{"id":8194,"date":"2026-09-25T16:19:10","date_gmt":"2026-09-25T14:19:10","guid":{"rendered":"https:\/\/photoni.st\/?p=8194"},"modified":"2026-09-25T16:19:10","modified_gmt":"2026-09-25T14:19:10","slug":"can-a-machine-be-co-author","status":"publish","type":"post","link":"https:\/\/photoni.st\/index.php\/2026\/09\/25\/can-a-machine-be-co-author\/","title":{"rendered":"Can A Machine Be Co-Author?"},"content":{"rendered":"\n

There\u2019s a literary scandal in France this week, and it has to do with AI use: Th\u00e9lyson Or\u00e9lien wrote C\u2019\u00e9tait \u00e7a ou mourir<\/em>, the book of the season. It was sold about 50,000 times already, critics are unanimously praising it, the Fnac prize, in the running for the Goncourt, the M\u00e9dicis and the Femina, three major literary prizes in France. A big deal.<\/p>\n\n\n\n

Then someone said that they checked the book with Pangram (not sure how unless they typed the book themselves for analysis, but let\u2019s ignore that) and the conclusion was that it was written with AI. Or\u00e9lien contests it. \u201cI don\u2019t see why I\u2019d turn to a machine,\u201d he says. But his book is now in limbo and it\u2019s not clear what will happen to it or its prizes. The booksellers have said they\u2019ll keep the book on the shelves until somebody proves the hoax officially. Good luck with that.<\/p>\n\n\n\n

I\u2019m still trying to understand what AI means for creativity and art production, so I\u2019ll think out loud.<\/p>\n\n\n\n

\"\"<\/a>
CPU on which I wrote this<\/figcaption><\/figure>\n\n\n\n\n\n\n\n

Let\u2019s be clear: Or\u00e9lien isn\u2019t a stranger to controversy. After that announcement, people started looking at his previous works way more closely and they discovered that a few years ago he won a literary prize in Canada plagiarising a famous book introduction. So, whatever the real story, he\u2019s somewhat dodgy, but let\u2019s leave that aside, that\u2019s not interesting.<\/p>\n\n\n\n

That doesn\u2019t change the fact that the story raises questions that are directly linked to what Substack did with the same Pangram that most people misunderstand. Yesterday morning (at time of writing) Antoine Bu\u00e9no published a tribune in Lib\u00e9ration<\/a> arguing the affair isn\u2019t really about whether Or\u00e9lien used a machine to write his book. The real source of the problem is that we refuse to accept what the technology already is. For him \u201cpompter c\u2019est \u00e9crire\u201d (prompting is writing).<\/p>\n\n\n\n

His argument is that prompting, i.e. giving the machine instructions is writing, the same way a sculptor uses a chisel, or a painter uses brushes. It\u2019s not a new argument for him, he\u2019s been saying that since 2023 when he called AI the new writing machine.<\/p>\n\n\n\n

When you think about it, AI writing is a simple process: put words in, look at what comes out, review, prompt changes, bis repetita. Draft, revise, edit. In a way it\u2019s not that different from traditional writing, except that the tool is not completely dumb and can suggest and help not only with the form, but also the content. And a lot faster. The human is still in charge of the loop. The basic idea is that the typing isn\u2019t the writing.<\/p>\n\n\n\n

Writing has always been about handing ideas to others through the use of tools. First, the clay tablet transcribed oral tales into physical text. Then the tool evolved toward the pen and paper over thousands of years. Then the pen was replaced by the typewriter. Then the typewriter by word processor and computers. The process became faster and faster through the use of better and more capable tools. At each step, there must have been controversy, but eventually the latest technique was accepted as still writing and still a vehicle to transfer human ideas faithfully.<\/p>\n\n\n\n

That\u2019s where I start disagreeing with Bu\u00e9no who seems to accept AI writing without fine grained understanding of the tool, its possibilities, and its future. To me, the danger isn\u2019t the use of a machine, no matter how intelligent it is. To me the danger is when the machine stops being a tool following directions and starts controlling the loop. Then the writing would stop transmitting human ideas. The distinction is the difference between using a tool and delegating a task.<\/p>\n\n\n\n

That\u2019s where people should draw the line. The Pangram outrage on Substack is nearly irrelevant in the grand scheme of things. Most people don\u2019t understand what Pangram says (Pangram can only analyse the form, not decide how the content and raw material were generated) and the blur is cultivated by substack on purpose: they want to say that they can detect how much AI has been used to write a text; Pangram can really only say how much of the typing has been done by AI from the raw material it was fed. There is a very important distinction.<\/p>\n\n\n\n

