When AI writes the words, who is the author? | Nature Africa

Email Bluesky Facebook LinkedIn Reddit Whatsapp X Save article View saved research Bertie Fielding Lire en français I am a fraud. That was the thought I walked with for most of last week. I had submitted an opinion piece to Nature Africa. It was not a research paper. It was a piece of public commentary, drawn from my own experience and written as I now write many pieces of this kind: with AI. Please read that carefully. I said with AI. I did not hand AI a topic and allow it to invent what I think. At least, that was the distinction I had been making. Then the handling editor wrote back. Two AI-detection tools had classified the piece as 100% AI-generated. She invited me, very politely, to rewrite it in my own words or, if the result was wrong, simply to confirm that I had written it. She did nothing wrong. She was careful, fair and open to the possibility that the tools were mistaken. Still, the message landed hard. In this case, however, the result touched something real. AI had generated almost all of the final prose. For a few days, the distinction between writing with AI and having AI write for me suddenly felt rather thin. I have been in academia since 1991, first as a student and then as an academic. That is 35 years. People often repeat that it takes 10,000 hours to become an expert. Academia can feel closer to a million. The number is not the point. Time spent near something does not make one an expert. You have to engage with it, practise it and, sometimes, live it. The ideas in that opinion piece were not borrowed from a machine. They came from years of leading, teaching, doing science, making mistakes and trying to communicate difficult work beyond my own discipline. I am not claiming expertise in how the underlying models are built. But I do claim expertise in the subject on which I was writing, and I do regard myself as a thought leader in that field. That claim rests on experience, judgment and original thinking, not on whether I personally typed the final sentences. The ideas came first My writing process usually begins with a mess. Sometimes I type fragments. At other times I speak into my phone and record the argument as it comes. At that stage, I do not care about punctuation, grammar or syntax. I am trying to catch the thinking before I make it respectable. I then give that material to AI and ask it to produce a draft. The first version is often polished. Sometimes it is much too polished. It can sound confident while completely missing what I meant. That is where the real work starts. I tell it what is wrong. I ask it to rewrite sections, restore details it has removed, add experiences it could not know, challenge the argument and reject neat conclusions that do not reflect the world as I have experienced it. This can take hours, days, weeks or, in some cases, months. I keep returning to the piece until I can recognise myself in it and until I am prepared to stand behind every claim. AI creates almost all the final sentences in that process. That is true. But it does not decide what I believe, what I am willing to say publicly or what I am prepared to defend. I make those decisions. Yet, approval alone cannot be the test of authorship. A senior person cannot simply sign off a text prepared entirely by somebody else and assume that the byline settles the matter. This is where I had to be more honest with myself. Did I type the final piece? For the most part, no. Did AI generate almost all of its wording? Yes. To describe that merely as light AI assistance would be dishonest. Was AI my ghostwriter? Yes. Plainly so. It did not merely correct my spelling, tidy my grammar or suggest the occasional sentence. It did almost all of the writing in the final piece. That means the detectors were not wrong about the prose. What they could not detect was where the ideas came from, whose experience informed them, who directed the argument and who was prepared to take responsibility for it. AI did not supply my years of experience. It did not make me an expert or a thought leader. It gave language and structure to thinking that existed before the final prose did. Perhaps I should therefore not claim to be the writer of the final prose. I can live with that. What I reject is the leap from “AI wrote the sentences” to “the expertise and thought leadership behind them must also be artificial”. Writing and expertise are not the same thing That distinction will irritate some people, especially professional writers. They may say that I am confusing having ideas with writing. That is a serious objection, and it deserves more than a defensive claim that AI is “just another tool”. Professional writing is not simply correct grammar and polished syntax. A good writer finds the shape of an argument, hears when a sentence is false to the speaker, understands the reader and knows what should be left unsaid. That is craft. It takes years to develop, and it has value. Being an expert or a thought leader does not automatically make me a professional writer. It means that I bring informed thinking, experience and judgment to the question. Writing is the craft that carries that thinking to other people, and that contribution should not be treated as insignificant. I understand why professional writers may experience AI not as support, but as something that is taking work away from them. A client may decide that fast and cheap is good enough. A writer who has spent years learning the craft has every reason to be concerned. But AI does not make that decision by itself. People and institutions decide whether speed matters more than craft, whether a competent draft is enough and whether a professional writer should be paid. We should not hide those human choices behind the technology. I am not asking professional writers to celebrate a tool they believe is devaluing their work. Nor am I claiming that AI-generated prose is equivalent to the work of an experienced human writer. Sometimes the right decision is to employ a professional writer, and we should be honest when that is the case. My claim is narrower. An AI-written piece can still carry genuine human expertise and thought leadership. But a prompt and a quick approval are not enough. The thinking, direction, sustained engagement, judgment and accountability must be real. Perhaps this is the distinction we need to debate: writing, authorship and thought leadership are related, but they are not identical. A human ghostwriter is still a writer, even when another person’s name appears on the piece. The named person may remain the source of the experience and argument, but saying “I wrote it” can hide the writer’s labour. We already live with versions of this distinction. Editors, writing coaches, communications specialists and speechwriters help other people bring