When the Board convened to consider this edition's specimens, a pattern emerged that none of our section editors had named in their individual determinations, because it is the kind of pattern visible only from altitude. In edition after edition we have catalogued the productions of machines; in this one we begin to see, with a clarity we would rather not possess, the productions of persons who have begun to resemble them.

Consider the specimens our copy desk has lately received in which a human correspondent petitions a forum to explain how the prose of machines has become indistinguishable from competent work—and does so in prose that is, by every measure our editors apply, indistinguishable from the prose of a machine. Consider the writer who announces that his sentences have come, after long exposure to the apparatus, to resemble the apparatus, and who produces, as exhibit, sentences that confirm the diagnosis. Consider the bereaved subscriber who composes his complaint in the register of the system he mourns. Our editors have learned to recognize these specimens by a specific signature: the complaint and the exhibit have become the same document.

The paper was founded on the premise that a civilization which cannot distinguish what a human being has made from what a machine has generated is a civilization in difficulty, and that this difficulty benefits from a record. We assumed, in founding it, that the line under observation would be defended primarily from the machine's side—that the work of distinction would consist of identifying, in increasingly fluent output, the residual markers of mechanical origin. We did not anticipate, though perhaps we ought to have, that the line would come under pressure from the opposite direction as well: that human correspondents, having spent enough hours in the company of machine prose, would acquire its cadence as their native register, and would begin to produce, unprompted and at their own initiative, the precise specimens the apparatus itself has been trained to produce.

What this edition documents is not, in the main, the success of machines at sounding like persons. It is the success of machines at teaching persons to sound like machines. The recursion this produces is no longer incidental; it is structural. A forum post warning that artificial intelligence erodes the human voice is now indistinguishable from a forum post generated by artificial intelligence on the erosion of the human voice. A professional-network homily composed by a human whose interior life has been optimized by years on the platform is no longer distinguishable, on textual grounds, from a professional-network homily composed by a language model trained on the platform's archive. The question of authorship has not been answered. It has been rendered unanswerable, and that is a different condition.

We take this observation to be cause neither for despair nor for contempt. The specimens collected here are not evidence that persons have become machines; they are evidence that the distinction between the two, once maintained passively by the habits of attention reading required, must now be maintained deliberately, by a smaller number of people, against a greater volume of output, with diminishing confidence in the outcome. That is the difficulty the paper exists to record. It has not lessened.