By Guillermo Fossati, PhD, psychologist and education researcher
Artificial intelligence can help students write better essays, solve problems and produce sophisticated explanations in seconds. But a more convincing answer does not necessarily mean better learning. Education is not simply about producing correct answers. It is about developing the knowledge and abilities that allow students to understand, reason and solve problems independently. Using AI to apply a skill we already possess is not cognitively equivalent to using it to replace the process through which that skill is acquired.
Three students, three learning experiences
Imagine three students are asked to analyse a complex text. The first reads it independently, struggles with its arguments and gradually constructs an interpretation. The second develops an initial interpretation, then uses AI to challenge it and suggest alternative perspectives. The third asks AI to summarise the text and write the analysis.
All three may submit satisfactory assignments. Yet their learning experiences differ. The first practices independent reasoning; the second combines reasoning with assistance; the third may obtain an excellent result without exercising the abilities the assignment was intended to develop.
AI does not inevitably undermine learning. Its value depends on which cognitive processes it supports and which it replaces.
When assistance comes too early
Reading, writing, remembering and reasoning are not merely ways of completing tasks. They are activities through which abilities develop. A child learning to write needs opportunities to construct sentences; a student learning mathematics needs to work through problems.
Not every difficulty should be eliminated. Some difficulties are part of learning itself. This does not mean leaving students without guidance: timely explanations and feedback are essential. But assistance that benefits an experienced learner may bypass necessary practice for a beginner.
Four questions for teachers
Before introducing AI into an assignment, educators should ask four practical questions:
1. What ability is this activity intended to develop? If students are learning to construct arguments, they need to identify evidence and develop conclusions themselves. AI can later help them examine weaknesses in their reasoning.
2. What should students attempt independently first? In mathematics, students might try a problem before asking AI to explain another solution. In writing, an initial draft gives them something of their own to revise. The sequence should reflect their prior knowledge.
3. Could AI challenge rather than replace their thinking? Instead of requesting a completed analysis, students could ask AI for counterarguments, then assess those suggestions against the text and defend their own interpretation.
4. How will we know learning has occurred? A polished AI-assisted assignment cannot establish what a student has learned. Later, without assistance, can the student explain the reasoning, retain relevant knowledge and apply it to an unfamiliar problem? These checks distinguish immediate performance from understanding and transferable learning.
These questions matter for teacher preparation and policy. Teachers need more than training in operating AI tools: they need to judge when assistance advances learning and when it replaces essential practice.
The measure that matters
AI can broaden access to explanations, feedback and educational resources. But its success should not be measured only by adoption rates, faster task completion or more polished assignments.
The purpose of educational technology should not be to make students increasingly dependent on more capable systems, but to help them become increasingly capable themselves.
Education must prepare students to work intelligently with AI while preserving their ability to question it. The ultimate measure of AI’s educational value is not what students can produce with it, but what they become capable of understanding and doing independently.