ideas into public. If a human writer had interviewed me and drafted the piece, an AI detector would not have revealed that assistance. That silence would not, by itself, have answered the ethical question. The boundaries are also different across forms of writing. In a poem or a novel, the words are not merely carrying the work; the words are the work. My argument about an opinion piece grounded in lived expertise should not be used as a convenient defence of every form of AI-generated writing. Nor am I speaking about undisclosed ghostwriting in scholarly publications. My process for academic papers is different. Research carries specific obligations around data, methods, analysis, citations, contribution and authorship, governed by research integrity standards and journal policies. What this means for students This experience also unsettled me because of what we are doing to students. I have heard colleagues say that they know a student used AI because the writing is simply too good, too polished or too perfect. Perhaps what happened to me was a useful warning. A polished sentence is not, on its own, evidence of dishonesty. Nor is imperfect grammar proof of original thought. Students can use AI to avoid thinking. They can also use it to organise their thinking, test an argument, improve expression and learn. Similar looking texts may conceal very different intellectual processes. Writing expert commentary with AI still requires expertise. You need enough knowledge to recognise when the tool is wrong, has flattened the argument or no longer reflects what you mean. Fluency is not knowledge. Perhaps we will get more ethical and effective use if we teach people how to work this way. Using AI is not automatically unethical. Teach disclosure, verification, judgment and accountability. Require people to explain and defend the final work. This discussion needs to happen in the open. A detector cannot reconstruct that process. At most, its score gives us a reason to ask questions, not a verdict. Declaring a student guilty because the writing is too good will not teach responsible scholarship. It may teach them only to make good writing look less good. The journal may still decide that the level of AI drafting in my piece is incompatible with its policy. It has every right to do so. Clear editorial standards matter, and authors must follow them or submit elsewhere. Towards better disclosure I am not writing this to criticise the editor. Her email forced me to confront a question I had perhaps made too easy for myself: when I say that a piece is in my own words, what exactly do I mean? Are words mine only when my fingers first place them on a page? Or can they become mine through the sustained intellectual work of choosing, rejecting, revising, checking and taking responsibility? I do not think the first keystroke is enough to settle authorship. We need better language for the space between the first keystroke and the final approval. There is a meaningful difference between using AI for spelling, asking it to restructure a paragraph, using it for substantial drafting and allowing it to substitute for one’s own thinking. Calling all of this either “human-written” or “AI-generated” conceals more than it reveals. We also need better disclosure. Saying only that AI was used may be technically correct while telling the reader almost nothing. We should be willing to explain what the tool did, what the author did and, where relevant, what a professional writer or editor contributed. For this piece, the honest disclosure is simple. I supplied the experience, the argument, the raw material and the direction. AI produced almost all of the final prose in response to my repeated input. I reviewed it, redirected it and decided what I was prepared to publish. I take responsibility for the final text. What I discovered last week was not that I am a fraud. I discovered that I was feeling like one. There is an important difference. The feeling was not wasted. It forced me to examine my process without hiding behind comfortable words such as “tool” or “assistance”. AI did almost all of the writing, and I should be transparent about that. But I will not allow that fact to erase the expertise and intellectual labour behind the piece. I know that some professional writers will still disagree with me. I want them to. They may say that I am the originator of the ideas but not the writer. I am prepared to hear that. My concern is the larger conclusion that, because AI wrote the sentences, the expertise and thought leadership behind them must somehow be fraudulent. Educators may think the boundary I draw is too loose. Editors may conclude that disclosure is not enough. I would rather have that debate openly than pretend that a detector score has already settled it. This opinion piece was made in the same way. It began with me speaking an untidy account of what had happened and what it made me think. AI did almost all of the writing. I questioned it, redirected it, removed what was too smooth, restored what mattered and decided what I could stand behind. The ideas are mine. The expertise is mine. The experience is mine. The responsibility is mine. That does not make AI’s contribution invisible, and it does not make the work of professional writers less valuable. AI did almost all of the writing. I remain the expert and the thought leader behind the piece. Does that also make me its author? I do not expect everyone to agree. But that is the debate I think we now need. doi: https://doi.org/10.1038/d44148-026-00236-3 Reprints and permissions Jobs Faculty - MD, PhD, MD/PhD, equivalent Texas Children's and Baylor College of Medicine - Open search for research faculty positions. All disciplines, MD, PhD, MD/PhD, equivalent. Houston, Texas (US) Texas Children's Hospital and Baylor College of Medicine Research Scientist We are a frontier artificial-intelligence laboratory, still in stealth, established for a single purpose: to build an AI capable of original discov... San Francisco, California The Infinity Fellowship Computational Research Scientist, Simmons Cancer Center Drive cancer immunotherapy breakthroughs using machine learning and multi-omic data in a senior role bridging research and care. Dallas, Texas UT Southwestern Medical Center Cancer Strategic Recruitment Initiative The University of Alabama at Birmingham (UAB) Heersink School of Medicine (HSOM) and the O’Neal Comprehensive Cancer Center are leading a recruitment Birmingham, Alabama (US) University of Alabama at Birmingham Tenure-Track Faculty- Dept of Neuroscience (Computational or Theoretical Neuroscience) The Dept. of Neuroscience at Yale University seeks tenure-track faculty members with interests in computational or theoretical neuroscience. New Haven, Connecticut Yale University School of Medicine - Department of Neuroscience