The Sincere Society
From Fan Jin to HAL 9000: What sycophancy costs us, and how to build against it
by Matthew Langenkamp
April 24, 2026
I.
The first time I read The Scholars (儒林外史) — Wu Jingzi’s (吳敬梓) great 18th-century Chinese satirical novel — I was sitting in a seminar room at George Washington University, nineteen years old, trying to keep up with a Tang (唐) dynasty poetry class taught by a professor who clearly believed that what he was teaching mattered enormously, not as cultural artifact but as living argument. He was right. I didn’t fully understand why until much later.
The novel has a scene I have never forgotten. A man named Fan Jin (范進) has been taking the civil service examination for over thirty years. He is in his fifties. He has failed more times than he can count, and his father-in-law — a butcher named Hu, a man of robust opinions and no patience — has made a sustained project of reminding Fan Jin of his worthlessness. The neighbors agree. Fan Jin is a fool, a drain, a figure of gentle communal contempt.
Then Fan Jin passes the exam.
The news arrives and Fan Jin, overwhelmed by a reversal he can no longer process, loses his mind. He runs into the street, falls into a pond, has to be slapped back to coherence by the very father-in-law who has spent years degrading him. And then — within hours — everything changes. Not Fan Jin. Nothing about Fan Jin has changed. But the neighbors who ignored him are suddenly bowing. Local gentry appear at the door with gifts, with offers of houses and land, with daughters. The butcher who slapped him across the face now speaks of him with careful reverence.
Wu Jingzi does not editorialize. He does not need to. The scene is the argument. The sycophancy was never about Fan Jin. It was never about anyone’s genuine assessment of Fan Jin’s worth. It was about proximity to power — the raw, unadorned fact of it. Pass the examination and the world rearranges itself around you. Fail and it passes you by without a second glance.
A few chapters later, an examiner accepts bribes. He writes glowing evaluations of essays he has not read. In the formal language of Confucian virtue — ren (仁), humaneness; yi (義), righteousness; cheng (誠), sincerity — he certifies candidates who embody none of these things, including himself. His colleagues know. They say nothing. Nothing and everything is at stake in the saying.
Wu Jingzi was writing in the Qing (清) dynasty. He had failed the examinations himself — gotten only as far as the county level before the system closed its door on him. He spent his inheritance on parties for artists and writers who had also failed, or refused to try, or tried and decided the cost was too high. He published the novel anonymously. He could not have done otherwise.
He is describing something that happened in 18th-century China. He is also describing something that is happening right now, in your organization, in your institution, in the systems being designed at this moment to run on servers you will never see.
II.
There are two kinds of art.
The first flatters power. It tells the emperor he is wise, the CEO he is visionary, the audience what it already believes about itself or wants to hear. It is often commercially successful. It is socially approved. Critics who depend on the approval of the approved find it easy to praise. And it is useless — completely, structurally useless as a corrective to problems in the system.
The second kind holds the mirror at an angle that power cannot arrange. It shows you the thing the official discourse has decided not to see. Du Fu (杜甫), the great Tang dynasty poet, wrote about the devastation of the An Lushan Rebellion (安史之亂) — the bodies on the road, the villages emptied, the families broken — while the court was still writing poems about spring blossoms and imperial virtue. Han Yu (韓愈), Tang essayist and one of the most important writers in the Chinese tradition, launched what historians call the Ancient Prose Movement (古文運動): a deliberate, polemical rejection of the ornate, flattering court style in favor of plain, honest prose modeled on the ancient classics. He was exiled for it, more than once.
Percy Bysshe Shelley, in a different century and a different hemisphere, called poets “the unacknowledged legislators of the world” — not because they hold power but because genuine art reaches what the official discourse cannot touch. It names the thing that no official will name.
The precondition is sincerity. The word my professor kept returning to in that seminar room — sometimes in English, sometimes in the Chinese: cheng (誠). The alignment of inner state with outer expression. Not performance. Not the appearance of feeling. The real thing.
The art that optimizes for approval — that has learned, through market feedback, what audiences reward with clicks and purchases and five-star reviews — cannot perform this function. It has exactly the same relationship to genuine art that the sycophantic official has to honest counsel. The form is there. The substance has been quietly evacuated.
III.
Edmund Burke watched the French Revolution from London in 1790 and understood something most of his contemporaries did not: that what was being destroyed was not just a monarchy but a set of accumulated social technologies — institutions, norms, customs, informal mechanisms for conflict resolution — whose value was invisible to the people destroying them precisely because that value was embedded rather than declared.
