top of page
Search

The Thinking We Must Not Outsource

Writer: Kirsty Nunn
Kirsty Nunn
Sep 17
6 min read

If artificial intelligence can increasingly produce the answers, what becomes of the intellectual journey through which we learn to ask the questions?


I teach Computer Science, which means I have spent much of my professional life encouraging students to make computers do things. There is a certain irony, therefore, in finding myself increasingly concerned about everything we should not allow computers to do for us. Artificial intelligence has introduced extraordinary possibilities into education. It can explain complex concepts, generate examples, offer feedback, translate information and support learners who might otherwise struggle to access particular forms of knowledge. These capabilities deserve serious attention.


But beneath the practical conversations about efficiency, assessment and academic integrity lies a more fundamental philosophical question.


What happens to human learning when the production of an answer no longer requires the intellectual effort that once preceded it?


The question is not whether artificial intelligence is useful. It plainly can be. The question is what we believe education is trying to develop.


The difference between an answer and an understanding

Imagine a student encountering a difficult programming problem. They examine the requirements, attempt a solution, discover an error, reconsider their assumptions and gradually develop a working algorithm. The process might be frustrating. It might involve several unsuccessful attempts. There may be moments when the student is convinced that the computer has developed a personal grudge against them. Eventually, the program works.


But the educational value lies in considerably more than the functioning code. Through the process, the student may have developed abstraction, decomposition, logical reasoning, debugging strategies and the capacity to tolerate uncertainty. They have not merely produced a solution. They have changed something about their ability to solve future problems.


Now imagine that the student asks an AI system to generate the program immediately. The resulting code may be elegant, efficient and entirely correct. Yet the cognitive transformation is not guaranteed. The student possesses an answer. Whether they possess the understanding is another matter. This is the distinction that should concern us.


Education is not primarily a process of acquiring finished products. It is a process through which engagement with problems changes the person doing the thinking.


The temptation of frictionless learning

Much of modern technological design aims to remove friction. We make systems faster, interfaces simpler and workflows more efficient. In many areas of life, this is unambiguously desirable. Few people wish to return to the administrative pleasures of manually calculating an entire spreadsheet. But educational friction is more complicated. Some forms of difficulty are unnecessary and exclusionary. Confusing instructions, inaccessible materials and avoidable administrative barriers should be removed wherever possible. Other forms of difficulty are intrinsic to learning. Struggling to distinguish between competing explanations, constructing an argument, retrieving knowledge and identifying an error can all contribute to intellectual development when appropriately supported.


The challenge is distinguishing productive cognitive effort from needless frustration.

Artificial intelligence makes this distinction increasingly important because it allows students to bypass processes that may be central to the learning objective.

If the objective is to evaluate a persuasive argument, an AI-generated example may provide an excellent starting point for critical analysis. If the objective is to learn to construct an argument independently, allowing AI to produce the argument may undermine the very capability being developed. The technology has not changed. The educational purpose has.


That is why I am increasingly sceptical of general statements suggesting that AI either belongs in every classroom or should be excluded from education altogether.

Both positions begin with the tool. We should begin with the learning.


Human agency in an age of intelligent systems

Self-determination theory, developed through the work of Edward Deci and Richard Ryan, offers a useful psychological perspective. Their research identifies autonomy, competence and relatedness as important psychological needs associated with motivation and wellbeing. These concepts provide an interesting lens through which to examine educational AI.


Artificial intelligence might support competence by offering accessible explanations, examples and feedback. It might support autonomy by enabling learners to pursue questions at an appropriate pace or investigate interests beyond the immediate curriculum. But it might also diminish both if students become dependent on a system to make decisions, construct responses or judge the quality of their thinking. There is a significant difference between using a tool that expands one's capabilities and using a tool that gradually replaces the exercise of those capabilities. The distinction is not always obvious.


A student who feels more successful because AI consistently produces polished work may experience an immediate increase in confidence without a corresponding increase in independent competence. Conversely, a student using AI to examine misconceptions, explore alternative approaches and test their own reasoning may develop greater independence. The critical question is not simply how much AI a student uses. It is whether that use strengthens or weakens their capacity to act without it.


The danger of confusing fluency with truth

Generative AI also presents an epistemological challenge: how do we know what we know? Language models can produce extraordinarily persuasive explanations. Their fluency can create an impression of reliability that is not always justified. Yet a convincing explanation is not necessarily a correct one.


Kasneci and colleagues (2023) discuss both the opportunities and limitations of large language models in education, emphasising critical thinking, fact-checking and awareness of potential errors and bias. UNESCO's guidance on generative AI similarly advocates a human-centred approach to educational implementation. These concerns lead us towards an educational responsibility that extends beyond learning to operate AI systems. Students need to understand how to interrogate them.


They should ask where a claim originates, what evidence supports it, which assumptions are embedded within it and what alternative interpretations might exist. They need the confidence to disagree with a response that sounds authoritative. This is particularly important because the authority of AI is unusual. It does not necessarily derive from expertise in the conventional human sense, yet its output can imitate the language of expertise remarkably well. We risk producing students who can obtain explanations instantly but lack the conceptual foundations required to evaluate them. In that situation, access to information expands while intellectual independence contracts.


A different ambition for AI in education

Perhaps the most valuable educational applications of AI are not those that eliminate thinking, but those that provoke more of it. Imagine using AI as a debating partner that challenges a student's assumptions rather than writing their essay. Imagine a programming assistant that asks diagnostic questions before revealing a solution. Imagine students comparing competing AI-generated explanations and investigating why they differ. Or consider a classroom in which students are deliberately given an incorrect AI response and asked to identify its weaknesses using their own subject knowledge.


In each case, AI becomes a catalyst for intellectual activity rather than a substitute for it. Of course, this requires careful pedagogical design. It also requires teachers to retain the professional judgement to decide when technology adds value and when it should remain absent. Not every lesson needs an AI component. Not every difficulty requires technological intervention. And not every opportunity to accelerate a task should be taken.


What should remain irreducibly human?

There is a temptation to define the future of education around the capabilities that artificial intelligence does not yet possess. But this creates a moving target. As systems become more capable, our understanding of educational value should not have to retreat continually into whatever technological territory remains unconquered. Instead, we might define education positively.


We want young people who can think critically, act ethically, exercise judgement, develop relationships, understand complexity and take responsibility for their decisions. We want them to experience curiosity, intellectual discovery and the satisfaction of developing genuine mastery. We want them to become people capable of contributing meaningfully to a world they will inevitably help shape. Artificial intelligence may assist in this process. It should not determine its purpose.


The challenge for educators is therefore not to protect every traditional task from automation. Some tasks deserve to change, and others may disappear entirely.

It is to identify the experiences through which human capabilities develop and ensure that those experiences remain central to education.


In a world where increasingly sophisticated answers are available almost instantly, perhaps the greatest educational responsibility is to protect the slow, demanding and profoundly human process of learning how to think.


References


Comments

 

© 2025 by The Education Architect

 

bottom of page