Responsibility Without an Author
When AI produces the reasoning, signing is no longer enough: responsibility needs a chain that can answer.
Ask yourself why a prescription has value. Not because it is medically accurate: an encyclopedia page can be that too. It is because a physician signed it, meaning an identifiable person, trained, insured, strikeable from the register, staked their responsibility on its content. The same goes for an architect's plans, accounts certified by an auditor, a deed drafted by a lawyer. In all these cases, the document's value does not lie first in what it says, but in the fact that someone answers for it. A profession is not a body of knowledge. It is a body of knowledge someone answers for.
This truth is so old we have stopped seeing it. It resurfaces because artificial intelligence now produces, at scale, an object new in the history of the professions: work without an author. Work that is competent, often excellent, perfectly presentable, and of which no one is the author in the full sense, the sense of someone who commits. The text exists, the quality exists, but the commitment does not. And the whole question, for trades that live on responsibility, is what work is worth when no one answers for it.
A profession is not a body of knowledge. It is a body of knowledge someone answers for.
What a signature really does
A misunderstanding about the signature must first be dispelled. We believe it certifies accuracy, that it is a kind of quality label affixed at the end. That is wrong, and the error is instructive. The signature does not guarantee the work is right; no one can guarantee that. It designates who will have to answer if it is not. It converts a document into a commitment, a content into a position, a text into an act. The physician who signs does not promise infallibility; they accept that the error, should it occur, will be theirs, with all that follows: their conscience, their reputation, their insurance, their right to practice.
This is why the signature is surrounded by institutions, not technologies. The professional body that admits and strikes off, the oath that binds publicly, the code of ethics, the insurance that compensates, the discipline that sanctions: this whole apparatus exists only to make the signature serious, which is to say costly. A signature that costs nothing is worth nothing. Society did not trust the professions because they were learned; it trusted them because they had organized, upon themselves, a system where being wrong has a price and someone pays it.
Seen this way, responsibility is not a burden weighing on the professions: it is their most precious asset. It explains why one pays a lawyer for an opinion a book would give, a physician for a diagnosis a manual would describe. What is bought is not the information, which has never been so abundant or so free. What is bought is the commitment of a person who puts their career behind their conclusion. At a time when information becomes an unlimited, costless flow, that commitment is precisely what remains scarce.
The signature does not certify that the work is right. It designates who will have to answer if it is not.
Orphan work
Now here is what AI introduces: work that has every property of professional work except that one. The generated memorandum is structured like a memorandum, argued like a memorandum, cited like a memorandum. But behind it, there is no one in the sense the institutions of responsibility intend. A model cannot be sanctioned, insured, struck off, or summoned before a disciplinary chamber. It cannot be ruined by its fault, nor feel shame, nor draw from it a lesson that binds it. The work is there; the author, in the full sense, does not exist.
It will be said that tools have always existed and the user's responsibility suffices: whoever uses the result answers for it, as the surgeon answers for the scalpel. The argument holds for tools that execute a gesture decided by a human. It weakens for a system that performs the reasoning itself. When the machine chooses the arguments, ranks the risks, formulates the conclusion, the person who signs no longer did the work: they received it. They can reread it, and they must. But rereading is not doing, and the responsibility they engage now bears on a reasoning they did not conduct, whose result they see without having traveled the path.
The danger is therefore not that responsibility disappears at once. It is that it empties from within while keeping its forms. The signature remains, the insurance policies remain, the professional bodies remain; but what the signature covers has changed in nature. To sign work one has done is to answer for one's choices. To sign work one has received is to answer for a reading. If the profession is not careful, the signature slides from commitment to registration, from "I answer for it" to "I saw it go by." And a signature that engages nothing but a rereading is a signature that, slowly, ceases to be worth what it was.
A model cannot be sanctioned, insured or struck off. The work is there; the author, in the full sense, does not exist.
The silent dilution
This slide has a natural slope every organization knows: when no one can answer fully, everyone answers a little, which is to say no one. The error born of assisted work will look for authors: the one who phrased the request, the one who reread, the one who chose the tool, the one who approved its deployment, the system's publisher, the model's provider. Each will have been a link; none will have been the author. Responsibility, which was a point, becomes a gradient. And a gradient of responsibility, for a client, a patient, a litigant, amounts to an absence.
The professions have met this slope before, and what they built to resist it is illuminating. Aviation did not dilute responsibility into cockpit automation: it maintained, against technological evidence, a single captain who answers for the flight, whatever systems assist them. The decision is sometimes almost entirely machine-made; the responsibility has not moved one seat. That choice is not technical, it is institutional: it was decided that a flight would always have an author. It is exactly the decision the intellectual professions now face, and it will not take itself.
When no one can answer fully, everyone answers a little. Which is to say no one.
Law, once again at the outpost
Law meets this question before the others, for a simple reason: responsibility there is not an adjunct to the work, it is its very substance. A lawyer's deed is worth the responsibility it engages; an opinion is worth the person who carries it; a pleading exists only signed by a name inscribed at a bar. The profession was built entirely on the idea that legal work is work one answers for, before the client, before the bar, before the judge. It is therefore the first to experience what becomes of this architecture when a growing share of the work is produced by what can answer for nothing.
But what plays out in firms prefigures what awaits every trade of commitment. The physician whose diagnosis is suggested by the system, the auditor whose anomaly is flagged by the algorithm, the engineer whose calculation is produced by the tool face the same question a few years apart: what does signing mean, when one no longer conducted the reasoning one signs? Law is not an exception protected by its conservatism. It is the outpost where the question arises in its purest form, because nowhere else does the value of the work merge so completely with the responsibility of its author.
Nowhere else does the value of the work merge so completely with the responsibility of the one who signs it.
Making the signature possible again
The answer cannot be to refuse assistance, and this must be said plainly: a professional who forgoes the best instruments also fails their obligations, and the average quality of assisted work makes its use inevitable and desirable. The question is not whether AI will produce legal, medical or accounting work. It already does. The question is under what conditions a person can still, honestly, answer for it. Not formally, by affixing a name at the bottom of a page, but really, in the sense that the signature commits something the signatory actually governed.
These conditions can be described. That the signatory can see the reasoning, not only the conclusion, failing which they sign a result and not a path. That the important choices were submitted to them as choices, with their alternatives, not buried in a smooth answer. That verification is organized to the measure of the stakes, not left to the residual vigilance of a busy day. That one knows, finally, what came from the machine and what came from the human, not to distribute blame, but so that the latter's commitment bears on what they actually held. Under these conditions, the signature becomes possible again: no longer the trace of a rereading, but the act of a person who governs work they did not entirely do, as the captain governs a flight they no longer fly by hand.
There will then remain to the professions a decision that belongs to them alone, because it is institutional and not technical: to reaffirm that all work that binds others will have an author, a real one, single and identifiable, whatever systems contributed to it. This is not resistance to progress; aviation showed it is the condition for progress to be accepted. The trades that keep an author behind every piece of work will keep what makes them exist: the trust of those who cannot verify, and who, precisely for that reason, need someone to answer. The others will discover that authorless work, however brilliant, is worth exactly what its last rereading is worth.
Work no one answers for is worth what its last rereading is worth. The professions exist to be worth more.
