The Direction of Trust

A woman applies for a loan.

She has done everything the form asked. She has entered her income, her history, her debts, her address. Somewhere, in a system she will never see, her application joins everything else that is known about her: years of transactions, the neighborhoods she has lived in, the times of day she shops, the company she keeps as inferred from a thousand incidental signals she never knew she was emitting. The system considers all of it. The system knows her better, in certain narrow and consequential ways, than her own family does.

The system says no.

She asks why. There is no why. The model is proprietary. The data sources are confidential. The decision logic is a trade secret, and possibly, by now, not fully explicable even to the people who built it. She is entitled to a form letter listing generic factors. She is not entitled to see what the system saw, to correct what it got wrong, or to understand what, exactly, tipped the judgment against her.

Nothing about this scene is exotic. Versions of it happen millions of times a day, in lending, hiring, insurance, immigration, policing, and welfare. And it contains, in miniature, the entire problem this final chapter exists to address: a moment of total asymmetry, in which one party is perfectly legible and the other is perfectly opaque, and the legible one is the human being.

We are frequently told that the crisis beneath all of this is a crisis of trust. People no longer trust institutions. Institutions no longer trust people. Neighbors do not trust neighbors, voters do not trust elections, patients do not trust doctors, and everyone, sensibly, distrusts what they read. The proposed remedies follow from the diagnosis: we must rebuild trust, restore faith, come together, heal.

The diagnosis is almost right, and the remedies are almost useless. This book has spent nine chapters arguing that our predicament is structural rather than moral, that it emerges from incentives rather than character, and that appeals to virtue cannot fix what architecture has broken. It would be a strange betrayal to end by recommending virtue. If the argument of this book is correct, then "people need to trust each other again" is not a prescription. It is a description of what a good prescription would produce.

So the question becomes: what would?

Begin by clearing away a comforting myth. There was no golden age in which human beings ran society on trust and technology ruined it. Pre-modern trust was real but parochial: intense within the village, and ending at its edge. Strangers were threats. Trade beyond the horizon required hostages, oaths, and kin. What actually built the modern world was the opposite of trust. It was verification: contracts enforced by courts, currencies backed by institutions, credit histories, audits, inspections, warranties, escrow. Each of these is a machine for making trust unnecessary, and together they accomplished something extraordinary. They allowed strangers to cooperate at planetary scale. You can send money to a person you will never meet, on a continent you will never visit, because a lattice of verification stands where trust would otherwise have to.

Verification, in other words, is not the enemy. It is the foundation. Any prescription that amounts to "verify less, trust more" is a prescription for going backward, and a serious reader should refuse it.

The problem is not that verification exists. The problem is the direction it flows.

Here is the principle that democratic societies discovered, slowly and at great cost, and that we are now abandoning without ever having voted to: power is supposed to be transparent to the people, and the people are supposed to be opaque to power. The state opens its books, publishes its laws, debates in public, stands for election, and submits to courts. The citizen, meanwhile, is entitled to a private life, a secret ballot, an unsearched home, an unmonitored conscience. Verification was aimed upward, at the strong, precisely because the strong are dangerous. Opacity was granted downward, to the weak, precisely because the weak need shelter. The arrangement was never perfect, but the vector was correct.

Over the past two decades, quietly and without a constitutional moment, we have inverted it.

The citizen is now the most legible object in the system. Every purchase, movement, search, association, hesitation, and habit is collected, correlated, and scored. Meanwhile the systems doing the watching have become progressively less visible: models that are proprietary, contracts that are classified, data flows that cross agencies and companies without a map, decision logic that is a trade secret when it is not simply beyond explanation. Recall the portraits of Chapter Eight. A satellite network on which a war depends, governed by one man's judgment, disclosed to no one. A surveillance-and-media estate whose owner cheerfully predicts that citizens will behave because they are constantly recorded. A software firm whose platforms mediate how the state itself sees, under contracts the public cannot read, running logic the public cannot audit. In each case the pattern is identical: watching flows down; opacity accumulates at the top. The same inversion appears, often more completely, in the Chinese surveillance stack and in sovereign AI infrastructure whose principals are not answerable to broad publics. Palantir can see you. You cannot see Palantir. Hikvision’s cameras and SenseTime’s models can see entire city populations; the populations cannot meaningfully inspect the systems that watch them.

This is the inversion, and once you see it, the so-called trust crisis stops looking like a mystery of the modern soul and starts looking like plumbing. Trust did not mysteriously evaporate. We re-piped verification to flow in one direction only, and trust drained out the bottom. Because trust was never the absence of verification. Trust is what mutual, bounded verification makes rational.

You trust your bank, to the extent you do, not because you believe bankers are virtuous but because examiners inspect them, regulators can shut them, deposits are insured, and failure has consequences. Nobody calls that naivety. It is a wager made reasonable by structure.

Distrust, likewise, is not a cultural mood to be talked out of people. It is the rational response to asymmetry. When the other side of every relationship can see you completely and be seen not at all, suspicion is not a pathology. It is arithmetic.

Which means the prescription is not to trust more, and not to verify less. It is to make verification symmetrical in proportion to power. Whoever holds the most capability owes the most legibility.

The watched must be able to watch back, not each other, but upward, at the systems and institutions that act upon them. The demand of this book is not gentler surveillance or friendlier terms of service. It is that the arrow be turned around.

And there is a second half to the principle, without which the first half breaks. Verification belongs where power lives. It must be kept out of the places where trust actually forms.

