Governance Through Uncertainty

Opacity is a property of a system. Uncertainty is a state of a person.

Jason C. Lau · July 14, 2026 · About a 13-minute read · This essay states, in public form, the argument of my paper at the 2026 ACM Conference on Fairness, Accountability, and Transparency.

One night in 2016, I was having dinner with friends in Sanlitun, one of the trendiest districts of Beijing. It was that night I first heard of Sesame Credit (芝麻信用). My friends showed me the app: a colorful gauge, a score between 350 and 950, perks unlocking as the number rose — discounted bike-sharing, fast-track visa applications, better placement on dating apps. Built by Alibaba from the data of its internet empire, it was designed to feel like a game; the name itself winks at “open sesame.” What determined the score, nobody could quite say. An Alibaba executive once suggested that buying diapers might raise it, since parenting signals responsibility, while playing video games for ten hours a day would likely lower it. The confidence with which behavior was mapped to character was striking. An explanation, it was not.

My friends loved it. Like many young professionals in Beijing — graduates of good universities, holders of well-paid jobs, beneficiaries of the boom — they had high scores, and the app repaid them in perks. They made fun of me for being backward about new technology. Then I asked whether any of the information these apps collected would be sent to the government.

“这个不好说 (zhe ge bu hao shuo),” one of them replied, and quickly changed the topic.

Bu hao shuo. The phrase literally means “hard to say” — expressing uncertainty. But it also means “not good to say” — implying something that cannot or should not be spoken. I spent years of ethnographic fieldwork in China coming to understand that this ambiguity was not an evasion. It was a description. No one knows exactly what is monitored. No one knows exactly how the scores are calculated. No one knows exactly what consequences follow from what behaviors, or where the line lies between what the state sees and what the platforms see. And no one can safely ask, because asking is itself the kind of thing that might be noticed. The uncertainty is not a bug in the system. It is the system’s governing logic.

I have come to call this governance through uncertainty: a mode of control in which indeterminacy about rules, boundaries, and consequences functions as the governing mechanism itself. A system with clear rules can be navigated, tested, gamed, resisted. A system whose rules remain uncertain produces something more useful to power: people who govern themselves, who internalize vigilance, who anticipate prohibitions before they are announced. Such a system does not need to watch everyone continuously. It needs only to make everyone feel continuously assessed, and to keep the consequences of transgression unpredictable.

The microworld

To see how such a system gets built, look at the world it encloses. I borrow a word from the historian of computing Paul N. Edwards: the microworld — his term for the simulated environments inside computers, worlds “without irrelevant or unwanted complexity,” internally consistent and externally incomplete. In my use, the term captures a paradox: a simplified system that has become an inescapable world. Micro, because the environment is radically simpler than the social world it mediates — life rendered as scores, flags, and categories, computationally manageable even when its inhabitants number in the billions. World, because this simplified environment has become the infrastructure of daily existence: payments, transit, communication, social life.

WeChat is the clearest case. It began as a messenger and became — in Tencent’s own words — “a way of living” (微信，是一个生活方式): how you pay, how you hail a taxi, how you reach the government, for over a billion people. Since 2017 it has been tied to your real, state-issued ID. To live in a major Chinese city without it is nearly impossible. Inside such an environment, the rules operate at several registers at once — the criteria that score you, the thresholds that admit you, the triggers that punish you — and each can be opaque on its own. You may know how to get in but not how you are ranked once inside; you may know that penalties exist but not what calls them down. And the stakes are real: digital wallets frozen without clear explanation; people blacklisted by the courts blocked from buying plane tickets more than seventeen million times in a single year.

Sesame Credit, it should be said, was not the government blacklist, and China never operated the single unified national score of Western imagination. Commercial scoring and state enforcement grew up alongside each other — and what mattered, for the people inside, was that no one could say where one ended and the other began. That, too, is the condition.

A microworld, then, has three defining features. It is rule-governed: behavior inside it is structured by logics that may not correspond to any rules outside. It is consequential: its categorizations reach your money, your movement, your life chances. And it is epistemically opaque: you know it has rules, but you cannot fully see what they are or when they change. That combination makes a microworld a space of governance, not merely participation.

When uncertainty becomes infrastructure

The most atmospheric case is censorship. People in China described it to me, again and again, as being like the weather — unpredictable, ever-present, something one learns to live with rather than understand. The metaphor is precise. A topic that circulated freely yesterday may be unsearchable today. A group chat that discussed current events for months may vanish overnight without explanation. And there is a structural reason the rules can never be collectively mapped: research on Chinese censorship has found that the system targets not criticism but coordination — and pooling knowledge about where the lines lie is itself a form of coordination. Everyone navigates the weather alone.

