I sometimes think that no one in Japan is better suited to generative AI than the staff of one particular national daily. For many of them, work stopped meaning “thinking for yourself” a long time ago.

The boss said so. That is how it has always been done. The other papers are doing it too. We wrote it that way last time. The editor changed it. The department head decided. It is company policy. There is no shortage of such explanations. The one question nobody seems to need is, “What do I think?”

When generative AI arrived, I doubt they saw it as much of a threat. It produces plausible prose without requiring them to think. It produces plausible explanations without requiring them to understand how a subject works. It will even supply plausible questions. All of this takes seconds. The Sinking Newspaper could hardly have asked for a machine better matched to its habits.

AI did not rob this organization of its capacity for thought. The organization had already stopped thinking for itself; then a machine arrived that looked as if it could think. That is what bothers me.

The absence of thought at this newspaper also makes me wonder about the education of the people who work there. A conspicuous number of reporters graduated from humanities or social-science programs at prestigious private universities, with Waseda and Keio at the top. I am not claiming that a private-university humanities degree makes a person incapable of logic. My quarrel is with Japan’s entrance-exam system and the habits it rewards.

Many private humanities programs have admissions routes that require no mathematics. A student can also avoid physics, chemistry, and biology altogether, gaining admission to a famous university on the strength of English, Japanese, and either Japanese or world history. For some, the last serious encounter with mathematics or natural science comes in middle school, or in the compulsory part of high school. They can graduate, acquire the label “highly educated,” and join a major newspaper with an understanding of mathematics, physics, chemistry, and biology that has barely advanced since their early teens.

What they often do very well is memorize dates, names, events, political systems, English vocabulary, classical Japanese vocabulary, and then reproduce all of it quickly under exam conditions. They can retain huge numbers of proper nouns, spot the right answer, recognize patterns in past papers, and anticipate what the examiner wants. Those skills are excellent for passing entrance exams.

A child from a relatively comfortable family in or around Tokyo may attend cram school from an early age and grow up surrounded by admissions advice. Such a child soon learns how to play the game: which school to target, which study guide to buy, which subject to drop, which route avoids mathematics, which university requires no science, and how to reach the most prestigious degree by the shortest path.

There is a rationality to this. It is simply different from the rationality needed to understand the world. The student learns how to optimize inside rules written by somebody else. The question, the correct answer, and the examiner are all supplied in advance. There is no need to formulate a question or challenge the rules. The winner reaches the examiner’s answer fastest.

Mathematics and science offer more than formulas and the periodic table. They teach a person to form a hypothesis, quantify, separate variables, distinguish correlation from causation, change the conditions, look for counterexamples, allow for measurement error, and ask whether a result can be reproduced. They also teach a harsher lesson: when the data contradict the explanation you prefer, you throw out the explanation.

A science degree does not automatically make someone logical, just as a humanities degree does not automatically make someone illogical. Still, Japan offers an elite educational track that allows students to avoid this kind of training from adolescence onward. The effects sometimes surface in the newspaper.

Japan’s national dailies can be obsessively detailed when they cover politics, crime, or accidents. Factional alignments, ministry rankings, police and prosecutorial procedure, the private language of Nagatacho and Kasumigaseki: terms understood only by beat reporters turn up with almost no explanation. Readers are expected to follow the reporting all the way down into the minutiae.

The tone changes with science. “Ordinary readers will find this too difficult,” an editor says, and the explanation is cut down until it explains very little. Some cuts go so far that they change the meaning. Certain stories leave the suspicion that the technical detail was not removed for the reader’s sake. The reporter may never have understood the subject well enough to explain it.

A political reporter may know factional alignments and party appointments in eye-glazing detail. A crime reporter may command an entire vocabulary of police organization and criminal procedure. A science reporter, meanwhile, can be sent into the field without a working grasp of probability, statistics, genetics, cell biology, biochemistry, or physics.

That reporter cannot assess the research. They may lack the background to understand the scientist, read the paper without knowing which part of the method matters, and have no idea what to examine statistically. The meaning of a control experiment, the importance of reproducibility, and the limits of what the data support may all be lost. Science reporting then turns into the collection of stories.

