There is a particular discomfort that appears when a machine gives us a sentence we might have written ourselves. The sentence is fluent. It says something close to what we meant. Perhaps it is even clearer than our first attempt. For a moment, we feel relief. Then comes the unease: if this was generated by a system trained on the words of others, and if I accept it, revise it and use it, in what sense is the thought still mine?

The question feels new because the technology is new. But the anxiety is older than artificial intelligence. AI has not invented our fear that ideas may not fully belong to us. It has made that fear visible.

We often imagine originality as a kind of purity. A real idea, we think, should come from within us, uncontaminated by imitation, borrowing or influence. Yet very little in human life begins this way. We learn language by hearing and repeating others. We inherit metaphors before we invent them. We absorb gestures, rhythms, methods, arguments and expectations long before we call anything our own.

This does not make thought fake. It makes it human.

As a hydrologist, I often think about this through models. A hydrological model can simulate the response of a river basin, estimate runoff from rainfall, or help us understand how a catchment might behave under changing climatic conditions. Some models are physically based, some conceptual, some statistical, and increasingly some are data-driven or assisted by machine learning. Whatever their form, however, they do not simply reveal the river as it is. They are built from assumptions, measurements, calibration choices, historical data, simplifications and uncertainty.

When a model produces a hydrograph, the curve does not speak for itself. A peak flow might look convincing, but the researcher still has to ask: Were the input data reliable? Was the calibration period representative? Did the model capture snowmelt, groundwater, land-use change or reservoir regulation properly? Is the result meaningful for this basin, this season and this decision?

We would not say that the model understands the river, or that it is responsible for the interpretation. The model produces an output. The researcher must decide what that output means.

This distinction matters for AI-assisted writing too. Generative AI can produce language that looks like thought. It can imitate tone, structure, argument and style. It can offer paragraphs that seem already polished. But current generative systems cannot stand behind what they say. They do not know why one sentence matters more than another. They do not carry the consequences of an argument. They cannot be embarrassed by an error, persuaded by a moral concern or held accountable for a misleading claim.

AI therefore unsettles us not simply because it imitates. Humans have always imitated. It unsettles us because it imitates without experience, commitment or responsibility.

Yet this does not mean that any contact with AI destroys authorship. The stronger question is not, “Did influence enter the work?” Influence always enters. The better question is: what did the human being do with it?

In scientific modelling, the human contribution is not the production of every number by hand. It lies in framing the problem, selecting the method, understanding the limits, comparing alternatives, interpreting uncertainty and deciding what can responsibly be claimed. A model result becomes part of scientific knowledge only when it is judged, situated and made answerable.

Something similar is true of writing. If a person asks an AI system to generate an essay and publishes the result without reflection, their contribution is thin. They have prompted production, but they have not necessarily authored meaning. By contrast, a person may begin with a question, use AI to test formulations, reject what is false or shallow, rewrite what is useful, add experience, change the structure and finally take responsibility for the argument. In that case, the human role is much stronger.

The difference is not cosmetic. It is the difference between outsourcing thought and using a tool to think more carefully.

This is why originality should not be understood as freedom from influence. Such freedom has never existed. A child does not become less human because their first words were copied. A musician does not become less original because they learned scales before improvising. A scientist does not become less creative because they use inherited equations, established methods or computational models. Human creativity begins inside forms we did not invent.

But it does not end there.

Originality happens after influence. It appears in selection, emphasis, misreading, adaptation, resistance and recombination. It appears when inherited material is reorganised by a particular mind, in a particular situation, under a particular responsibility. A thought becomes ours not because nothing entered it from elsewhere, but because we have worked on what entered – and because we are prepared to answer for the result.

This also means that not all borrowing is innocent. AI systems are trained on human-made materials, and questions of consent, copyright, attribution and compensation remain serious. To say that creativity has always involved influence is not to say that every use of another person’s labour is acceptable. Transformation does not erase obligation.

But the ethical difficulty of AI should not push us back into a false myth of pure originality. If we insist that a truly human idea must come from nowhere, then AI will make every act of writing feel contaminated. Worse, it will blind us to the way human culture has always worked: through transmission, repetition, revision and responsibility.

The hydrological model offers a useful analogy because it reminds us that tools can extend thought without replacing judgement. A model may calculate faster than a person. It may detect patterns that would otherwise remain hidden. But the scientist must still ask what the model cannot know: what matters here, what has been excluded, what uncertainty is acceptable and who may be affected by the decision.

In language, AI forces the same question with greater intimacy. The output is not a graph or a number; it is a sentence that may sound like us. That is why it feels so disturbing. It enters the space where we usually locate the self: voice, expression, judgement and thought.

Perhaps this disturbance can be useful. It can push us towards a more demanding idea of authorship. To call something mine should not mean that it came from an untouched interior. It should mean that I understand something of what I have inherited, that I have transformed it with care, and that I am willing to take responsibility for where it goes next.

AI has not killed originality. It has exposed the weakness of a myth we should have abandoned long ago. Originality was never about starting from nothing. It was about what happens after influence: the human act of turning what has been received into something that can be answered for.

Artificial Intelligence, Creativity, Psychology, Science