During the pandemic, this logic materialized into infrastructure that governed physical movement itself — an account I take from documentary reporting and informal conversations rather than systematic fieldwork. By March 2020, entering even a convenience store in Beijing required scanning a QR code. The code — green, yellow, or red — appeared on your phone within seconds and determined whether you could proceed or be turned away. Within weeks, health codes (健康码) were mandatory across China: transit, office buildings, residential compounds, hospitals. The system stitched together databases that had previously been separate — health records, telecom data, transportation logs, identity — into an algorithmic layer. A person could wake with a green code and discover, hours later, that it had turned yellow or red, without explanation or any apparent change in circumstances. Because the system computed an overall “risk” rather than detecting specific exposures, the codes were basically unfalsifiable: there was nothing to check your categorization against, and no way to effectively contest it.

Then, in 2022, the infrastructure showed what else it could do. When depositors traveled to Zhengzhou to protest frozen accounts at rural banks, their health codes turned red on arrival — immobilizing them. By the local discipline commission’s own count, 1,317 depositors received red codes. Some had never traveled at all; they had merely scanned a location code that other depositors shared in a chat group. No new system had to be built. The existing algorithmic layer simply received different inputs.

Local officials were later punished for what state media called an “abuse” of the health code system. But notice what the punishment was for: this particular misuse becoming visible — not the system’s capacity to enable it. The infrastructure remains. The uncertainty about whether any given code change is legitimate health policy, bureaucratic error, or political targeting persists. Individual abuses can be sanctioned; the condition that enables them stays intact. This is the point on which everything else turns: governance through uncertainty does not require a conspiracy. It requires infrastructure.

Beyond the panopticon

It is worth being precise about what is new here, because arbitrary power is old. The modern theory of surveillance is Foucault’s panopticon, and the panopticon also runs on uncertainty: the prisoner never knows whether the guard is watching, and so behaves as if he always is. But that is uncertainty about surveillance, not about norms. The prisoner knows perfectly well what the rules require; only the watching is in doubt. In Foucault’s account, modern power works through a double legibility — people are made knowable to power, examined, counted, classified, while the norms are made knowable to people, so that they can regulate themselves. Governance through uncertainty keeps the first half and dismantles the second. You are more legible to the system than Foucault could have imagined: real-name registration, platform data, algorithmic profiling. But the norms, the criteria, the thresholds are illegible to you. You know the direction of acceptable conduct — be compliant, be cautious. What you cannot know are its coordinates: what counts as compliant enough, which behaviors move which scores, where exactly the red line (红线) falls.

And computation adds the twist that closes the last exit: personalization. Your score, your code, your feed. Under older forms of arbitrary rule, people could at least learn from one another’s fates — watch what happened to a neighbor and adjust. When evaluation is individualized, even that is gone. Arbitrary power could always make the rules unknowable. Only computation makes them unknowable and individual. The result is a system that does not only produce subjects who follow rules. It produces subjects who never stop trying to find them.

Involution and lying flat

That last sentence is not a flourish; it is a description of daily life, and the Chinese internet has given the experience two names. 内卷 (neijuan, “involution”) — a word borrowed from the anthropologist Clifford Geertz, who used it for agriculture that grows ever more intricate without becoming more productive — exploded in Chinese discourse around 2020, naming the grueling competition among students and workers: longer hours, more credentials, ever-higher stakes, no improvement in prospects. The anthropologist Xiang Biao described it as a spinning top, whipping itself to keep spinning in place. Most discussions tie involution to scarcity — too many people chasing too few places. That is accurate but incomplete. Life inside the microworld is thoroughly quantified: the scores are visible, the rankings are visible, the distance between you and everyone else is a precise and public fact. But the rules that generate the numbers stay hidden. Because everything is measured, the pressure to optimize is inescapable; because the criteria are unknowable, optimization is impossible. If any behavior might count, and none is known to count more than the others, the only safe strategy is total effort. Involution, understood this way, is compulsory striving under epistemic deprivation — the exhausting imperative to optimize without knowing what optimization means. It has other causes too: economic scarcity, demographic pressure, expectations around achievement. The not-knowing does not replace them; it sharpens them, because effort that cannot be targeted cannot be finished.

The counter-move also has a name: 躺平 (tang ping, “lying flat”) — the refusal to keep optimizing a game whose rules cannot be read. Not exit from the microworld; there is no exit. Withdrawal from striving within it. And in 2025, China’s internet authorities announced a campaign against online “negative emotions,” targeting narratives such as “effort is useless” and “studying is useless” — not because these expressions mobilize protest, but because they articulate non-participation as a legitimate choice. That campaign tells you what a system like this actually fears. It does not fear people who play badly. It fears people who stop playing.

Familiar ground

The temptation, for a Western reader, is to file all of this under authoritarianism and set it safely elsewhere. I want to resist that — not because the political differences are small, but because the Chinese case makes something visible that is much harder to see at home. Here is the uncomfortable symmetry: Western systems maintain the rhetoric of transparency while becoming steadily less transparent. Chinese systems have more openly abandoned the pretense. In China, opacity is not hidden as failure. It is acknowledged as governance.