A young researcher, a breakthrough, a world first, struggle, dreams, adversity, a white coat, a laboratory. These are the familiar pieces of the newspaper’s human-interest story. A reporter can assemble them without understanding much of the science.

The early coverage of STAP cells is a good example. Newspapers and television celebrated Haruko Obokata and STAP cells on a huge scale. Several things fed that response: RIKEN’s authority, the news value of a discovery said to overturn conventional wisdom, the appeal of a young female researcher, the imprimatur of Nature, and competition among news organizations. Scientific illiteracy alone does not explain what happened.

Even so, would the mythology have been built quite so eagerly if more reporters in those newsrooms had been able to judge the research for themselves? Instead of testing the scientific claim, the media took a story from a prestigious institute, “A world-changing discovery has apparently been made,” and made it bigger.

The habit extends beyond science reporting. An authoritative source gives an explanation. The reporter assumes they understand it, fits it into the profession’s standard form, and sends it back out. This resembles the entrance-exam habit of finding the sanctioned answer and reproducing it correctly. It also happens to be useful for surviving inside a large company.

On the entrance-exam circuit, students learn to read the examiner’s preferred answer; at the newspaper, employees read the boss’s. School rewards a higher ranking score; the company rewards a higher personnel rating. Students study the tendencies of an exam; employees study the mood of the organization. The classroom has become an office, but the game is much the same.

The company has correct answers, and the boss knows what they are. Personnel reviews are grading, promotions are ranking scores, and assignments are pass-fail decisions. An employee learns what to say, whom never to contradict, which department offers promotion, which superior is safest to work under, and which assignment will earn points. The ability to solve the given problem remains useful.

What an employee must not do is question the problem itself. Why are we doing this work? Why is this person important? Why does this system still exist? Why is this number our target? Why are we still producing this newspaper? Ask those questions and you stop being a person who solves problems. You become a person who causes them.

I half-jokingly call this type the shidai-bunkei ojisan, the “elite private-university humanities guy.” The phrase names a Japanese type shaped by an exam track that can largely avoid mathematics and science. The academic discipline itself is beside the point. He memorizes, reproduces, locates the authoritative answer, senses what the room wants, and reads the examiner. He leaves the premises alone.

Much of what generative AI now automates is work he had already been doing by hand. Store existing information, consult old patterns, infer the answer the situation demands, and turn it into plausible prose. The only difference was whether the processor happened to be a person or a machine.

Generative AI performs that sequence far faster and on a far larger scale. It searches, writes, translates, handles mathematics and statistics, and processes enormous datasets. Competition is almost beside the point. The machine automates the work he has long regarded as his specialty.

What ought to remain is the work that needs a human being: doubt, unease, an imagination for people with less power, resistance to the organization’s assumptions, and questions nobody else has asked. Those are the very capacities the Sinking Newspaper has neglected most.

People who possess them are often treated as nuisances. They contradict superiors, question irrational customs, point out when someone with less power is being treated unfairly, refuse to accept “that’s just how things are done,” and mention facts the company would rather avoid. None of this helps in the promotion game.

The people who thrive are often those who read the internal hierarchy quickly and absorb it into their moral judgment. A department head must be important. A bureau chief must be talented. Anyone who failed to rise must have something wrong with them. A non-regular employee ranks below a regular employee. Someone the company does not value can safely be treated as lesser. A bureaucratic ranking quietly becomes a ranking of human worth.

The imagination required to question that substitution disappears. People stop seeing the advantages that shaped their own lives or the protection they receive inside the organization. With the same shoe that happened never to step on them, they step on someone below. Often they do not even recognize the act as cruel. “It’s work.” “It’s the company.” “That’s how evaluations work.” “It can’t be helped.” No further thought is required.

This kind of organizational intelligence fits generative AI frighteningly well. Suppose a superior says, “Come up with ideas for making newspapers appealing to young people.” The first questions should probably be: Do young people need newspapers? Is the problem with young people, or with the newspapers that fail to offer them anything worth reading? Why do we assume anyone should pay for what we produce?

Such questions make the assignment inconvenient, so it is easier to ask AI: “Give me ten innovative ideas for marketing newspapers to Gen Z.” Back come social-media integration, short-form video, explainer content, university partnerships, youth events, and influencer collaborations. The same familiar proposals arrive instantly in a tidy list.

