Writing is thinking: understanding the real risks of AI in scientific writing

Researchers today are under huge pressure, with serious workloads and overwhelming publication pressure. So it’s no surprise that they’re falling prey to AI’s apparent lifeline for the overworked and overstressed.   

It’s a growing temptation for researchers, whether native or non-native speakers, to turn to AI to write their papers. Large language models (LLMs) such as ChatGPT can take on the challenging task of writing a scientific article in a fraction of the time, so why not?   

Because, it turns out there’s a lot more to lose than there is to gain. In this post we look at what Nature Portfolio’s expert trainer Dr Jeffrey Robens has to say about the implications of outsourcing your writing to AI.   

The value of the writing process

The process of writing is in itself one of the most important parts of your research journey. Writing an article is about more than just communicating what you know to your readers. When you write you’re organising your thoughts and formulating your arguments – it helps you work out what you think.  

So if you delegate this whole process to AI, you’ll be missing out on a whole raft of things. Not only the writing experience but the entire critical thinking process, which is the heart and soul of your research. 

Here’s what you have to lose… 

#1 Your voice 

A scientific paper is your opportunity to showcase your research to the community. It’s where your ideas and insights come together in a way that shows readers what you think, why it matters, and what its place is in the field.  

Your credibility and reputation rest on the quality of the paper you produce. So you need to demonstrate your thinking clearly and let your originality shine through. And this can only come from you, not an LLM. You want your readers to hear what you think, not what ChatGPT thinks.

#2 Quality and accuracy 

LLMs are unreliable. You can’t count on them for the facts because they aren’t built with factual accuracy in mind. They’re built on predicting patterns which reflect the structure and content of their training data. However authoritative they might make something sound (and they do), this can often be based on false information, including non-existent papers, authors and data. This is known as “hallucination” and occurs when LLMs generate apparently convincing content based purely on statistically plausible wording.  

Anything generated by AI is completely unverifiable. Which creates a lot of work for you if you decide to use it to write your article. You’ll need to revisit every sentence, checking the accuracy of every fact and the validity of every reference.  

Does that really save time? Definitely not – it would be quicker and more efficient to write the paper first yourself and use AI to fine tune the end result. LLMs can, however, be a valuable support in the writing process, editing the grammar and refining the text to achieve good flow and readability.  

#3 Understanding your ideas  

“Writing is thinking in slow motion” Yean Foong 

One of the most important reasons for writing your own paper is that it forces you to think, and really know, your research. When you write it up, you need to structure your thoughts in such a way that your audience can clearly understand your thoughts and ideas. As we’ve already seen, it’s about more than writing it down. It’s an exercise in itself about clarifying your thoughts to formulate arguments, and identifying any gaps in your logic or evidence. When you’re later called upon to defend or explain your research, you’ll be glad you didn’t outsource the real thinking to AI. 

This is why revisions are important too. They aren’t so much about correcting anything that was “wrong” in your first draft, they’re about refining your understanding as you see your work again with fresh eyes. 

#4 Essential writing practice 

Writing up your research can be challenging for anyone – academic English is difficult. Even more so if English isn’t your first language. But the experience of writing is one of the best ways to improve, not only for future papers but more widely too. 

Writing develops your ability to explain complex scientific ideas clearly and confidently. This is an essential skill for conference presentations, job applications, international collaborations and leadership opportunities in the global research community.  

And the more you write, the better you’ll get. Your future self will thank you for putting in the effort now. 

 

Where AI really helps with writing

LLMs do however have their uses when used in the right way. They can be great assistants for: 

• Brainstorming phrases or alternative sentence structures; 
• Correcting grammar;  
• Improving flow and readability; 
• Summarizing long blocks of text. 

But the key word here is assistant. You should remain the author of your research. You should be doing the thinking, which as we’ve seen is core to the writing process. Ultimately it’s your voice, your reputation and your ideas that will further your scientific career.  

 

Final thoughts

If writing feels hard, that’s because thinking is hard. But that’s also what makes writing powerful – it helps us slow down, reflect, clarify and grow.   

So next time you sit down to write, don’t look for shortcuts. AI won’t do a good job, so embrace the process instead. Writing is thinking, and it’s one of the most valuable tools you have as a researcher.  

Building on the ideas in this post, Nature Masterclasses offer further advice and training, including their Harnessing Generative AI in Academic Publishing live workshop and  Writing a Research Paper: 2nd Edition, which introduces the power of narrative tools and the principles of scientific writing. Complementing this, Publishing a Research Paper: 2nd Edition navigates the editorial and publishing process, with a focus on how to submit your research paper. Our live workshop Getting Published: Effectively Communicating Your Research provides insights and strategies to help researchers increase their chances of publication and maximize the impact of their research. 

Dr Jeffrey Robens is Head of Community Engagement at Nature Portfolio and conducts training workshops worldwide to help researchers improve their publication quality and impact. With 20 years’ experience in academia, he has published prolifically and received numerous awards. 

Strengthen your scientific writing — without losing your voice.
Explore Nature Masterclasses to build confidence in scientific writing, develop a clear authorial voice, and learn how to use AI responsibly as a support tool, not a substitute for thinking. Or contact our sales team for further discussion.