A Tool, Not an Oracle: Using AI Without Surrendering Judgement

Can we borrow the power of a tool without slowly becoming dependent on it? What happens to judgement when we stop exercising it? And when a machine speaks with such fluency and apparent confidence, how do we remember that fluency is not the same thing as wisdom?

These are not merely technical questions. They are questions about the relationship between human beings and the instruments they create, a relationship that has always carried risk alongside its extraordinary promise. The arrival of artificial intelligence into everyday intellectual life is, in many respects, the latest chapter in a very old story: the story of what we gain when we extend our reach, and what we quietly give away in the process.

Let us be clear about what artificial intelligence, in its current and most familiar form, actually does. It recognises patterns across enormous quantities of text and produces responses that conform to those patterns with remarkable statistical sophistication. It does not understand, in any meaningful sense. It does not have experience, curiosity, or conscience. It does not know what it does not know, and this last point is perhaps the most consequential of all. A tool that cannot signal its own limits is a tool that requires the user to supply those limits from outside. That responsibility belongs to the human being holding it.

Yet there is something seductive about fluency. When a piece of writing arrives fully formed, in measured sentences, with apparent breadth of reference, the mind of the reader, or in this case the person who prompted it, can slip into a comfortable passivity. The cognitive effort that ordinarily accompanies thinking, the friction of uncertainty, the slow building of an argument from first principles, all of this is bypassed. And it is precisely that friction, that effortful engagement, which builds the capacity we call judgement. To bypass it repeatedly is not a neutral act. It is a form of gradual erosion.

This is not a counsel of abstinence. The question is not whether to use these tools but how to use them without surrendering the very faculty that makes their output meaningful. A hammer does not decide where the nail should go. A calculator does not determine whether the calculation serves a good purpose. Similarly, an artificial intelligence system does not and cannot decide whether its output is true, appropriate, ethical, or wise in a given context. Only the person using it can do that, and only if that person remains actively in the role of judge rather than passive recipient.

What does it mean, in practice, to keep that authority firmly human? It begins with posture. The person who approaches an AI system as a starting point rather than a finishing point is in a fundamentally different relationship with the tool than the person who treats its output as a draft answer to be approved or rejected wholesale. To use the tool well is to interrogate its output, to ask what it has missed, to notice where it has been vague, to bring to bear the knowledge and values that the system cannot possess. This requires intellectual confidence, and intellectual confidence is something that must be cultivated and protected.

There is also the question of attribution and responsibility. When a decision is made, when a piece of work is produced, when advice is given, the question of who is accountable does not dissolve simply because a machine contributed to the process. Responsibility remains with the human being who chose to use the tool, who selected what to accept, and who put the result into the world. The danger is not merely practical but moral. If we allow ourselves to think of AI as the author of our conclusions, we have not only deceived others; we have begun to deceive ourselves about the nature of our own agency.

Rhetoric and critical thinking, the disciplines at the heart of the work of organisations like this one, are particularly vulnerable to misuse in this context. Rhetoric is the art of constructing persuasive, well-ordered argument suited to a particular audience and purpose. That suitability, that attunement to context, is irreducibly human. A machine can generate text that resembles a well-constructed argument, but it cannot feel the weight of a room, cannot sense where scepticism lies in a particular audience, cannot know the specific history and values that make one framing right and another tone-deaf. The rhetorician must supply all of this. To hand the task entirely to a machine is to produce something that looks like rhetoric but performs none of its real work.

Critical thinking faces a subtler danger. The purpose of critical thinking is not merely to evaluate the arguments of others but to develop the habit of self-examination, to become aware of one is own biases, assumptions, and blind spots. This development happens through struggle. It happens when we sit with a difficult question long enough to feel its edges, to notice where our reasoning becomes circular or where we are relying on authority rather than evidence. If an AI system resolves that struggle too quickly, supplying an answer before the productive discomfort has done its work, the development does not occur. We may walk away with a plausible answer and a slightly weaker mind.

None of this means that these tools are without genuine value. They can surface perspectives one had not considered, help to organise complex material, accelerate the early stages of research, and assist those whose relationship with written language involves barriers that the tools can helpfully lower. These are real benefits, and they deserve acknowledgement. The issue is one of calibration and of maintaining a clear sense of where the authority lies.

So perhaps the question to hold onto is this: when we have finished using the tool, is the thinking more ours or less ours than when we began? If the answer tends toward less, then something important is being quietly traded away, something that, unlike processing power, cannot simply be recovered by switching to a better model. The oracle, after all, does not think for us. That remains, as it has always been, entirely our own responsibility.