Artificial intelligence has changed the way we learn, work, create and solve problems. With a simple prompt, an AI tool can explain a difficult concept, summarize a lengthy document, generate an essay, write computer code, create an image, analyse information and even assist with research. Tasks that once required hours of searching, brainstorming and preparation can now be completed within minutes. There is no doubt that this is a remarkable technological advancement.

However, as artificial intelligence becomes increasingly integrated into our daily lives, there is an important question that we need to ask: Are we designing AI to make humans more capable, or are we gradually designing ourselves to become more dependent on AI?

The answer may depend less on whether we use artificial intelligence and more on how we design and use it.

AI should not be viewed as the enemy of learning. In fact, it has enormous potential to transform education. The concern begins when convenience replaces the learning process itself. When a student can ask an AI tool to answer every question, write every assignment and solve every problem, there is a risk that the student may obtain the right output without developing the knowledge and skills required to produce that output independently.

This creates an important distinction between getting an answer and gaining understanding.

When Convenience Becomes Dependency

Imagine a student who is struggling to understand polymerase chain reaction (PCR). The student could ask an AI tool to explain PCR, receive a detailed explanation and read through it. That is a useful application of AI.

But imagine another student who asks AI to write an entire assignment on PCR, copies the response and submits it without understanding denaturation, annealing, extension, DNA polymerase or why the technique works.

Both students have used AI.

However, only one of them has actually used it as a learning tool.

This difference is becoming increasingly important. The OECD's 2026 Digital Education Outlook notes emerging evidence that general-purpose generative AI can improve students' performance on tasks without necessarily producing corresponding learning gains. It also warns that excessive cognitive offloading to chatbots may contribute to disengagement and weaker development of skills. In contrast, GenAI used with deliberate pedagogical purposes can support learning and skills such as critical thinking, creativity and collaboration. This suggests that the problem is not simply that students are using AI. The deeper issue is whether AI is helping students think or thinking instead of them.

From Answer Generator to Learning Partner

Perhaps the next generation of educational AI should be designed differently.

Instead of immediately giving students the answer, an AI learning system could first ask what the student already knows.

For example:

Student: "Explain antimicrobial resistance to me."

Instead of immediately generating a long explanation, the AI could respond:

"Before we begin, explain what you understand by antimicrobial resistance."

The student gives an answer.

The AI identifies what the student understands and where the gaps are. It then provides an explanation suited to that student's level of knowledge. Afterwards, it asks questions to determine whether the student actually understands the concept.

"Why do you think inappropriate antibiotic use can contribute to antimicrobial resistance?"

The student responds.

The AI evaluates the reasoning, identifies misconceptions and asks another question.

The goal is no longer simply to provide information. The goal is to build understanding.

This is the kind of shift we need: from AI that simply provides answers to AI that actively participates in the learning process.

UNESCO's AI Competency Framework for Students similarly emphasizes a human-centred approach, critical judgement of AI outputs and the development of students as responsible users and co-creators of AI. The framework includes competencies across human-centred thinking, AI ethics, AI techniques and applications, and AI system design.

What If AI Made Us Think Before It Helped Us?

A learning-focused AI could introduce what might be called productive friction.

Today, many AI systems are designed around speed. Ask a question and receive an answer. Give a prompt and receive an output.

But learning does not always happen through speed.

Sometimes, struggling with a problem, making a mistake, trying again and discovering the answer independently are essential parts of learning.

Therefore, an educational AI could deliberately delay complete answers when doing so would benefit the learner.

If a student asks:

"What is the diagnosis in this case?"

The AI could respond:

"Before I tell you, what findings in the case led you to your answer?"

If the student gives an incorrect answer, the AI could provide a clue rather than immediately revealing the correct one.

If the student still struggles, it could provide another clue.

Only after the student has attempted the problem could the AI provide a complete explanation.

This approach would make AI more like a personal tutor than an answer machine.

The Goal Should Be Understanding, Not Completion

One of the biggest problems with AI-assisted learning is that completing a task can easily be mistaken for learning.

A student may submit an excellent essay without being able to explain the arguement contained in it.

A student may submit computer code without understanding how the program works.

A student may produce a sophisticated presentation without understanding the subject.

A student may even obtain an excellent research paper without knowing whether the sources cited in it are genuine or whether the conclusions are scientifically justified. The output looks intelligent.

But the person producing it may not have developed the intelligence required to produce it independently.