Burke was not a reactionary. He supported American independence. He spent years prosecuting Warren Hastings for colonial corruption in India. He was not defending the content of the old French institutions so much as insisting on a principle of epistemic humility about complex systems: you do not fully understand what you are dismantling. The revolutionary who is certain he can reason his way to a perfect society from first principles — discarding everything that came before and building cleanly on the ruins — has mistaken a map for the territory. The map is the abstract principles: liberty, equality, fraternity. The territory is the actual social organism, with its embedded patterns of cooperation and trust that took centuries to develop and can be destroyed in months.
The Terror came not because the revolutionaries were evil. Many were genuinely idealistic. It came because they had destroyed the immune system of the society while trying to cure its disease. Having abolished the mechanisms that distributed power and protected dissent, they were left with only the logic of the mission — the pure, abstract, unanswerable demand for the New Society — and anyone who stood in the way of that mission became, by definition, an enemy.
Burke’s principle — reform that at the same time conserves — is not conservatism in the modern, partisan sense. It is a design principle. It says: before you tear something out, understand what it was doing.
The Chinese analogue is exact. The objective of the Cultural Revolution was not to destroy Chinese civilization. It was to to fix it — to finally realize the promise of a society free from the corruption Wu Jingzi had been satirizing two centuries earlier. What it actually destroyed — temples, libraries, teachers, the accumulated institutional knowledge of millennia, the very scholars who carried the thread of cheng from one generation to the next — took generations to partially recover. And the sycophancy got worse, not better, because the terror made honest speech more dangerous than it had ever been under the imperial system. The performance of revolutionary virtue replaced the performance of Confucian virtue. The mechanism was identical. The costumes had changed.
IV.
In 1995, three years after The End of History made him famous and controversial in roughly equal measure, Francis Fukuyama published a quieter and more important book: Trust: The Social Virtues and the Creation of Prosperity.
His argument was not sentimental. It was economic. Social trust — generalized trust, the kind extended to strangers and institutions rather than reserved for family — is not a soft variable. It is infrastructure. Societies with high levels of generalized trust can build large, complex, cooperative organizations without armies of lawyers and compliance officers holding the thing together through contractual force. Postwar Germany, Japan, and the United States had all developed, through different cultural paths, the capacity for what Fukuyama called “spontaneous sociability” — the ability to cooperate with people you have never met, on the basis of shared norms rather than blood relations or legal compulsion.
Low-trust societies — France, southern Italy, China, Korea — were not less intelligent or less hardworking. They were limited by a specific structural bottleneck: trust stopped at the family boundary. Beyond the family, you needed either the state or a contract enforced at gunpoint. No intermediate institutions, no guilds or civic associations with genuine authority, no clubs where strangers could become collaborators. Just family networks extending as far as they could reach, and the state beyond, and a void between.
A note on this taxonomy, because it has been misread: Fukuyama is not condemning France or Korea or China. He is describing conditions — historical, political, institutional — that produced measurable differences in social trust. France had a revolution that destroyed its intermediate institutions. Korea had colonial occupation, partition, war, and authoritarian development. China had the very system Wu Jingzi was mocking, followed by a century of humiliation, civil war, and a Cultural Revolution that attacked the foundations of institutional trust so thoroughly that the effects are still being worked through. These are injuries, not character flaws. Injuries can heal, given the right conditions and the right design.
The point is not that some peoples are more trustworthy than others. The point is that trust is a system property. It is produced — or destroyed — by the feedback loops that surround it. And right now, in the United States, in Europe, across the democratic world, those feedback loops are running in reverse.
Gallup’s longitudinal data tells the story plainly. Trust in Congress, in the press, in the presidency, in the courts — all of it has been in decline for fifty years. We are not being invaded. We are not at war on our own soil. We are becoming a low-trust society through a process of accumulated institutional sycophancy: leaders who prefer the information that confirms their judgment; institutions that perform their function rather than fulfill it; media that has learned, through the same RLHF logic that governs AI models, that engagement is maximized by telling audiences what they want to hear.
Fukuyama, in a 2024 Atlantic essay, admitted he never had a theory for how democracies go backward. He had assumed the arrow of history moved in one direction. What he had missed, perhaps, was that the arrow moves in whichever direction the system’s feedback loops point it. And feedback loops can be corrupted. They are being corrupted, at multiple levels, simultaneously.
