A note for readers outside Germany, followed by the essay itself.
The German debate, in brief
Last week Die Zeit reported that Germany’s Digital Minister, Karsten Wildberger, had published newspaper op-eds and delivered speeches that were written largely — in some cases almost entirely — by AI. Using the detection tool Pangram, the magazine put an April op-ed in the Handelsblatt at roughly 99.3 percent machine-generated, a March piece in the Frankfurter Allgemeine Sonntagszeitung at about 78 percent, and a Washington address to the Atlantic Council at, reportedly, one hundred percent. None of it had been disclosed to the editors. The tell-tale markers were stylistic: an excess of em-dashes (25 in a single piece), rule-of-three triads, negation used as emphasis, and a few empty, mixed metaphors. The ministry defended the practice, calling AI a “supporting tool, comparable to a research database,” with a human always checking and bearing responsibility for the final text. (Note the irony: the ministry’s own defense already contains half of the argument below.)
A debate about transparency in political communication promptly erupted. In the middle of it, the philosopher Arnd Pollmann posted that the outrage “is getting a bit much,” adding that — apart from a piece by Axel Springer CEO Mathias Döpfner — he had yet to read a single “fulminant” (rousing, blistering) defense of AI ghostwriting, and wondered, half-mockingly, why not: “Text as text, the death of the author, and so on. Surely that must be possible at the push of a button.” What follows is an answer to that challenge — and, perhaps, a small demonstration of it.
Arnd Pollmann thinks the whole thing “is getting a bit much,” yet misses — “apart from Döpfner” — the rousing defense of all this AI ghostwriting, and asks, with audible mockery, why nobody has bothered to write one: “Text as text, the death of the author, and so on. Surely that must be possible at the push of a button.” Very well, I’ll pick up the gauntlet — and whether this defense was produced at the push of a button or hammered out line by line by hand, you, dear Herr Pollmann, will not be able to say for certain by the end. And that is already the first part of the answer.
Ever since Die Zeit‘s investigation into Digital Minister Wildberger began making the rounds, I keep reading the same indignant reflex everywhere: that a minister had his speeches and op-eds “written by the AI” without disclosing it, that this is a scandal, a fraud on the reader, ultimately an attack on authorship itself — and that, with all sympathy for the unease, strikes me as far too simply thought.
Because let us first look at what Die Zeit actually found: it ran the texts through detection software, counted 25 em-dashes, flagged triads, negations and a handful of crooked metaphors, and then announced that all of it “reads” like a machine — whereas what it did not find was a single factually false sentence, a single lie, a single real error of thought. So the minister was not shown to be wrong; he was shown to sound like a machine. But that is a verdict about taste, not about truth.
So let us take your flippantly tossed-off “death of the author” at its word, because it is far more serious than you probably meant it: in 1967 Roland Barthes announced that a text does not belong to the one who wrote it but to the one who reads it, that meaning is born not at the source but in the recipient — and for half a century we humanities people happily recited this, as long as it concerned Balzac and Proust. Now that a machine takes the conclusion literally and lays a text before us as nothing but text — no suffering artist’s soul behind it, no sleepless nights, simply finished — we suddenly rediscover the sacred creator we ourselves buried in the seminar room. You cannot celebrate the death of the author and at the same time demand his resurrection the moment you dislike whose — or what — hand guided the pen.
And in any case: since when do we disclose who really writes? No chancellor and no minister of this republic has ever penned their own major speeches — since Adenauer the ministries have been staffed with advisers, speechwriters and whole press offices that formulate anonymously in someone else’s name, and not once has anyone demanded a footnote reading: “This paragraph was drafted by department head so-and-so while the minister was at lunch.” The human ghostwriter — invisible, uncredited, often badly paid and still at the desk at three in the morning — has always counted as perfectly self-evident; the machine one, which does the same thing, only faster and with no claim to overtime, is suddenly treated as betrayal. IMHO the difference there lies not in ethics but in feeling: one ghost is made of flesh, the other of statistics, and that wounds us.
And yet what we demand of a public text was never that a particular hand type it, but that a human being stand behind it with their name — the authorship that actually counts is not a keystroke but a liability. Whoever puts their name underneath is saying: I vouch for this, I bear the consequences, I let myself be measured by it; and whether the tool was a pencil, an adviser or a language model changes nothing about that liability. Responsibility, after all, can be delegated to no machine — and that is precisely why the machine is not the problem.
At this point a thought comes into play that I happen to be working on in other contexts anyway: if one accuses the AI of “only reproducing” what humans fed into it, one ought, in fairness, to add that we humans too, our whole lives long, basically only reproduce and recombine what was fed into us — the voices in the womb, the first light after birth, the taste of breast milk, and later the songs, the images, the films, the first beer and the binomial formulas. Out of all of it we cobble together our picture of the world, give back what we have learned, and now and then fit it into something new — which is just about exactly what a language model does too. If the mere drawing on what one has learned devalues the AI text, then it consistently devalues the human one as well; and I want neither — I simply plead for an end to the double standard.
And since I am already on the push of a button you mock so nicely: I mean it seriously, and as praise. It is no impoverishment that good writing becomes possible at the push of a button, but a liberation — the notion that a thought is worth something only once it has been bought with toil, sweat and a little bad conscience is, in the end, the guild ideology of the writing trade, which likes to sell its monopoly on phrasing as morality. Those who think clearly but agonize over the writing — and that is a surprising number of clever people — have so far preferred to keep quiet, and now they may finally join in. That is not the downfall of language but its democratization.
And so to the uncomfortable question I put to myself, and which you, while reading, may already have put to yourself too: would you actually have been able to tell whether I typed this text entirely by hand or merely prompted it, shook it out and reworked it? Provocatively put, the answer is irrelevant — if the text convinced you, it did what texts are for, and if you could not tell it apart with certainty from a “real” one, then the line everyone is currently fretting about runs only inside your own head.
You ask why nobody has written the rousing defense before now. The perhaps most honest answer is at the same time the most unpleasant: because it can in fact be produced at the push of a button — and the mere suspicion that this is exactly what happened here is enough to expose the whole outrage for what it is at its core: a wounded ego, not an argument.
I could write a great deal more on this whole subject, and I am working on it in other contexts anyway — but for today I will leave it at this food for thought. I know that for many this is an uncomfortable, perhaps even an irritating idea. But it grounds you enormously to make clear to yourself that in the end it is the sentence that counts, and not the certification of its origin. 😉
Leave a Reply