Because the inversion did not merely aim the cameras the wrong way. It let measurement seep into domains that measurement destroys. A friendship that is monitored is no longer quite a friendship. A conversation that is recorded becomes, subtly and then not subtly, a performance.

Deliberation under observation turns into positioning; candor becomes liability management; dissent becomes a risk calculation. The Oracle founder's prophecy, that citizens will be on their best behavior because everything is recorded, is offered as a promise of order, and it would deliver order of a kind. But best behavior is not trust. It is compliance, which is what remains when trust is no longer required and no longer possible. A society of perfect mutual verification would not be a high-trust society. It would be a society that had made trust obsolete, and human beings do not flourish as compliance engines. The felt experience of the past two decades, the sense many people have of drifting mentally apart from one another even as the tools of connection multiplied, is at least partly this: we have been teaching each other, through infrastructure, that observation is the normal condition and that unguarded presence is a risk.

So the rule has two clauses, and they are the same rule seen from opposite ends. Watch the powerful. Shelter the powerless. Aim verification at systems, institutions, and concentrations of capability, relentlessly, and build hard walls around the interior spaces, the private lives, associations, and conversations, where trust between people is actually manufactured. We currently do the reverse on both counts. That reversal is the disease. Its correction is the cure.

A principle without machinery is a sentiment, and this book has been hard on sentiment. So here are three mechanisms, offered not as a complete program but as an existence proof, evidence that structural answers are available to those willing to build them.

The first is audit with teeth. Not transparency reports written by the audited, and not ethics boards appointed by the observed, but independent examination on the model we already apply to banks: standing bodies with statutory authority to enter, inspect, and compel. Any algorithmic system deployed against the public at scale, in lending, hiring, policing, benefits, sentencing, or targeting, should be subject to examiners who can see the training data, query the model, read the logs of who searched for what, and publish findings the operator cannot veto. The precedent is a century old. We decided long ago that institutions holding other people's money must open themselves to inspectors precisely because the temptation and the stakes are too great for self-report. Institutions holding other people's data, and making decisions about other people's lives, have crossed the same threshold. The objection will be trade secrecy, and the answer is the same one we gave the banks: examiners can keep secrets. What they cannot do is allow the secret to include how you were judged. This is verification pointed upward, and nothing about it requires anyone to trust anyone. That is its virtue.

The second is expiration. Chapter Five described the true engine of the surveillance state, which is not decree but accretion: each sensor cheap, each expansion incremental, each new baseline normalized before anyone thought to object, until the infrastructure exists and the question of consent is moot. The structural answer is to make surveillance infrastructure the one category of public machinery that is not permanent by default. Sunset clauses, real ones: every camera network, data-sharing agreement, retention regime, and monitoring program lapses automatically after a fixed term unless publicly re-justified and re-authorized, with its actual usage records, the queries run, the errors made, the mission creep accumulated, on the table at renewal. This changes the physics of habituation. Today, watching persists unless someone summons the extraordinary energy to dismantle it. Under expiration, watching ends unless someone can publicly defend it. The burden of proof returns to where it belongs, on the watcher, at regular intervals, forever.

The third is a place where power must sit with the governed. Audits discipline systems and sunsets discipline infrastructure, but the deeper wound described in this chapter, the mental moving-apart, is not healed by paperwork. Trust between strangers is rebuilt the only way it has ever been built: through repeated, consequential contact under fair rules. This is what citizen assemblies are for, and they are not a utopian invention; they are running. Ireland handed its most intractable moral questions to randomly selected panels of ordinary citizens, given time, evidence, and the standing to be taken seriously, and those assemblies produced deliberation of a quality that professional politics had failed to produce for decades, on questions politicians had been too frightened to touch. Apply the instrument here. Before an algorithmic system is deployed on a population, a jury drawn from that population examines it, questions its makers under oath, and issues findings the deploying institution must publicly answer. The point is not that random citizens will out-analyze the engineers. The point is the direction of the encounter.

For one procedural moment, the builders of the machine must explain themselves to the people the machine will act upon, rather than the reverse. Every mechanism in this chapter is finally that same rotation, performed at a different altitude: the arrow of accountability, turned to point up.

None of these three is sufficient. All of them are possible. That is the claim that matters, because the counsel of despair, the shrug that says the technology is here and the asymmetry is inevitable, is itself one of the forces this book has been describing, and it is the cheapest of them. Asymmetry is not a law of nature. It is a set of contracts, defaults, and statutes, every one of which was written by someone and can be rewritten.

Return, finally, to the woman and her loan, because the test of any prescription is what it changes in the room where the harm occurs.

In the world these mechanisms describe, the system that judged her has examiners, and the examiners have seen its insides, and somewhere there exists a finding, written by people who do not work for the lender, about what this model does and to whom. The data that fed it carries an expiration date and a record of every hand it passed through. Before it was ever switched on, its makers sat across a table from a room of people like her and answered for it. And she can ask why, and the question has an address.

Notice what has not happened in this world. No one has become kinder. No executive has grown a conscience; no citizen has taken a workshop on faith in institutions. The people are the same people. Only the direction of the arrows has changed. And yet in that world, when she is asked whether the system can be trusted, she can answer yes without being a fool, because her trust is no longer a donation to the powerful. It is a wager with collateral.

That is the whole prescription, and it was never really about machines. We do not need people to trust each other more. We need to build a world in which trusting is no longer irrational, and then let human beings do what they have always done when the structure permits it. Trust is not the premise of a decent society. It is the yield.