Nobody is offended. Neither the newspaper as a product nor the company’s reason for existing comes under examination. Yet the organization retains evidence that “we thought about it.” The most dangerous use of generative AI is giving up the question itself. It allows an organization to proceed without ever having asked one. Journalism was supposed to ask what nobody else had thought to ask, rather than collect the largest number of answers.

Having written that, I wonder whether I am romanticizing the old newspaper. Journalism was once supposed to distrust the plausible stories circulating through society: government announcements, corporate publicity, police statements, political speeches, market narratives, public opinion. A reporter would ask whether the number was true, whether the causal claim held, who released the information and with what interest, whether contrary evidence existed, where the primary source was, and whether another method could verify it. The plausible statement had to be taken apart, tested, and put back together.

The Sinking Newspaper lost much of that ability long ago. The government announces, a ministry explains, investigators speak, a politician comments, and the newspaper publishes. Reporters do further reporting and contact other sources, of course. That still falls well short of fact-checking in the stronger sense: questioning a claim’s premise and testing it against independent evidence.

I once heard an executive at the Sinking Newspaper say, “We fact-check too.” What he described was, as far as I could tell, what Japanese reporters have traditionally called uratori, or corroboration. One source says something, another source is asked about it, a person involved is contacted, a number is checked. These are necessary parts of reporting. They are not the whole of modern fact-checking.

A fact-checker breaks a claim into propositions that can be tested, then consults primary sources, public data, original documents, and evidence independent of the people making the claim. The aim is to decide whether the claim is true, false, or misleading. “Someone else told me the same thing” does not settle it. Three bureaucrats may repeat an explanation because all three read the same document, serve the same institutional interest, or share the same error. Asking one representative of power to confirm another representative’s account does not produce independent verification.

Before asking whether the Sinking Newspaper fact-checks, one may need to ask what its executives think fact-checking means. This goes beyond terminology. What does it mean to doubt an official announcement? What does independence from a source require? How do testimony and evidence differ? What separates “several people said this” from “this has been established as fact”? These questions concern the newspaper’s idea of journalism itself.

Without them, a procedure begins to look like truth. Source A said it. Source B said it too. Their accounts match, so publish. Procedure completed; correct answer. The world is not an entrance exam. Its answer is not waiting somewhere on the question sheet, and the person who wrote the question may be lying. A journalist’s work was supposed to begin by remembering that possibility.

Journalism in that sense has been weakening at the Sinking Newspaper for years. This is why talk of AI “destroying newspapers” sounds slightly wrong to me. Much of what might be destroyed has already been hollowed out. AI may instead make the existing damage harder to see.

It can summarize government announcements, organize ministry documents, transcribe press conferences, search related coverage, draft stories, propose headlines, and write social-media copy. The newspaper can produce more material, faster and more cleanly. It may even look smarter. The arguments are tidier, the prose smoother, the typographical errors fewer, and foreign-language sources may appear. But the reporter’s own question is still missing.

AI is very good at answering a question it has been given. Trouble starts when people also delegate questions such as “Why are we asking this?”, “Who defined the problem this way?”, “What if the premise is wrong?”, and “Who disappears when we describe the issue in these terms?” Once those are handed over, the human being becomes an approval mechanism.

AI generates ten ideas and a person selects one. AI writes the proposal and a person pastes it into a document. AI drafts the story and an editor polishes it. AI proposes the headline and someone adjusts it. AI summarizes the result and a person checks it. Eventually the job may involve no judgment at all. The human role shrinks to looking at the output and saying, “Yeah, this seems fine.”

The meetings will continue and the documents will still be produced. Projects will acquire names such as “AI Transformation,” “Next-Generation Newsroom,” “Digital First,” or “AI-Native Editorial.” The name hardly matters. AI can supply that too.

Inside the organization, nothing changes. Those at the top issue instructions and those below carry them out. Failure travels downward; credit travels up. The hierarchy survives, dissenters remain troublesome, and people with less power absorb the burden. Nobody questions the premise. The only improvement is that the organization stops thinking faster.

AI is often sold as a way to “free people for creative work.” Give routine tasks to a machine and people will have more time for serious thought. The logic is sound. But what if the people involved never wanted serious thought? Will they use the extra time to think, or schedule more meetings and produce more reports? Perhaps one AI will write a document, another will summarize it, and a third will reduce the summary to three bullets for a person to read and say, “I see.”

