When the Police Are the Public, AI Governance Becomes Everyone’s Job

Jul 19, 2026

11 min read

0 views

84%

The old boundary is disappearing

Here is a troubling question: what happens when the line between those who govern and those who are governed starts to blur? In one domain, the answer has long been familiar. The public is not just watched by the police, the public is, in a sense, the police. In another domain, the same logic is now arriving through technology policy: every agency is being pushed to appoint a chief AI officer, because AI is no longer a niche tool for specialists but a general capacity that can shape how the state sees, decides, and acts.

That conjunction matters more than it first appears. We usually think of policing as enforcement and AI as innovation, one reactive and one futuristic. But both are really about distributed power. Both ask the same deeper question: who gets to perceive reality on behalf of everyone else, and what happens when that power is spread across many hands instead of concentrated in one office?

The most important insight here is not that AI will help government work faster. It is that AI is becoming part of the machinery by which modern institutions define the public itself. And once that happens, governance stops being something done to people by experts behind closed doors. It becomes a shared civic architecture, for better or worse.

The public as infrastructure, not audience

The phrase “the public are the police” sounds provocative because it overturns the usual image of authority. But it also reveals something broader about how modern systems actually work. Institutions do not operate only through top down command. They operate through participation, surveillance, reporting, delegation, and feedback loops. A neighborhood watches, reports, and normalizes behavior. A bureaucratic system processes forms submitted by citizens, teachers, doctors, contractors, and agencies. The boundary between state and society is always thinner than we want to believe.

AI intensifies that reality. A chief AI officer is not just a new executive title. It signals that machine learning models, data pipelines, and automated decision systems are becoming part of the state’s nervous system. They can help a benefits office answer questions faster, a health agency identify outbreaks, a transportation department spot risk patterns, or a climate team model interventions. But they also change what it means for the public to encounter the state.

Consider a food assistance application. In the old model, a person fills out forms and waits for a human caseworker, whose judgment may be slow, inconsistent, or biased. In the AI mediated model, the person may encounter automated verification, eligibility scoring, document extraction, or fraud detection before any human sees the case. The state becomes more responsive, but also more abstract. The public is still the public, yet now it is legible to government through patterns, probabilities, and predictions.

That is the crucial shift: the public is no longer only the audience of institutions, it is the data substrate those institutions read from.

When institutions begin to see society through models, the first political question is not whether the model is accurate. It is whose reality gets counted as legible in the first place.

This is why the analogy to policing matters. Policing is never only about badges and patrol cars. It is about classification, attention, and interpretation. Who looks suspicious, who gets stopped, whose complaint matters, whose story is believed. AI systems inherit exactly this kind of power, but at scale, and often invisibly.

The promise of AI in government is real, and so is the trap

It would be a mistake to dismiss the pro public case for AI. Government is full of brittle processes, inaccessible forms, fragmented data, and slow coordination. A well designed AI system can reduce friction in ways that actually matter to people. It can translate services into multiple languages, help people navigate benefits, detect duplicative paperwork, surface patterns in public health data, and make institutions more usable for citizens who have historically been excluded by complexity.

Imagine three concrete examples.

First, a parent applying for childcare support. Instead of deciphering a maze of state websites, an AI assistant could walk them through the process, flag missing documents, and explain deadlines in plain language. That is not flashy innovation. It is dignity.

Second, a county health department monitoring a flu surge. AI could help synthesize hospital signals, pharmacy sales, school absences, and weather patterns to identify a local outbreak earlier. That is not abstract efficiency. It is preventive public health.

Third, a housing agency trying to detect discrimination in lending or rental patterns. Properly constrained AI could help auditors find anomalies that human reviewers would miss. That is not replacing democratic oversight. It is extending it.

The temptation is to stop there and declare that the only real challenge is deployment speed. But that would miss the deeper issue. Every one of these examples depends on a hidden social bargain: citizens must surrender more of their behavioral traces, institutions must interpret those traces correctly, and people must trust that the system is helping rather than sorting them.

This is where the trap emerges. The same tools that reduce friction can also reduce friction against power. A system that makes government easier to use can also make it easier for government to profile, exclude, nudge, or automate away human judgment. Efficiency is not a neutral virtue. In a public institution, efficiency always asks: efficient for whom, and at what cost to accountability?

That is why AI governance cannot be treated as a technical compliance exercise. It is a constitutional problem in practical clothing.

The real risk is not machine error, but moral outsourcing

People often worry that AI in government will make mistakes. That is a valid concern, but it is not the deepest one. The deeper risk is that institutions will use AI to outsource difficult judgments while preserving the appearance of objectivity.

A model can score risk, rank priorities, detect anomalies, or generate recommendations. What it cannot do is absorb responsibility. Yet organizations often behave as if it can. When the system denies a benefit, flags a person, or prioritizes one neighborhood over another, the machine becomes a shield. “The model said so” can sound like a neutral explanation, even when it is just a convenient way to hide contested value judgments.

This resembles the logic of distributed policing. In a neighborhood where everyone is encouraged to watch, report, and suspect, no single person feels fully responsible for the climate of fear. Authority gets diffused into routine behavior. AI can do something similar to governance: it diffuses judgment into workflows, dashboards, and scores until no one feels like the author of the decision.