This is why educational AI should measure something deeper than task completion.

It should ask:

Can the student explain it?

Can the student apply it to a new situation?

Can the student identify an error?

Can the student defend their reasoning?

Can the student solve a similar problem without assistance?

These questions are much closer to genuine learning.

AI Should Strengthen Critical Thinking

Another important concern is that AI systems can produce answers that sound convincing even when they are inaccurate, incomplete or poorly supported.

This makes critical thinking more important, not less.

A student should not simply ask:

"What did AI say?"

The student should also ask:

"Is this correct?"

"What evidence supports it?"

"What could be missing?"

"Can I verify this from a reliable source?"

"Are there alternative explanations?"

"Does the evidence actually support the conclusion?"

This is particularly important in scientific and healthcare education.

UNESCO's student AI competency framework explicitly identifies critical thinking as fundamental to meaningful engagement with AI and emphasizes that students should understand both the capabilities and limitations of AI systems.

The ability to challenge AI should therefore become part of AI literacy.

In other words, being good at AI should not only mean knowing how to prompt it. It should also mean knowing when to question it.

The Healthcare Perspective

This issue becomes even more important in healthcare.

Healthcare professionals are increasingly using technology to support diagnosis, documentation, research, patient management and decision-making. AI can reduce repetitive workloads and help professionals process large amounts of information.

But healthcare is not simply about producing an answer.

It involves judgement, context, ethics, communication and responsibility.

Consider a medical laboratory science student learning about laboratory diagnosis. Instead of asking an AI tool to provide the answer to a clinical case, an educational AI could present the case and ask the student to reason through it.

What is the likely organism?

What specimen should be collected?

Which laboratory test would you perform?

What result would you expect?

What alternative diagnosis should be considered?

What could produce a false-positive or false-negative result?

This approach encourages the learner to develop clinical reasoning instead of memorizing AI-generated responses.

The same principle can apply to nursing, medicine, pharmacy, public health and other healthcare disciplines.

AI should support the professional's thinking, not eliminate the need for professional thinking.

Research Also Requires Human Intelligence

Research is another area where AI has enormous potential.

Researchers can use AI to help identify literature, organize information, generate research questions, summarize papers and improve the clarity of writing. These applications can save considerable time.

But research is not simply the process of putting information together.

Research is about asking meaningful questions, evaluating evidence, identifying gaps, challenging assumptions and contributing something to existing knowledge.

If a researcher asks AI to generate a literature review and publishes it without checking the references, verifying the claims or reading the original studies, the research process has been reduced to text generation.

That is dangerous.

A researcher should be able to ask:

Does this source actually exist?

Does the study support this claim?

Was the study population appropriate?

Are the methods reliable?

Are the conclusions justified?

What limitations were identified?

What does existing evidence say that AI may have overlooked?

AI can accelerate research, but it cannot remove the researcher's responsibility to think.

UNESCO's guidance on generative AI in education and research therefore emphasizes a human-centred approach and the need for ethical, safe and meaningful use of these technologies.

We Should Not Try to Remove AI From Education

The solution is not to tell students to stop using AI.

That is neither realistic nor necessarily desirable.

AI is becoming part of education, healthcare, research and the workplace. Students need to learn how to use it responsibly because they will eventually work in environments where AI is integrated into everyday tasks.

The better question is:

How can we design AI so that using it makes people better thinkers rather than weaker thinkers?

This requires a change in design philosophy.

Instead of designing AI around:

Prompt → Answer → Done

we could design learning AI around:

Question → Assess → Guide → Practice → Challenge → Verify → Reflect → Understand

The difference may appear small, but its consequences could be significant.

A Future of Human-AI Collaboration

The future should not be about humans competing against AI. It should be about humans learning how to work intelligently with AI.

AI can handle repetitive tasks, process large amounts of information, identify patterns and assist with activities that consume time. Humans can provide judgement, creativity, curiosity, empathy, context, ethical reasoning and critical evaluation.

The objective should therefore not be to make humans unnecessary.It should be to make humans more capable.

Imagine an AI learning system that knows when to answer and when to ask a question. An AI that recognizes when a student has misunderstood something. An AI that refuses to simply complete an assignment when doing so would prevent learning. An AI that challenges assumptions, presents alternative viewpoints and asks the learner to defend their reasoning.

Such a system would not simply make learning easier. It could potentially make learning deeper.