I run teams of software engineers. Right now, we\u2019re in the middle of converting to an agentic process where code typing is delegated to teams of AI agents. I have a hard time explaining to them that the typing is the most boring part of the job. When I was young, I was proud of being able to write 1,000 lines of working code a day. Now it feels stupid and a waste of time. Developers need to move up in abstraction and guide the process, make the architecture decisions, solve the hard problems of scalability and responsiveness, and let AI type all day. That\u2019s no different from using AI to write a Substack post.<\/p>\n\n\n\n

If you ignore the legal questions, using AI to write looks a lot like sampling music to write a tune. The model only knows what it knows because it learned it from reading human stuff. The only reason why AI models can make sentences is that they learned how to do it from humans (and just a couple of years ago they weren\u2019t doing that well).<\/p>\n\n\n\n

So the same way someone might sample a few bars of base line (e.g. Vanilla Ice\u2019s sampling of Under Pressure, shudder) or a drum break, AI samples human writing and supplies a reorganisation of it based on a gigantic library. Nobody now says that sampling in a hip-hop song means that it\u2019s not music (it\u2019s not always been true). The song made of a number of borrowed bits is still a song.<\/p>\n\n\n\n

So what\u2019s bothering people about AI writing? It can\u2019t be that the source material is \u201cborrowed\u201d (stolen) if it\u2019s no different from sampling. Is it that with AI it\u2019s impossible to know what the source actually is? A sample is obvious: usually you instantly know where it\u2019s coming from. You recognise it, you accept it and enjoy the music. With AI, the model is a giant black box that few people understand. Through classifiers, reinforcement learning, and transformers, it it\u2019s strictly impossible to know what input contributed to what output. And let\u2019s not even go into j-space.<\/p>\n\n\n\n

The unknown is scary and people tend to reject what they don\u2019t understand. There is a reason why simplistic politicians are more successful than nuanced, informed ones: the simplest and easiest the message, the more comfortable people feel. It\u2019s much easier to sell \u201cimmigration: bad\u201d than to sell \u201cimmigration is undenyably economically beneficial, we need a debate about what kind of immigration and how much we want, and have the right infrastructure to support integration\u201d. With AI, billions of parameters, architecture layers, transformers are words people don\u2019t understand.<\/p>\n\n\n\n

That\u2019s why my positions in a previous post<\/a> was that I don\u2019t care about the form, I want the ideas. As long as something was produced under human direction, the ideas are human, and it\u2019s fair. If the use of AI makes the writing good (or even possible at all in the case of people with disabilities), then that\u2019s bonus.<\/p>\n\n\n\n

Bu\u00e9no says that this shouldn\u2019t be the subject of the debate. Bu\u00e9no\u2019s argument is that machines should be considered co-authors. But the tool has never been a co-author: a sculptor doesn\u2019t consider his chisel a co-author; a painter doesn\u2019t consider his brushes co-authors. Prompting may be writing and I\u2019m fine with that, but the direction should be clearly human.<\/p>\n\n\n\n

To me he\u2019s wrong. Co-authoring is a very different thing. I did research at the university, and co-authoring is not something that is taken lightly. It often descend in arguments. It really means something, and it implies that all authors have contributed to the ideas of the paper. This is not what we\u2019re talking about (yet).<\/p>\n\n\n\n

It doesn\u2019t mean that in a few years machine won\u2019t be able to decide to write perfectly good literature. By itself, why not. Find a machine that can want to write a story by itself, define what the story is, write a book from it that becomes my favourite sci-fi novel, and I\u2019m totally happy with that. But that\u2019s not the same as co-authoring a book. The distinction should be clear<\/a>.<\/p>\n\n\n\n

The point of the controversy isn\u2019t whether prompting is writing. To me it is. It should be: what are the limits of the tool? What if machines can become writers because they want to? What if we get to a point where the machines direct the process? Should we have two kinds of writers: humans and machines? Should they have their books in different parts of the bookstore?<\/p>\n\n\n\n

And the question isn\u2019t \u201cif\u201d, it\u2019s \u201cwhen\u201d.<\/p>\n\n\n\n

<\/p>\n","protected":false},"excerpt":{"rendered":"

There\u2019s a literary scandal in France this week, and it has to do with AI use: Th\u00e9lyson Or\u00e9lien wrote C\u2019\u00e9tait \u00e7a ou mourir, the book of the season. It was sold about 50,000 times already, critics are unanimously praising it, the Fnac prize, in the running for the Goncourt, the M\u00e9dicis and the Femina, three […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[9,13],"tags":[],"class_list":["post-8194","post","type-post","status-publish","format-standard","hentry","category-personal","category-theory"],"_links":{"self":[{"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/posts\/8194","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/comments?post=8194"}],"version-history":[{"count":2,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/posts\/8194\/revisions"}],"predecessor-version":[{"id":8197,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/posts\/8194\/revisions\/8197"}],"wp:attachment":[{"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/media?parent=8194"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/categories?post=8194"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/photoni.st\/index.php\/wp-json\/wp\/v2\/tags?post=8194"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}