> ## Content Index > Fetch the complete content index at: https://icefrontiers.com/llms.txt > Use this file to discover other available public pages before exploring further. # Writing with AI - URL: https://icefrontiers.com/writing-with-ai/ - Published: 2026-10-13T01:42:29.000Z - Updated: 2026-10-13T01:57:54.000Z - Description: Navionics drew a neat magenta line to Kodiak. The track we actually sailed was different. I consider myself the author of what I write with AI because the responsibility of what I publish stays with me. - Author: Jean Peccoud - Tags: Letters from the Edge While planning our passage from Akun Island to Kodiak, Alaska, I started where every passage starts: with a departure point and a destination. I marked both on the chart in my Navionics application and added a few intermediate waypoints to shape the route I had in mind. Only then did I ask the software to calculate a route. Within seconds, it produced a neat magenta line across the chart, connecting the waypoints I had chosen. There was just one problem. Parts of the proposed route passed much too close to shore and dangerous reefs for my comfort. So I edited it. But even the edited line was only a proposal. Over the following days, the GPS recorded our actual track, and it never quite matched the magenta line. Wind, current, and sea state pushed us off it. At times I followed the line closely; at other times I chose to leave it. Every mile of that track reflects a decision made aboard, including the decision to follow the line where it made sense. This is how modern navigation works. The software proposes; the skipper interprets the proposal, adapts it to conditions the software cannot anticipate, and remains responsible for the boat and crew throughout. The software doesn’t know everything about the boat, the conditions we may encounter, or the level of risk I am prepared to accept. And it won’t bear the consequences if something goes wrong. The magenta line belongs to the software. The yellow track is mine. --- That experience provides an interesting perspective on a question I have been considering recently: what does it mean to be an author when technology, and particularly artificial intelligence, helps produce the text? This blog has been an experiment in using AI extensively and responsibly. After writing more than 100 articles, it is time to clarify my own approach and disclose it to the people who read me and the people I work with. I describe my writing process in detail on a separate [AI disclosure page](https://icefrontiers.com/ai-usage-disclosure/). Here, I want to explain why I believe that process still makes me the author of what I publish. ## Writing has always involved technology We have been using technology to assist with writing for more than a century. The transition from handwriting to the typewriter, and later to word processors, fundamentally changed the writing process. A typewriter made producing legible text easier, but revising a manuscript remained laborious. Word processors made it possible to move paragraphs, replace words, and reorganize entire sections without retyping pages. Even the simple ability to copy and paste changed how we develop and organize our thoughts. Writing became a more iterative process, with fewer practical constraints on revision. Spell checkers introduced another form of assistance. They began by identifying spelling mistakes, then evolved to suggest grammatical corrections, alternative formulations, and improvements in style. Tools such as Grammarly have incorporated increasingly sophisticated AI capabilities, progressively blurring the distinction between correcting what a writer has produced and suggesting what the writer might say. Each of these technologies has changed how we write, and some have changed how we think while writing. Yet I doubt anyone today would argue that authorship requires putting words on paper with a pen, or even typing every word without assistance. ## When someone else writes There is an even more extreme example of the separation between producing text and claiming authorship: ghostwriting. Many memoirs, autobiographies, and books attributed to public figures are written, in whole or in part, by professional writers. The person whose name appears on the cover may contribute the experiences, ideas, interviews, and editorial direction without writing much of the manuscript. This practice is widely accepted in publishing, even when the ghostwriter’s contribution is not explicitly acknowledged. Ghostwriting raises legitimate questions about authenticity and disclosure, but it also illustrates that authorship has never been determined solely by who produces the words. ## Lessons from scientific publishing Scientific publishing provides another perspective on authorship. Over thirty years as a scientist, I have published more than a [hundred research articles](https://scholar.google.com/citations?user=AnVSQhoAAAAJ&hl=en&ref=icefrontiers.com), most of them with multiple coauthors. Early in my career, I was a contributing author providing data that would be integrated into the paper. Since then, I have mostly served as corresponding author, the senior author who coordinates the write-up and acts as interlocutor with the journal. Writing a scientific article is a deeply collaborative process. Graduate students working on their first paper sometimes make the mistake of treating it as a personal writing assignment. They sit in front of a blank page, struggling to find the right words, when they would be better served by discussing the paper with their colleagues. The process often begins with brainstorming the main conclusions, developing an outline, and deciding which figures will best support the argument. There is even an argument for building the paper around the figures, but that’s an entirely different story. From there, the manuscript evolves through writing assignments, discussions, and successive rounds of revisions. Ideas emerge during these exchanges. Arguments that initially seemed convincing are challenged. New analyses may become necessary. A colleague who was expected to make a minor contribution may end up providing an essential insight. Another who was expected to make major contributions may disappear entirely. The process can span months, sometimes years. It is therefore difficult to predict how it will unfold. Often, the contributions of different people only become clear