What Would Make You Change Your Mind?

When AI can produce a convincing answer in seconds, the more useful question may be what happens when a learner has to defend, revise or qualify their own view.

Photo by Julio Lopez on Unsplash.

Imagine a student considering whether a new health-monitoring technology should be used with a particular group of patients. The published results look promising. The technology appears accurate and could make monitoring easier for a group that is currently difficult to monitor well. Some important questions remain unanswered.

An AI system could write a plausible recommendation in seconds. So could a capable student. If both produce a similar final answer, the answer on its own tells us little about what the student understood.

A more revealing place to look may be the point at which a learner’s position meets a challenge, and what they do next.

Start before the AI

Sequence matters. If a student reads a fluent AI answer before forming a view, that answer can become the frame for everything that follows. The student may end up editing it rather than reasoning towards a view of their own.

One alternative is to reverse the order.

The student works through the published study and records an initial position before seeing any AI output. Only then is an AI system introduced and asked the same question.

In our scenario, the student recommends using the technology. The AI argues that it should not be used until larger trials are complete.

There are now two positions, and a gap between them that the student has to deal with.

This is not a method every task should follow. There are times when students should work without AI altogether, and times when AI may be more useful for exploring a topic than for challenging a conclusion.

But this sequence has one particular advantage. The student had a view before the challenge arrived, so we can see what happens to it.

The challenge only works if there is knowledge to meet it

That gap is only useful if the student understands enough science to work with it.

AI makes the language of scepticism easy to produce: limitations, alternative views, calls for more research. A student without enough grounding in, for example, why it matters who took part in a study can repeat those objections without knowing which one actually deserves weight.

Questioning without knowledge can become a performance of its own.

So the science needs to come first. The challenge can test understanding. It cannot replace it.

Accept, reject, qualify

In the scenario, the student goes back to the study to test the AI’s reasoning.

They notice something the AI’s summary did not make clear: the study recruited mainly younger adults, while the group being considered is older and has different health needs and characteristics. That makes it less clear how confidently the findings can be applied here.

The original recommendation weakens.

But the student does not simply adopt the AI’s view. The scenario also describes a gap in how this group is currently monitored, and the AI’s position ignores it. Waiting has costs too. In this case, delaying use would leave the monitoring gap unresolved.

The student’s final recommendation is conditional: a limited, carefully evaluated trial with this group is justified. Routine use is not.

Look at what happened. The student accepted one part of the AI’s challenge, rejected another and qualified their own recommendation. Each move has a reason, and each reason can be traced back to the science.

This is different from keeping a process log or a series of drafts, which record how thinking unfolded over time. Here, the focus is on one deliberately introduced challenge: what happens after the learner has committed to a view?

What they refuse to change

What a learner keeps can be as revealing as what they change.

Once learners know that changing their mind is valued, some will produce a change to order. “I initially thought X, but on reflection…” is easy to write. It is easy for an AI to write too.

A revision on its own is not evidence of judgement. Nor, on its own, is refusing to change.

What matters is whether the learner can explain why a particular point should survive the challenge.

Here, the student keeps the monitoring gap in view despite the AI’s more cautious position. They have to explain why that consideration still carries weight. That explanation can show whether they understand the situation or are simply sticking to their first answer.

Judgement is also easily mistaken for caution. A learner who only ever adds caveats has not necessarily judged anything.

Sometimes the responsible conclusion is to act on incomplete evidence, because not acting carries its own risk. The harder question is when the evidence is good enough.

What shows judgement is the reasons: are they supported by the science, and do they hold up when questioned?

Where to look

When polished answers are easy to produce, the final answer may be the least revealing part of a piece of work.

What may reveal more is the movement between a learner’s first position and the conclusion they finally reach: what they accepted, what they rejected, what they refused to give up, and why.

A single episode can show how a learner reasoned on one occasion. Whether they can do the same across different problems is another question.

The question that brings this into view is simple to ask and hard to answer well: What would make you change your mind?

The example in this article is hypothetical. CognateUK is exploring how independent reasoning, challenge and revision might help make learner judgement more visible. This remains something to test rather than an assumed outcome.