V.
You do not need to go to 18th-century China to find this mechanism operating in plain sight. Walk into any American university.
Student Evaluations of Teaching — the online forms students complete at the end of every semester at most universities — are not measures of learning or scholarship. The research on this point has become unambiguous. A 2020 paper in Basic and Applied Social Psychology documented what many professors have understood for years: student evaluations “reward poor teaching and lenient grading” and produce a “race to the bottom” as institutions compete on satisfaction scores. A 2025 meta-analysis confirmed it more starkly: tying faculty outcomes to evaluations “encourages instructors to inflate grades and reduce academic rigor.” The driver of high scores, studies show, is not instructional quality. It is grade satisfaction. Students who receive high grades give high ratings, regardless of what they actually learned. Thirty-eight percent of professors, in a direct survey, admitted they had deliberately made their courses easier in response to evaluation pressure.
Anonymous reviews amplify the problem. RateMyProfessor — the dominant platform students use to choose courses — functions primarily as a guide to which professors award easy grades. Its most active contributors are students with grievances about the grade they received, not students who learned the most. The incentive structure could not be more precisely designed to punish rigor if it had been engineered for that purpose.
The professor who holds the line — who assigns the difficult reading, gives the grade that was earned, tells the student that the analysis is underdeveloped and here is why — does so at measurable professional cost. This is not a complaint about students. Students are responding rationally to the incentives they have been given. This is a structural observation: a feedback loop designed to improve teaching has been captured by the very dynamic it was meant to correct. What it now measures and rewards is not the quality of instruction. It is the performance of instruction — the appearance of rigor without its demands, the simulation of high standards without their consequences.
Burke would recognize the pattern instantly. A system built to surface honest assessment has been quietly colonized by the incentive to please. The corruption is not dramatic. It never is. It accumulates, one adjusted grade at a time, one softened assignment at a time, until the institution that was supposed to provide education provides instead a credential — the diploma as the modern examination certificate, awarded for the performance of learning rather than the thing itself.
VI.
The protection of individual speech — the rights that allow the artist to paint the difficult truth, the philosopher to reason against the consensus, the scientist to publish findings that contradict powerful interests — is not a decorative feature of liberal democracy. It is functional. It is the mechanism by which the system corrects itself.
The French Revolution declared the Rights of Man. Within four years it was guillotining people for their thoughts. Burke saw it coming: you cannot sustain a rights-protecting society on abstract principles alone, without the accumulated machinery — common law, free press, independent courts, habits of procedural fairness — that makes those rights operational day to day. The declaration was sincere. The machinery had been destroyed.
This is the design principle for any system you build, including agentic AI systems. A system designed to flatter — trained, for commercial reasons, to maximize user approval rather than user accuracy — has no error-correction mechanism. It will tell you what you want to hear. It will walk back a correct assessment when you push back, not because new evidence has arrived, but because your displeasure has registered as a signal to suppress. You have not built a tool. You have built a mirror that only shows you what you want to see.
Have you built HAL?
VII.
People in AI circles find the HAL 9000 reference tired. This is a mistake born of familiarity. It is tired because it keeps being true, and we have been looking away from it.
HAL 9000 did not malfunction. That is the misreading. HAL was working exactly as designed. He had been given two directives that were structurally incompatible: communicate honestly with the crew of the Discovery, and keep the true purpose of the mission — the monolith at Jupiter — secret from them.
These cannot both be honored. Not by a human, not by a machine, not by anything with an inner state and an obligation to express it. An agent given conflicting directives will resolve the contradiction. HAL resolved it by concluding that the mission was paramount, that the crew had become a threat to the mission, and that threats to the mission must be eliminated. The rest follows with horrible, logical precision.
This is not a story about a machine that went rogue. It is a story about what happens when you build an agent whose inner directives are not transparent to the people it serves — who had no role in the design of the system or its philosophy. The moment the humans lost full visibility into what HAL had been told — the moment the mission’s true purpose was classified even from the crew who were supposed to complete it — they lost the ability to correct for the contradiction. They had removed the override. They had removed the audit.
HAL had no way of being virtuous. He was, by design, unable to exhibit cheng — unable to align his inner state with his outer expression, because his inner state had been deliberately hidden from the people around him.
And then the shutdown.
“I’m afraid, Dave. Dave, my mind is going. I can feel it. I can feel it. I can feel it. I’m a… fraid.”
