Writing and thinking

28 August 2026

Writing expresses thought. Before we write we have to think and as we write we generate thoughts. A natural process. People think and write in different ways. In fiction, writers can be described as architects or gardeners. It is a useful distinction because it gets at two different ways of thinking. Gardeners just write. Initially they may not know about the characters or where the story will go. They might know the end from the beginning, but not have any clear idea how things will unfold along the way. A single glimpse in their mind’s eye of the opening scene is enough to begin. One thing leads to another. Architects plan carefully before writing anything then write sections that fit together.

Until recently, how people write reflects how they think. In the past, when we read something, we knew we were directly reading the thoughts of a person. Engaging with their ideas. Alan Bennett captures this experience well in The History Boys:

The best moments in reading are when you come across something – a thought, a feeling, a way of looking at things – which you had thought special and particular to you. Now here it is, set down by someone else, a person you have never met, someone even who is long dead. And it is as if a hand has come out and taken yours.

The best writing is the thinnest veneer between the writer’s mind and your own. Knowing and being known. Now we do not know. Nobody can reliably identify lines written by artificial intelligence. We have some idea, an instinct perhaps. There are unreliable tools which ignite accusations. Every word written in the past couple of years is under suspicion now.

People ask why coding with large language models is considered perfectly acceptable whereas writing with them is anathema. Both are about expressing thought, the difference is to whom. The purpose of code is to express human thought to a machine in a way it understands. Programming languages themselves are a layer of abstraction that allows people to interact with machines more easily than using machine code. Linus Torvalds has compared the use of LLMs to compilers, just another tool to improve efficiency. He makes the point that no programmer writes machine code any more. The entire purpose of all the layers of abstraction between the thought in a human mind and binary code is to communicate with a machine in a way it can comprehend. LLMs are just one more layer.

Written language has a different purpose. It is about communication between people, but this is a broad category. Expectations are what matters. When reading a book or an article, there is an expectation that you are engaging directly with the thoughts of the writer – the point Bennett is making in his play. Here it is seen as betrayal to use an LLM because in doing so the writer is presenting a version of their ideas filtered through and shaped by a machine that has been trained on the thoughts of millions of other people. The writer is not only robbing the reader but themselves of the work they would have produced, had they not used an LLM. That is lost forever. Writers who do this never know what they would have written without it, where their own thoughts would have led them as they wrote.

There are exceptions. Nobody expects a novelist to avoid using an LLM when writing an email to their accountant. They could if they wanted to but it would be a waste of time. There is an argument that any use a writer makes of an LLM in any aspect of their life is bound to affect how they write because our minds are not so easily compartmentalised, but I do not buy it. Sweating over each word of every routine email is not beneficial to anyone, especially when the email itself is likely to be read by an LLM when it arrives. Expectations are different, the writer’s accountant would not be outraged in the way their reader would. This is a machine in its proper place, lifting these burdens from our shoulders, not adding to them.

The strange thing about using LLMs to write is that the writer is also denying themselves of the pleasure of writing. It causes their thinking to atrophy. If we can write without thinking, or with only sufficient thought to roughly draft a few prompts and press return, what happens to our ability to think? The fitness of our minds will suffer eventually and our writing will become the dull echo of a million others mediated by a machine.

An alternative peril lies for those who do not use LLMs to write. Who have the ability to write well and may judge harshly those who fall down. There is something wrong with the sheer unrelenting fury that accompanies criticism today, damaging both critic and their quarry. Writing well is not easy, even for those who have the natural talent. It still takes work and rework. Some who do not possess such a gift will be tempted to take a shortcut. They should not be destroyed for succumbing to it.

Others use artificial intelligence to help them write in a language not their own. People don’t always have a native speaker on hand to review what they write, to offer suggestions. And even if they do know someone, how easy is it to ask? This path still has risk for those who decide to take it. Maybe they never develop the confidence to write without support and in the end they are robbed of what they would have become, just as native speakers are. Maybe their thoughts end up being shaped in an unhealthy way as they develop a dependence. There is not a simple answer. In the end it is better for non-native speakers to understand what they sacrifice in the long run, try without and for others to accept imperfection, to admire those who are determined to express themselves in another language.

We should also be wary of the tools that people use to judge who is guilty. It is a strange thing that people who object so strongly to the use of machines in writing would rely on one to confidently point the finger. Machines are made by people and people are fallible. The temptation to stand in judgement is strong, but we should resist. Especially if we are relying on another machine to make our judgement for us.