The Whole Premise Of Checking For Human Writing Is Daft

Mo Shehu opens by granting the case for writing by hand, at length and without irony: it focuses thought, the first sentence is usually rubbish and the seventh is often better, repetition teaches cadence and argument, and there are things you don't know you believe until you try to explain them. None of that, he says, explains why the physical production of every word has become a test of whether a thought is human.

His objection is a category error. The demand takes one useful method of thinking and promotes it into the only respectable one, confusing the exercise with the outcome and the labour with the intelligence. It asks whether someone typed the words when the questions that matter are whether they had something useful to say, whether they understood it, whether they made the important decisions, whether they checked the claims, and whether they can defend the result.

The example he builds the argument on is a pair. One person types 3,000 words by hand with no thought passing through their head beyond finishing before lunch. Another dictates an idea, argues with an LLM for an hour, rejects six drafts, rearranges the piece, rewrites half of it, and publishes something they understand down to the final comma. Under the emerging rules, he writes, "the writing police see the first person as a human writer and the second as a suspicious package."

The tool-panic pattern

Shehu's framing analogy is fire, and he plays it out as comedy: one band of homo erectus waking up each morning with a new idea for what to burn, cook, harden, or illuminate, and a second band watching from a neighbouring cave with concerns. Cooked meat would weaken the jaw. Warm fires would keep the young awake after sunset. Flame-hardened weapons would deskill the sharpened-stick industry, and before long nobody would remember the honourable craft of chewing raw antelope.

The concessions are real ones. People got burned, someone set fire to the wrong patch of grass, one band attacked another — though the defenders also had fire, which levelled things. The dangers called for rules, better tools, and respect. What nobody concluded was that the existence of danger made warmth illegitimate.

He then lists the same accusation arriving on schedule: the calculator weakening arithmetic, search and maps weakening memory, spellcheck weakening spelling, photography weakening painting, and recorded music, sampling, digital illustration, coding frameworks, and predictive text all serving time as ways to bypass the proper work. The tool becomes ordinary and the outrage moves on.

The accessibility argument

The strongest section is the one about whose body the demand assumes. Its model writer has two working hands, ten obedient fingers, good vision, fluent spelling, enough muscular control to operate a keyboard, and a mind that converts thought into polished prose through sustained typing. Anyone outside that model starts the conversation under suspicion, because assistance has been defined as contamination.

Stephen Hawking is the uncontroversial case. He produced his later work through a computerised communication system and voice synthesiser built with developers, and nobody reads A Brief History of Time wondering whether the machine deserves half the credit. Shehu's harder case is the MouthPad, a fitted mouthpiece that drives a phone or computer through tongue, head, and breath movement. A quadriplegic user might scroll with their tongue, select letters by mouth gesture, complete words with predictive text, dictate other sections, and ask software to expand compressed input into fluent sentences.

His questions about where that user crosses the line are the point of the section, and they don't have answers: is a finger human but a tongue artificial, is autocomplete permitted because we stopped noticing it while an LLM is forbidden because the machinery is still visible, how many letters must a person select before the sentence qualifies for citizenship. "Authorship has never lived in your tendons."

Where it does live, on his account, is intention, judgment, and responsibility. Who wanted to say this. Who supplied the ideas. Who chose what stayed and what disappeared. Who recognised when the machine produced nonsense. Who accepted the consequences of publication. Who can explain the argument without asking Claude to attend the meeting on their behalf. Those questions work for a pen, a secretary, feet, a tongue-driven cursor, eye tracking, a screen reader, or a rewritten paragraph, and they also catch the people who publish pages they haven't read and hide behind automation when challenged. He compresses it to: not human versus machine, but judgment versus surrender.

The carve-out

Shehu does not claim unaided writing never needs testing. Schools need to know whether students can form arguments, an employer may want assurance that someone can perform when the tools fail, and a novelist may decide that writing every sentence is part of the art. His condition is that you name the ability being tested, build the conditions for it, and assess that. What he rejects is promoting unaided typing into a general certificate of humanity, which he calls technologically backward and needlessly exclusionary: it protects one production method by pretending the method is the mind.

Where this sits against the vault's other positions

The judgment-versus-surrender line is nearly the same test as credibility-as-slop-test, which holds that AI-assisted text can't be judged from its surface and that the working question is whether the author would read it aloud, unashamed, staking their name on it. slop-marker-convention runs the same logic for code, defining slop by the production process rather than by how the output looks. And ai-detector-arms-race supplies the empirical half of Shehu's complaint: surface detection fails in both directions, its tells change with every model release, and its stable outcome is writers degrading their work to pass. human-made-disclosure covers the declare-human marker, which is unverifiable by construction and useful as a stance rather than as evidence — which is roughly the status Shehu would assign it.

The real disagreement is with a-voice-from-nowhere, which argues from Leo XIV's encyclical that generated work copies the form of thought without the path underneath it. Shehu's opening concedes that the path exists and matters — the seventh sentence being better than the first is exactly that claim — but he treats it as one route to a destination rather than the destination itself, and he is not obliged to explain what happens to a writer who never walks it. That question belongs to cognitive-debt and dont-outsource-learning, which document the cost of skipping the process, and it is the same cost the schools carve-out silently concedes. His piece is about how to judge a finished text and by whom it is owned, not about what producing it does to the producer, and the two arguments are only in conflict if you read either as answering both.