Taimi Journal
AI Can Help Children Learn — But Only If It Doesn't Give Away the Answer
AI is often talked about as either a miracle cure or a threat. The research tells a calmer story: AI can support learning if it gives hints instead of ready-made answers. In this article, we walk through what the research says
— and seven questions every parent can use to evaluate any learning app.
- Intelligent tutoring has been studied for decades — it can support learning when it guides step by step.
- Without guardrails, AI can harm learning if it gives answers too easily.
- A good learning tool gives hints, questions, and feedback — it doesn't do the child's work for them.
- Parents don't have to be home tutors; their most important role is to supervise, encourage, and set boundaries.
- At the end of this article, you'll find seven questions to evaluate any learning app.
How do you know if AI actually helps?
Your child is doing math homework. They get stuck. Do you open an AI app?
If you do, you're faced with a question that sounds small but changes everything: do you let the AI give the answer — or guide the thinking?
Most parents haven't considered that there's a difference. That AI can be by a child's side in two completely different ways: either ready to take a shortcut, or ready to help the child pause and think for themselves.
This exact difference has been studied — a lot, and for a long time. The research gives a surprisingly clear direction on what kind of AI supports learning and what kind can actually hinder it. In this article, we walk through what we know — and you'll come away with seven questions to evaluate any learning app, whether it's Taimi or something else.
AI as a learning tool is not a new idea
Many people were surprised that AI could even be used for teaching. Researchers weren't. Intelligent Tutoring Systems — ITS — have been studied for decades, and one of the field's key meta-analyses was published back in 2011.
Intelligent tutoring can support learning — when it guides step by step
Kurt VanLehn's meta-analysis gathered decades of research on computer-assisted tutoring. The results were strong: at their best, intelligent tutoring systems reached a similar range of effectiveness to one-on-one human tutoring.
This doesn't mean any AI works. Effectiveness depends entirely on how the system is designed — whether it gives room for thinking or shortcuts too much.
What this means for family life: The question isn't "is AI good or bad?" It's: what kind of guidance does the tool give? Does it help the child think, or does it do the thinking for them?
— VanLehn, 2011
A more recent systematic review (Létourneau et al., 2025) confirmed the same direction: ITS effects were positive but modest. At the same time, the review highlighted something important — ethical questions remain largely underexamined.
Hint versus answer — the whole difference
At the heart of ITS research is a simple insight: computers don't get tired. They can ask the same question again, give immediate feedback, and adapt to a child's skill level.
But the same qualities that make a system effective also make it dangerous — if it isn't designed right.
When AI gives the answer too easily, practice may not turn into learning
In a recent experimental study, students used generative AI for math practice. When AI was used without constraints, students' performance improved during practice — but on an independent test, they performed worse than students who hadn't used AI at all.
The AI made practice too easy. It gave ready-made answers instead of guiding thinking. The child moved through the exercises — but their thinking didn't strengthen.
What this means for family life: If a tool gives a ready answer the moment a child faces difficulty, it may look helpful but bypass learning entirely. Even a good AI tool can be harmful when it is used in the wrong way.
What this means for Taimi: In Taimi, the AI is designed to give step-by-step hints — not to solve the problem for the child.
— Bastani et al., 2025
Fortunately, there's also research on what happens when this principle is followed. Pardos and Bhandari (2024) studied ChatGPT use in math instruction specifically so that the AI gave step-by-step help, not solutions. The result: students who received this kind of support performed better than the control group.
In learning research, this is called formative feedback — feedback whose purpose isn't just to tell whether an answer was right or wrong, but to help the learner adjust their thinking for the next attempt (Shute, 2008). Good AI support isn't an answer dispenser. It should give just enough support for the child's thinking to continue — but leave the solving work to the child.
What does the difference look like in practice? Imagine a 10-year-old facing the multiplication problem 6 × 4.
Parents don't have to be home tutors
At this point, many parents think: "Well, I can help my child with homework. Isn't that the same as tutoring?"
Research points in another direction.
Unsolicited help can inadvertently turn a parent into a home tutor
A large study examined the effect of parental homework help on learning outcomes in a dataset of 2,176 students. Especially intrusive or unsolicited homework support was linked to weaker math development.
It's not that parents don't know how — they do. And it's not that parents don't want to help — they do. It's about the dynamic: the parent-child relationship is different from the tutor-student relationship. The parent gets frustrated. The child gets frustrated. Homework becomes a negotiation.