when the manuscript is getting ready to be submitted. For this reason, I have developed a personal rule: I don’t put authors’ names on a manuscript until we are ready to submit it. As corresponding author, I consider it my responsibility to identify potential coauthors, ask whether they believe their contributions justify authorship, and discuss the basis for their claims. I let people claim their coauthorship rather than assign it based on my recollection of their contribution to the work and eventually its description in a manuscript. Over the past few decades, scientific publishers have worked to clarify authorship criteria. The [CRediT contributor taxonomy](https://credit.niso.org/?ref=icefrontiers.com) identifies fourteen distinct contributions, including conceptualization, data analysis, software development, funding acquisition, and writing. Not every contribution necessarily qualifies someone for authorship, but the framework recognizes that scientific knowledge is produced through many different activities. Scientific publishing also emphasizes accountability. Authors are expected to approve the final manuscript and take responsibility for the integrity of their contributions. This experience has shaped my understanding of authorship. A scientific article is not simply the product of several people contributing separate pieces of text. It is the outcome of an iterative and collaborative rewriting process. And by the time a manuscript is ready to be submitted, nobody remembers who wrote what part of the manuscript. Most of the text has been rewritten multiple times by multiple authors. Nor are coauthors the only people who shape a manuscript. Editors may propose fundamental changes to a book’s structure or challenge its central argument. Scientific peer reviewers routinely identify weaknesses and request revisions that can substantially change a paper’s conclusions. Their contributions can be intellectually significant, yet they do not lead to authorship, because the author must decide which suggestions to accept, which to reject, and how to incorporate them into the finished work. Editors and reviewers make suggestions to improve a manuscript; authors are free to follow them or not. They bear the responsibility for the final product. Editors and reviewers don’t. ## Using AI without surrendering authorship My writing process today is surprisingly similar to the process I follow when writing scientific papers with colleagues. I bring the ideas, experiences, and arguments, often as notes and rough paragraphs. AI helps me organize the material and develop it into a complete text, which I then edit, rewrite, and reorganize. As the manuscript approaches completion, I submit it to several models for critique, much as I circulate a scientific paper among colleagues and reviewers. To return to the chart: the waypoints are mine, AI proposes a line between them, and the text I publish is my track. When I started working on this article, ChatGPT was very eager to write the entire text for me as soon as I gave it a topic. I politely asked it to moderate its enthusiasm and listen to the argument I was going to build. The distinction matters. The purpose of AI assistance is not to replace my thinking with plausible prose. It is to help me develop, challenge, and communicate my own ideas. ![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/2026/10/image.png) Screenshot showing ChatGPT writing an entire article based on a short prompt. Unfortunately it cannot retrieve this first iteration. ## Originality and conventions Every writer is influenced by what they have read. We learn vocabulary, style, structure, and ways of developing arguments from other authors. Much of what makes writing effective is precisely our ability to work within conventions that readers recognize. AI-generated text tends to follow recognizable patterns of language, argument, and structure. This is often criticized as a weakness, and sometimes rightly so. But familiarity can also be an advantage. Scientific articles, for example, follow remarkably standardized conventions. Their purpose is not to impress readers with original prose, but to communicate new ideas and findings in a form that other scientists can readily understand and evaluate. Shared conventions make it easier for knowledge to circulate across large communities. Different forms of writing make different demands on originality. I have published [short poems](https://icefrontiers.com/good-night/) written without AI assistance, in which the particular choice and arrangement of words are central to the work. Their meaning may be deliberately ambiguous, leaving readers to develop their own interpretations. Much of my writing for Ice Frontiers lies somewhere in between. I want to express personal experiences and original reflections, but I also want those ideas to be accessible to people who may have very different backgrounds. AI can help me find familiar forms that make unfamiliar ideas easier to understand. That is not necessarily a loss of originality. It can be a way of making originality communicable. This does not relieve the author of the responsibility to have something meaningful to say. Nor is it an endorsement of the industrial production of generic content, increasingly referred to as AI slop. ## Standing behind the work I do not consider AI tools to be coauthors. They are capable of making substantial contributions to the writing process, but they cannot assume moral responsibility for what they produce. They have no standing as authors in the way people do, and they frequently remind their users that their outputs can be inaccurate and need to be independently verified. My position is not fundamentally different from the one I occupy as the corresponding author of a scientific paper. In that role, I am the person who decides when the manuscript is ready for submission. My name appears on the page. I am the one who presses the Submit button. On Ice Frontiers, I am the one who presses Publish. There is a risk in this, because no matter how cautious I may be, there is no way for me to verify everything included in a publication. As a corresponding author, I can’t check every dataset included in a paper or independently verify every script used to analyze the data. Instead, I evaluate the manuscript the way reviewers do, by examining the coherence of the story. Is it internally consistent? Is it