Math101 · Independent Ontario learning library
AI is not one educational treatment. A carefully designed tutor that asks questions, limits answer dumping, and uses verified course content is different from an open chatbot completing an assignment.
Current evidence supports a conditional answer:
AI can improve learning and access in some settings. Benefits are more likely when the system scaffolds thinking, gives accurate feedback, and prepares students to perform without it. Poorly designed or poorly used AI can create errors, dependence, privacy risks, and weaker independent learning.
What a promising classroom experiment found
A 2025 randomized study in an undergraduate physics course compared a carefully designed AI tutor with an active-learning class experience over two lessons. The 194 students using the AI tutor showed larger learning gains in less time and reported stronger engagement and motivation.
Important limits:
- It was one university physics context.
- The intervention covered two lessons, not a full degree.
- The tutor was intentionally scaffolded and constrained.
- The result does not apply automatically to a generic chatbot prompt.
The study demonstrates possibility, not a universal replacement for teachers.
What a large high-school math experiment warned
Research involving nearly 1,000 high-school mathematics students found that access to a general GPT-style interface improved performance while students had the tool. But some students then performed worse when the tool was removed. A more guided tutor design reduced that negative effect.
That pattern matters: assisted performance is not the same as learning.
If AI produces the solution path, the student may submit correct work without practising retrieval, method selection, or error recovery.
What broader reviews suggest
Recent meta-analytic work reports positive average effects of generative AI on learning outcomes, but studies vary in subject, age, duration, tool, comparison condition, and quality. Fast-changing products make long-term evidence difficult.
Treat the average as a reason to design and test carefully—not as proof that every AI feature helps.
AI is strongest at
- generating additional examples at a chosen level
- rephrasing an explanation
- asking Socratic questions
- giving immediate low-stakes feedback
- simulating practice conversations
- helping plan and organize work
- supporting accessibility when privacy and accuracy are managed
AI is weakest or riskiest at
- acting as the final authority on factual or mathematical correctness
- completing graded work the student is expected to author
- inferring a learner’s emotional, disability, or family context
- giving medical, legal, or financial decisions without professional review
- replacing the sustained human relationship of teaching and mentoring
- protecting sensitive data when students paste private records
A five-part test for an educational AI feature
- Learning target: What independent skill should improve?
- Cognitive work: What thinking remains with the student?
- Accuracy: How are content and calculations checked?
- Transfer: Can the student later perform without the tool?
- Safety: Are privacy, bias, academic integrity, and escalation addressed?
Why KamranBot is designed around hints
Math101’s goal is not maximum message length. KamranBot should use the current course, ask for the student’s attempt, reveal one step at a time, render mathematics correctly, and encourage a no-tool retry. A direct answer can still be given when requested, but it should show the reasoning and invite verification.
That design follows the strongest lesson from current evidence: guardrails change what students do with the tool.
Can AI replace a teacher or tutor?
It can automate some explanation, practice, and feedback, but current evidence does not justify treating it as a complete replacement. Teachers and tutors diagnose context, build relationships, design assessment, notice disengagement, and exercise professional judgment.
Does using AI make students less intelligent?
Not automatically. The effect depends on the task. Using AI to receive a hint and then solving independently can support learning; repeatedly outsourcing the target skill can reduce practice and create dependence.
Is AI safe for children?
Safety depends on the product, age, supervision, privacy, content controls, and use. Families and schools should review terms, avoid sensitive data, and provide human oversight. UNESCO recommends a human-centred and age-appropriate approach.
