThe ScienceLearning Still Takes Effort: Why Easy Is Not the Same as Learned
If a tool makes the work feel smooth, it is tempting to conclude that learning has happened. The feeling is often wrong.
Nov 15, 2026 · 5 min read