consistent with what I know of the field? The corresponding author, the editor, the reviewers are part of a review process that strives to establish warranted confidence by testing coherence across multiple, partly independent forms of evidence. More definitive verification is generally relegated to post-publication, when the results are used, challenged, or reproduced by other researchers. A manuscript must withstand that level of scrutiny to be published. The same criteria apply whether a contribution comes from a coauthor or from AI. I am attaching a transcript of a session with Gemini to verify this blog post. [verificationverification.pdf147 KBdownload-circle](https://icefrontiers.com/content/files/2026/10/verification-1.pdf "Download") What makes me the author is not that I personally generated every sentence, but that I have contributed ideas, exercised judgment throughout the writing process, and decided to associate my name with the entire text. And on rare occasions, I have declined offers to associate my name with papers because I lacked confidence in the writing process, long before AI was part of it. Contributing ideas and exercising judgment is not a promise of quality or originality, regardless of how the text is ultimately produced. We did not need to wait for AI to encounter writing that made little meaningful intellectual contribution or lacked originality of form. Yet the poor quality of such work never prevented people from claiming authorship, at their own risk. Because the risks of authorship are real. The most common is rejection by editors who do not want to publish the manuscript. The risks don’t stop once the manuscript has been published. The author of a journal article may face a correction or retraction after discovering that data produced by a collaborator were unreliable. A book author may receive devastating reviews or discover that readers interpret the work very differently from what was intended. These possibilities are part of the uncertainty of writing and publishing. We cannot know how a manuscript will be received before releasing it into the world. What we can decide is whether we have done enough to stand behind it. ## Choosing a course Which brings me back to the route from Akun Island to Kodiak. Navionics proposed a route. I examined it, recognized hazards, and changed it. Then we sailed it, and the track we left behind was mine. Even so, I could not guarantee a safe passage. The weather could change, equipment could fail, or I could make a mistake. Taking authorship of the track did not eliminate those uncertainties. It meant accepting responsibility for navigating through them. I believe the same principle applies to writing and to the use of artificial intelligence. We can investigate, test, revise, and challenge our assumptions. We can reduce uncertainty, but we cannot eliminate it. Publishing this article is itself an exercise in accepting responsibility under uncertainty. I have started working on a book about the 2026 Ice Frontiers expedition, and I know that some editors and publishers have policies that are incompatible with my approach to AI-assisted writing. I don’t intend to change how I write to fit those policies. I would rather be transparent and work with people who share my understanding of authorship, even if that means closing some doors before I have had the chance to knock on them. A few people have already asked whether I use AI and expressed their disappointmentt when I explained that I do. And yet, they keep reading, and they still tell me what they think. I take those reactions seriously. I understand that some people regard the personal act of writing as inseparable from authorship. Others may question the originality or authenticity of ideas developed with AI assistance. I do not expect everyone to agree with my position. AI raises difficult questions about originality, responsibility, and the limits of human control. I am publishing this article not because I believe I have resolved every question about AI and authorship, but because I have examined my approach and am prepared to stand behind it. I welcome disagreement, criticism, and suggestions that might change my thinking. Just as with the route from Akun to Kodiak, I have considered the alternatives, made my adjustments, and chosen a course. I cannot guarantee where it will lead. But it is my decision, and I am prepared to take responsibility for it. I encourage constructive comments below. I will reply to or acknowledge them, and I am open to evolving my AI usage practices if they help me discover flaws I might have missed. --- ## Related Posts [AI Usage Disclosure: How I Write with AII use ChatGPT, Claude, and Gemini in my writing: from my own notes to AI-assisted drafts, multi-model review, and final authorship.![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/icon/ice-frontiers-logo-60x60-008f0382-c626-4c6d-863d-2a8aa249e2c4.png)Ice FrontiersJean Peccoud![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/thumbnail/policy---1-6d74d40b-b0f4-4d11-8a33-6ea29c85824b.jpg)](https://icefrontiers.com/ai-usage-disclosure/) [Ethics at the Edge of Uncertainty: AI, Climate, EducationWhen the Pope and an AI company stand together to address the ethical risks of artificial intelligence, something has shifted. Irreducible uncertainty transforms technical problems into ethical questions — and ethical questions, ultimately, are personal ones.![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/icon/ice-frontiers-logo-60x60-e382cddb-a7ef-442e-8be6-2ebca249f4b2.png)Ice FrontiersJean Peccoud![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/thumbnail/ethics---1-1-a5235a8a-124f-4f6d-b3fe-e87e5749bdb8.jpg)](https://icefrontiers.com/ethics-at-the-edge-of-uncertainty/) [No Stake in the Game — A Reader WritesA reader wrote in about the Magnifica Humanitas piece — admitting he knew nothing of the encyclical, the event, or Chris Olah. What began as a note on borrowed legitimacy turned into something harder: not whether AI understands, but who bears the cost when it decides.![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/icon/ice-frontiers-logo-60x60-3cc1b164-8a17-4698-b08a-8db1c48e3375.png)Ice FrontiersGrant McNamara![](https://storage.ghost.io/c/83/df/83df04cc-ccc4-44fa-aa45-4c4e7c599a17/content/images/thumbnail/shaft---1--12--b802cc38-deac-47a9-8367-8e1d141ac29c.jpg)](https://icefrontiers.com/no-stake-in-the-game/)