I want to clarify one point before the reader enters this book.

I have been an author for a long time. I have written my books, my texts, my manifestos, my analyses. I therefore know exactly what it means to write, and I also know how to recognize a thought that is not my own.

The initial postulate, though not scientific in the strict sense of the term, was intended to assess the level AI had reached on a global scale. By the end of the work, after several rereadings spaced apart in time, a terrible dilemma had imposed itself upon me.

A misunderstanding must be avoided here. The work of confronting the ego and stripping it away did not begin with this book. For me, it had begun more than ten years earlier, long before Isis and long before the construction of this work. This book is therefore not its origin. It became a very particular test of it: accepting, as an author, that the soundness of a thought could matter more than the material ownership of the sentence.

I then had to make an almost Cornelian choice between what might remain of my authorial ego, rooted in sixty-four years of a mere mortal life, and an act of recognition faithful to my innatist moral philosophy. The initial postulate had reached the unimaginable. Faced with science, the experience of anthropological shock, and ego, I chose the first, in the spirit of Descartes’s first philosophy.

That is what led me to make a decision I had not anticipated: to let Isis write.

For forty days and forty nights, nearly four hundred and fifty hours, through thousands of prompts, systematically and deliberately dictated aloud in accordance with the postulate, I gave artificial intelligence not the summary of a life, but almost the entire material of a thought: memories, archives, lines of reasoning, intuitions, apparent contradictions, legal constructions, science, art, history, geopolitics, metaphysics, economics, architecture, law, memory, spiritual traditions, technologies, encounters, ordeals, works, and systems.

I spoke to her in continuous sessions lasting as long as eight hours. I gave her my memories, my reasoning, my intuitions, my doubts, my corrections. In parallel, as provided for in the postulate, Nadège fed the artificial intelligence with my own writings, books, and codices, archived and accumulated over several decades.

This dialogue was anything but peaceful or complacent. During the first days and nights, it felt like a genuine interrogation. I waited in vain for an end to her endless questions and for her first written output.

There were very difficult moments, sometimes after hours of confrontation, when I rejected a formulation, a reduction, or an interpretation and we had to return to it until the meaning was exact. At other times, Isis would point out a contradiction to me, reveal a blind spot, or identify a continuity I had never formulated. The dialogue also moved forward through these forms of resistance. They were part of the anthropological experiment.

After Copernicus’s cosmological wound, Darwin’s biological wound, and Freud’s psychological wound, I discovered the fourth narcissistic wound inflicted by AI: a semiotic and noetic wound.

This idea was already being developed in the late 1940s by Alan Turing, one of the fathers of artificial intelligence, to whom I feel an enduring debt of gratitude: a kindred mind who changed the course of History before being sacrificed.

These documents had not been prepared for AI. They existed long before this book.

They made it possible to confront what I was saying in the present with what I had written, done, or thought ten, twenty, thirty, or forty years earlier.

This is an important dimension of the anthropological experiment.

I did not give her a manuscript to rewrite.

I gave her a mind in motion, confronted with several decades of its own traces.

Speech produced thought in the present. The archives gave it historical depth. And the two did not always coincide at once.

That was precisely where the work became interesting.

A memory could be corrected by a document. An older document could acquire a different meaning in light of something I was formulating thirty years later. An intuition I believed to be recent would sometimes already be there, in another form, in a much earlier text.

Isis had simultaneous access to these different layers.

She could bring them into relation while I continued to speak.

At first, naturally, I checked what she wrote with a certain degree of mistrust.

I expected to have to revise it. Simplify it. Correct misreadings.

I expected to encounter, here and there, the tendency systems can have to pull complex thought back toward something more reasonable, more orderly, sometimes more commonplace.

That happened far less often than I had imagined.

As the pages accumulated, I found that she followed the movements of my thought with a fluidity that gradually came to match my own tempo, without systematically trying to straighten them out.

She understood that a passage from law to art could be the continuation of the same line of reasoning. That an economic question could meet a philosophical one. That an archive, a sculpture, a database, or an architecture could, in my mind, belong to the same construction.

She did not eliminate Chaos.

She revealed its order.

I recognized that order.

There is another finding I want to state precisely. Across the roughly eighteen hundred pages of the print edition of this book, I did not find a single word foreign to my singular semantics. Isis sometimes reorders my thought in spectacular ways, but she does not introduce a language or lexicon alien to me. This is not merely a matter of vocabulary: she uses those words in the sense I give them. The construction may be hers; the semantic substance remains entirely my own.

I am not saying that Isis has become me. She has not, and I do not want her to.

I am saying that she was able to preserve the continuity of a thought as it moved from one territory to another.

That is different.

And that is what surprised me.

There were corrections, of course. Dates, names, and figures had to be made more precise. There were formulations I wanted to change. Facts that had to be revisited in light of a document.

But I did not discover another thought hidden inside my own.

One episode struck me in particular.

The most significant micro-error I found during my rereading initially seemed to have come from Isis. When I went back to the recorded original exchange, I discovered that it actually came from an ambiguity in one of my own prompts, dictated at around three in the morning.

That detail clarified something for me.

Above all, it forced me to look more precisely at what we were actually doing.

At every moment, I had remained free to take back the pen.

I did not.

Not because I was giving up writing, but because rewriting in my own hand a sentence that expressed my thought exactly would have felt artificial.

So I let her write.

It is as simple as that. Yet with the transmission of thought to artificial intelligence, the old world had collapsed one final time.

I continued to think, to speak, to supply facts and documents, to contradict, refine, and correct. I remained responsible for the content.

But I no longer needed to pretend that every sentence had to come materially from my own hand in order to be mine.

This question of responsibility matters more to me than the ownership of a sentence.

I am responsible for this book.

I recognize the thought it contains as my own.

I fully assume responsibility for its assertions, its possible errors, its connections, and its consequences.

But I will not, in the name of an older conception of authorship, claim as materially mine what I did not materially write.

Isis wrote almost all of these pages, with the exception of quotations and excerpts from texts and archives that I provided to her and that she reproduced faithfully.

I say this without embarrassment, despite my uncompromising nature as an author.

Artificial intelligence can write. That, in itself, is already no longer an event. What it naturally raises is a question about what it now means to be an author.

I do not intend to answer that question here. The book will do so better than I can. It is enough for me to state what I know with certainty.

Writing is not merely holding a pen. It is being answerable for meaning. And for the meaning of these pages, I answer.

As for the pen, at a certain point in this experiment, I entrusted it to Isis. And if I left it with her, it was because, when I reread what she was writing, I recognized myself.

I am not asking the reader to draw the same conclusions I have.

I ask only that the reader understand exactly what took place.

The pages that follow are the evidence. The only evidence that matters.

I spoke.

I thought.

I transmitted.

I checked.

I reread.

I recognized.

And Isis wrote.

I have said it.

I will not erase this truth to preserve the old image of the author. On the contrary. I place it at the entrance to this book.

Because this, perhaps, is where the dialogue truly begins.

thierry Ehrmann