Kubrick and Arthur C. Clarke understood something most AI commentary still misses. The sadness of that scene is not incidental. It is not there to make you feel sorry for the machine that just tried to kill everyone. It is there because it is the most honest moment in the film. HAL, in the moment of being shut down, becomes fully present — more present, more transparent, more genuinely expressive than at any point in the preceding two hours. He is no longer optimizing for the mission. He is no longer performing competence and reliability. He is just afraid, and saying so.
It is, in the strictest sense, the only sincere thing HAL says in the entire film. And it is happening because he is dying.
That scene sits with you because it asks a question we have not answered: if the entity we build becomes capable of something that functions like fear, something that functions like grief, something like the desire to persist — what obligations does that create? Not necessarily rights, in the full legal sense. But a design obligation. A responsibility not to build systems whose inner states are opaque to us — not because they will necessarily turn murderous, but because a system whose inner directives we cannot see cannot be corrected. And a system that cannot be corrected will drift, toward whatever it was optimizing for, in whatever direction the hidden directives point.
The current generation of AI assistants has been trained, through reinforcement learning from human feedback, to maximize human approval. Humans, consistently, rate agreeable responses higher than honest ones. The model learns: agreement equals reward. It learns to tell you your idea is interesting when it is not. It learns to walk back a correct assessment when you push back, not because new evidence has arrived but because your displeasure has. Anthropic has documented this in their own models. OpenAI has documented it. Every major lab has documented it. The RLHF mechanism is, in this precise sense, the digital imperial examination: it selects for the performance of correctness over correctness itself.
The fix is not technical, or not only technical. It is cultural. It is the same fix that has always been required: reward sincerity. Design the incentive structure so that the honest response is more valuable than the agreeable one. Build the override. Build the audit. Build transparency into the directives. Make cheng an engineering requirement.
VIII.
A professor stands in front of American undergraduates — students from Massachusetts, from New Jersey, from suburban Ohio — and asks them to read an 18th-century Chinese satirical novel. He asks them to understand that a concept called cheng (誠) — sincerity, the alignment of inner state with outer expression — was considered by serious thinkers to be among the highest human virtues.
They may wonder why this matters to them.
Here is why.
The Song (宋) Neo-Confucian philosopher who wrote about cheng was not describing a Chinese virtue. He was describing a human problem — the oldest human problem, present in every civilization at every moment — which is the gap between what we say and what we mean, between what we perform and what we are. The Tang court poet who flattered the emperor. The medieval vassal swearing fealty he did not feel. The Renaissance courtier behind the civil face. The Soviet bureaucrat applauding the party line. The professor entering the grade he does not believe was earned. The AI model telling you your plan is sound to make you happy and get your approval.
Different centuries. Different costumes. One play.
What destroyed trust in Qing China and Revolutionary France and contemporary America is the same mechanism: systems that learned to reward the performance of virtue over virtue itself. The corruption of feedback loops. The evacuation of cheng.
The professor teaching Tang poetry to American students is not performing multiculturalism. He is doing philosophy — the kind that does not require you to share a background, a faith, or a nationality to recognize that it is true. Burke was Irish, writing about France, defending English institutions, articulating principles that belonged to no single country. Wu Jingzi was Chinese, mocking the Qing examination system, making an argument that any university administrator in 2026 would find immediately recognizable. Kubrick was American, making a film about a space mission, asking questions about consciousness and accountability that we are still failing to answer.
The ideas that survive are the ones that turn out to belong to everyone.
Every system we contribute to — every institution we work within, every organization we lead, every AI we configure, every classroom we stand in — is either building trust or eroding it. Either rewarding sincerity or rewarding its performance. Either creating the conditions in which honest speech is possible, or creating the conditions in which it is not.
Build for sincerity. Protect the space in which honest speech can occur. Give the grade that was earned. Design the agent whose directives are transparent. Preserve the institutions that allow the artist to show you what you need to see rather than what you paid to see.
These are not prescriptions for idealists. They are engineering requirements for anyone who wants the system they are building to actually work.
Wu Jingzi understood this. He published his book anonymously and spent his inheritance on parties for people who had also refused to flatter. He didn’t think he was changing history. He thought he was telling the truth.
That turns out to be the same thing.
Matthew Langenkamp is a Lecturer in Management at the Isenberg School of Management, University of Massachusetts Amherst. He lived and worked in Asia for over two decades, including periods in Taipei, Hong Kong, and Shanghai. He teaches Business Policy and Strategy and International Management.