So run the test on familiar ground. A content creator watches her reach collapse and cannot learn why. A rideshare driver watches the payouts shrink, ride after ride, and cannot learn what changed. The outcomes are perfectly visible — the score, the lower payout, the silence where an audience used to be. The criteria that produced them are not. Community guidelines are stated at a level of generality that guides virtually nothing; enforcement diverges from any reasonable reading of the rules; policies shift without notice, outdating whatever you thought you had figured out. And the platform can infer far more about you than you can ever infer about it. Outcome visibility, rule illegibility, dynamic shifting, epistemic asymmetry: the four properties of the condition I documented in China, present in structurally analogous form. What differs is the weight, and the institutional surround. The stakes differ; the exits differ; the recourse, when it exists, differs. In liberal democracies the cost of an opaque decision is usually economic rather than existential, and some way to object, however imperfect, exists. That difference is precisely what is worth protecting — which is why it matters to understand the mechanism now, while the difference holds.

Who can afford illegibility

Seeing the mechanism changes the question we ask about fairness. For a decade, the central worry about algorithmic systems has been bias: whether the outputs treat some groups worse than others. That worry is real. But governance through uncertainty stratifies in a way no audit of outputs will catch, because it was never just in the outputs. It stratifies through the differential capacity to navigate opacity. Those with resources, connections, and standing bear uncertainty better: they have informal channels, they can absorb a mistake, they know people who know things. In my fieldwork the contrast was stark: the young professionals at that dinner moved through the system with ease; the migrant delivery workers I spoke with could not. The weather of censorship is the same weather, but some people have umbrellas. And opacity is not even neutral in how it distributes harm: for the well-positioned it is a resource — an edge over everyone who must simply bear it — which is part of why the dark is so durable. Governance through uncertainty is never equally uncertain.

Accountability without full transparency

If this is right, it asks something specific of the people who build, study, and regulate these systems. Start with what the concept does not claim: opacity alone does not make a microworld. What turns opacity into a governing mechanism is opacity that cannot be escaped and cannot be appealed — that is the combination to watch, in any system, anywhere.

It also matters where opacity comes from — and increasingly, it comes from the technology itself. Some opacity is political: information withheld for control, as in much of the Chinese case. Some is commercial: proprietary systems and trade secrets. But some is structural — what Jenna Burrell identifies as opacity arising from machine learning itself: systems whose operative logic cannot be rendered in terms a person could use, and whose effective criteria shift as models are retrained, in ways difficult for users — and sometimes for their own designers — to interpret. Yet the mechanism I have described does not care which kind of opacity it faces. Whether the rules are hidden for political reasons, protected for commercial ones, or illegible by architecture, the person underneath is in the same position: navigating a consequential system whose criteria cannot be known. Computer science has a rich literature on strategic classification — what happens when people can see the rules well enough to game them. Governance through uncertainty is the inverse problem: high-stakes evaluation under persistent rule illegibility, where the dominant response is not targeted evasion but diffuse, exhausted intensification. As AI systems take over more consequential decisions, that inverse problem becomes the default condition of algorithmic life. Uncertainty is not a defect at the system’s edge, waiting to be patched with disclosure. It is a property of the design — and it should be analyzed as one. The question to ask of any algorithmic system is not only “how can we make it transparent?” but “what work does its opacity do?”

Which is why the remedy cannot only be transparency. Some systems can be disclosed into accountability. Others govern through their illegibility — and for those, explanation will arrive stale or not at all. What can be delivered is different, and older: contestability. The right to challenge a consequential decision and have someone answer for it, even when no one can see fully inside the system that made it. You do not need to understand how a score was computed to deserve a way to appeal it. Accountability cannot wait for transparency.

My friends at that dinner in Beijing were not evading my question. They were answering it — describing, in three syllables, the epistemic condition of the world they lived in. The rest of us are entering that condition more gradually, with gentler stakes and better exits, and with a window in which to build the protections whose necessity the Chinese case makes plain. The place to begin is to name the condition precisely, because we have been calling it by the wrong name. Opacity is a property of a system. Uncertainty is a state of a person. That is why I call it governance through uncertainty — and why the question that should organize the next decade of technology governance is not only how to make these systems legible, but who can afford illegibility, and who pays for it.

@inproceedings{lau2026governance,
  author    = {Lau, Jason C.},
  title     = {Governance Through Uncertainty: What Chinese Algorithmic
               Systems Reveal About the Limits of Fairness, Accountability,
               and Transparency},
  booktitle = {Proceedings of the ACM Conference on Fairness,
               Accountability, and Transparency (FAccT '26)},
  year      = {2026},
  pages     = {4061--4077},
  publisher = {ACM},
  address   = {Montreal, QC, Canada},
  doi       = {10.1145/3805689.3806495}
}