How AI is flattening how we think, write, and communicate

In Vietnam and around the world, AI has rapidly become part of the writing process in classrooms, businesses, and newsrooms. Its adoption is widespread, but its effects on how we think are still unfolding.

The tone of AI-assisted writing is familiar. The transitions are neat. The conclusion sounds reasonable. Yet something human is often missing.

The problem with AI-assisted writing is not only repeated phrases, over-tidy contrast, or inflated conclusions. Those are symptoms. The real problem is that it’s now easy to confuse fluency with real thought.

The real issue is not bad writing

Large language models generate text by drawing on patterns in language, training data, prompts, and design choices made by developers. They are very good at producing the next plausible sentence from what has been said before. If many people ask similar systems to make writing clearer, more professional or more neutral, the results will often converge.

A skilled user of AI can ask for different tones, structures, and levels of formality in writing prompts. But the default setting of AI writing is usually not specificity or domain specific language. It tends to produce sentences that most people would recognise as acceptable.

For routine work, that may be enough. A meeting summary, a polite email, a simple announcement or a first pass at translation does not need to sound original. But public argument, journalism, academic commentary, and professional advice are different. In these forms, the writer is explicitly deciding what deserves attention through emphasis, example, objection, and phrasing. Those are the parts of writing that an AI assistant may smooth away.

An AI system may treat a local or overly specific phrase as something to fix. In a policy note, research abstract or professional report, that phrase may be important. It may carry a caution, a local assumption or a way of framing responsibility that polished generic language would otherwise minimise.

AI produces fluency before the writer has done the more difficult work. When a tool can produce a clean first draft in seconds, the temptation is to let it begin. The user then becomes an editor of the machine’s first thought. Once the machine has supplied the structure, tone and order of ideas, the human author is already working within an inherited frame of reference that they might be reluctant to change.

Use AI after judgement, not before it

The important question is not whether to use AI. For many people, that question is already unrealistic. The better question is where AI enters the process.

The weak workflow is to ask AI to do the whole task first, then edit the result. For example, a prompt that reads “Write me an article about AI writing” will likely produce readable text. It may even look professional. But the central decisions have already been made, which means the flow of thought has been implicitly determined by the machine.

A stronger workflow starts before the prompt. With the same intention as the above example, the writer should first write an outline, decide the claim, choose the example, identify the reader, and the desired flow of thought.

At that point, the AI tool can be useful. It can identify weak arguments, unclear transitions in the author’s thought, missing context, and sentences to be made clearer without losing detail.

This is a practical rule for ensuring original thought and structure are explicitly prioritised over aesthetic fluency.

How tools and institutions can play a role

Tool design matters too. AI systems should help users test, compare, and revise their own judgement, not only produce polished output. But the first discipline belongs with the user. Do not ask the machine to supply the judgement you have not yet made.

Meanwhile, universities, newsrooms, and workplaces should not treat all AI assistance as cheating or all AI output as acceptable. The meaningful distinction is between support and substitution. For example, AI support with editing, critique, translation support, and structural feedback can strengthen the writer’s work. But outsourcing the first judgement will likely weaken it.

Vietnam does not need to copy Silicon Valley’s assumptions about productivity, or Western universities’ panic about student misconduct. It can develop its own tools and practical approach to AI-assisted writing in both Vietnamese and English – clear enough for global communication but still grounded in Vietnamese concerns and ways of thought.

AI can make writing and communication clearer. It can help Vietnamese thinkers enter global conversations with more confidence. But clarity is not the same as authorship.

The future skill is not communication without AI. It is using AI without surrendering intent, specific nuance, or the awkward phrase that made the argument worth raising in the first place.

Story: Dr Byron Mason, Senior Lecturer in Robotics and Mechatronics Engineering, RMIT University Vietnam

If AI influences how we write and organise ideas, the subsequent question is more confronting: what happens when it begins to influence who we are inside? In the next commentary, RMIT Senior Lecturer in Computer Science Dr James Kang examines AI’s broader impact on human character. Read the article: “The quiet loss of human character: How AI is making us more alike”.

---

Thumbnail image: patpitchaya – stock.adobe.com