What this means for family life: Parents don't have to be teachers. Their most important job is to be a safe, encouraging, and present adult.
What this means for Taimi: In Taimi, the parent supervises and encourages — but doesn't have to become a home tutor. A well-designed tool can take on the patient, step-by-step guidance that is often hard to provide in the middle of family life.
— Park et al., 2023
When 10-year-old Ella does math homework and gets stuck, a parent can get frustrated as quickly as the child. "Just look at it, this is so easy" — and the situation escalates. An AI tutor, by contrast, can ask: "What have you tried so far?" without frustration. The parent can be the parent. The computer can be the patient feedback-giver.
What do children themselves want?
One of the most interesting studies in this field isn't directly about AI. It's about children — and what they themselves say they need to support their learning.
In a recent co-design study, 10–12-year-olds identified three needs: organization, focus, and fun (Amaefule et al., 2025). Not "make it easier." Not "give us ready answers." But "help me stay organized, concentrate, and keep going."
These three — organization, focus, fun — are exactly the things a good learning environment provides. They're things a well-designed digital tool can support. Children aren't looking for shortcuts. They're looking for structure.
Seven questions to evaluate any learning app
Finnish education authorities frame AI in education in a similar way: it should support the learner's own thinking, be transparent, and leave adults with a clear role (Finnish National Agency for Education & Ministry of Education and Culture, 2025).
From these principles — and from the research described above — comes the following checklist. Use it to evaluate any app meant to support learning.
- Does the tool give hints before answers? A good tool guides the child to think for themselves — it doesn't offer a ready answer immediately.
- Does the child have to explain their own thinking? If the tool asks "how did you arrive at that?" and not just "was the answer correct?", it's likely supporting learning.
- Can the parent see how AI is being used? Transparency means the guardian knows when the child has used AI and for what purpose.
- Is the child's data minimized and protected? The tool shouldn't collect more data than necessary, and data must not be used for child profiling.
- Does the difficulty level adapt without the tool doing too much for the child? A good tool challenges appropriately — it doesn't lower the bar too far or carry the child too much.
- Does the tool encourage trying, not just getting the right answer? Feedback like "good effort, shall we try again together?" is better than just "wrong / right."
- Is the role of AI clear to the child? The child should know whether they're talking to AI or a human, and understand that AI is not an authority but a tool.
These questions aren't opinions. They're practical applications of research, feedback pedagogy, and the principles that guide how Taimi's learning features are being designed.
What we don't yet know
The research gives us a strong foundation for what kinds of principles work. But it doesn't tell us whether a specific app works — not ours or anyone else's.
We don't yet know whether a child learns math from the specific app a child opens today. We don't know whether skills learned in an app transfer to everyday life. We don't know how AI affects learning motivation over months or years.
That's why we don't want to promise too much. Taimi is not a scientifically proven learning intervention. It's a family tool being built on research, designed around three principles:
- Step-by-step guidance — not ready-made answers
- Preserving the child's own thinking
- The parent as supervisor, not tutor
Sources
- VanLehn, K. 'The Relative Effectiveness of Human Tutoring, Intelligent Tutoring Systems, and Other Tutoring Systems.' Educational Psychologist, 2011. Meta-analysis.
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. & Mariman, R. 'Generative AI without guardrails can harm learning: Evidence from high school mathematics.' PNAS, 2025.
- Park, D., Gunderson, E.A., Maloney, E.A., Tsukayama, E., Beilock, S.L., Duckworth, A.L. & Levine, S.C. 'Parental Intrusive Homework Support and Math Achievement: Does the Child's Mindset Matter?' Developmental Psychology, 2023.
- Pardos, Z.A. & Bhandari, S. 'ChatGPT-generated help produces learning gains equivalent to human tutor-authored help on mathematics skills.' PLOS ONE, 2024.
- Létourneau, A., Deslandes Martineau, M., Charland, P., Karran, J.A., Boasen, J. & Léger, P.M. 'A Systematic Review of AI-Driven Intelligent Tutoring Systems in K–12 Education.' npj Science of Learning, 2025.
- Shute, V.J. 'Focus on Formative Feedback.' Review of Educational Research, 2008.
- Amaefule, C.O., Britzwein, J., Yip, J.C. & Brod, G. 'Children's Perspectives on Self-Regulated Learning: A Co-Design Study on Children's Expectations towards Educational Technology.' Education and Information Technologies, 2025.
- Finnish National Agency for Education & Ministry of Education and Culture. 'AI in education — legislation and recommendations.' 2025.