Technology does not make an organization intelligent by itself. It magnifies whatever was already there. A curious newsroom can test more hypotheses. A critical thinker can use AI to challenge their own assumptions. A good reporter can compare masses of documents, analyze data, conduct open-source research, and support verification.

In a company that prizes obedience, AI makes obedience faster. A precedent-bound company can reproduce its precedents more quickly. In a company that avoids responsibility, “AI says so too” becomes one more excuse. When an organization has stopped thinking, fluent text on a screen helps conceal the fact. A person can think nothing and look thoughtful, understand nothing and sound informed, make no real decision and still possess a “strategy.”

Now the private-university humanities guy has a problem. For years, he made a career from memorized information and past examples. He anticipated what his superiors wanted, read the room, found the approved answer, arranged information in the expected format, and produced plausible prose. People called this the work of an intelligent professional.

He had a famous degree. He could write and knew difficult words. He remembered the names of politicians and bureaucrats, understood the office’s internal politics, and could say something suitably impressive in a meeting. He could therefore think of himself as a knowledge worker.

Generative AI did the same work with humiliating ease. It stores and retrieves enormous amounts of information, recognizes patterns, organizes the main points, writes, translates, summarizes, proposes projects, and invents headlines. It works faster than a person, does not tire, handles mathematics and statistics, writes code, and reads documents in foreign languages.

That raises an unpleasant question: Was what he had always called “intelligence” really intelligence? Perhaps much of it was the ability to memorize information, fit it into existing patterns, and produce the answer somebody else expected. Generative AI is exceptionally good at that.

The human contribution still matters. AI simply exposed how much supposedly human work had never needed a person in the first place.

What remains is the capacity to doubt, sense that something is wrong, question a premise, challenge an authoritative explanation, think about causes rather than repeat numbers, imagine a life wholly unlike one’s own, formulate a question nobody supplied, and accept personally inconvenient facts. It also means looking at an organizational “truth” and asking, “Is it?”

That may have been the real value of human intelligence all along. Yet the Sinking Newspaper has spent years refusing to reward it. Employees who contradict superiors, question precedent, object to company policy, see an issue from a minority’s perspective, or reject “everyone does it” as an argument are commonly dismissed as difficult.

The Sinking Newspaper rewarded the abilities AI can replace and suppressed those it cannot easily replace. Memorize, summarize, organize, write plausible prose, give the boss the answer he wants: high marks. Doubt, object, ask questions, depart from convention, resist authority: low marks. Then a machine arrived that performs the first group of tasks far faster than any of them.

AI may do more than put the private-university humanities guy out of work. It does something more brutal: it abruptly lowers the value of what he spent decades believing was his intelligence. Article summaries, headline suggestions, project proposals, meeting papers, emails, ceremonial speeches, interview questions, translations of overseas coverage, archive searches, routine argument maps: AI can produce them in seconds. One question remains for the human being. So what do you think?

The Sinking Newspaper may find this the hardest question of all. A newspaper was supposed to distrust society’s plausible stories. Having lost much of that ability, this one is now acquiring a machine that can produce plausible language without limit. If its users have forgotten the difference between plausibility and thought, calling this modernization would be generous. It would merely finish hollowing journalism out.

I have no objection to newspapers using AI. They should use it aggressively. They should also bring their own questions. Why are we writing this story? Why interview this person? Why use this number or trust this source? Why is this on the front page and that is not? Why repeat the government’s explanation? Why does this company exist? Why have people stopped reading newspapers? Why are we so sure we are right?

What if we are the ones who are wrong? The ability to ask that question marks the difference between using AI and hiding behind it.

The Sinking Newspaper needs something more primitive than the newest AI. It needs the ability to doubt, investigate, falsify, read numbers, think about causation, resist authority, question the common sense of its own group, imagine the lives of people with less power, and think for itself. It must also treat its own newspaper as an object of investigation.

If it can do that, generative AI may help save journalism. If it cannot, AI will be the perfect technology for the Sinking Newspaper: a way to forget that it is sinking, imagine that it is modernizing, produce more intelligent-looking language than ever, and go under with unprecedented efficiency.