That is why the appointment of a chief AI officer is symbolically important. If done well, it creates a visible locus of responsibility for systems that otherwise disappear into procurement, vendor claims, and technical jargon. If done poorly, it becomes a ceremonial title that gives institutions cover to expand automation without a real ethical center.

A useful mental model here is the difference between using AI as a microscope and using AI as a verdict machine.

A microscope helps humans see better. It does not replace judgment.
A verdict machine pretends that better seeing is the same as deciding.

Government should use AI like a microscope: to surface patterns, reveal bottlenecks, and illuminate hidden inequities. It should not use AI like a verdict machine that converts uncertainty into bureaucratic finality.

The public should never have to feel that they are arguing with an invisible statistical authority. They should be able to ask: who designed this system, what values were encoded in it, what data shaped it, and how can it be challenged?

A new civic model: shared intelligence with accountable judgment

The surprising connection between public policing and AI governance is this: both force us to decide whether modern institutions are trying to control society or become more intelligible to it. The first path breeds suspicion. The second path, if built with care, can deepen legitimacy.

The right goal is not “more AI” and not even “smarter government” in the abstract. The goal is shared intelligence with accountable judgment. That means building institutions where technology expands what the public can see and do, while humans remain clearly answerable for the choices that shape lives.

Here is what that looks like in practice.

1. Make systems explainable to ordinary people

If a government AI tool affects eligibility, access, or enforcement, the explanation must be understandable without a technical degree. Not a vague notice about algorithms, but a plain language account of what data was used, what the system can and cannot do, and how a person can contest the result.

2. Separate assistance from adjudication

AI can help prepare, triage, summarize, and identify patterns. It should not silently become the final judge in cases where rights, benefits, or penalties are at stake. Assistance is useful. Adjudication requires human responsibility.

3. Build public counter power

If agencies use AI to detect fraud, communities should have tools to detect bias. If governments use models to allocate resources, independent auditors should be able to test whether the system leaves some neighborhoods behind. Shared intelligence only works when oversight is also shared.

4. Treat data collection as a democratic decision

Every AI system depends on what it can see. That means the most important policy choice may be what data the state should never collect, not just what it can analyze. Restraint is not anti innovation. It is the precondition for trust.

5. Ask whether the system makes the public more legible or more vulnerable

This is the best test of all. If a tool helps people navigate government, it increases legibility in a positive sense. If it turns citizens into risk profiles that they cannot inspect or challenge, it increases vulnerability. Legibility and vulnerability can look similar from the inside. Governance must learn to tell them apart.

A democratic AI system is not one that knows everything. It is one that knows enough to help, but not so much that it can dominate.

This reframes the role of the chief AI officer. The title should not mean “the person who gets AI into every corner of government.” It should mean “the person accountable for ensuring that machine intelligence strengthens public life without eroding public agency.” That is a very different mission.

The future of trust will be built on visible limits

The hardest part of AI governance is not capability. It is trust. And trust in government has never come from promises of perfection. It comes from visible limits, transparent procedures, and the ability to challenge power when it gets things wrong.

That is where the two ideas finally converge. If the public are the police, then civic order depends on distributed participation. If AI becomes part of government, then civic order depends on distributed scrutiny. In both cases, the institution only works when people can see themselves inside it without feeling trapped by it.

The temptation in moments of technological change is to imagine that the answer is better tools. But better tools without better norms simply scale old habits. A bureaucracy that already mistrusts people will use AI to mistrust them faster. A bureaucracy committed to service will use AI to make service more humane, more accessible, and more accountable.

So the real challenge is not whether government should adopt AI. It already will. The challenge is whether citizens will demand that AI be used to widen participation rather than deepen opacity. That is a political choice, not a technical inevitability.

If the public is part of the machinery of governance, then public oversight must be part of the machinery of AI. Otherwise, the state will become easier to operate and harder to trust. And that is a bargain no democratic society can afford.

Key Takeaways

Treat AI in government as a legitimacy issue, not just an efficiency project. Ask whether it makes services more accessible and decisions more accountable, not only faster.
Use AI as a microscope, not a verdict machine. Let it surface patterns and reduce friction, but keep final judgments visibly human and contestable.
Demand plain language explanations for any AI system that affects rights or access. If citizens cannot understand how a decision was made, they cannot meaningfully challenge it.
Insist on independent audits and public counter power. If a system allocates benefits or flags risk, there should be external checks for bias, error, and hidden exclusion.
Set limits on data collection before problems appear. The most democratic AI policy is often knowing what not to know.

Conclusion

The most radical thing about AI in government is not that it can make the state smarter. It is that it forces us to redefine what a public institution is for. A healthy democracy does not merely deploy power efficiently. It makes power legible, contestable, and shared.

That is the hidden link between the public as police and the state with a chief AI officer. Both are reminders that modern governance is never simply top down. It is a web of perceptions, judgments, and responsibilities spread across people and systems. The question is whether that web becomes a net that catches people, or a fabric that holds the public together.

The future will not be decided by whether AI enters government. It already has. It will be decided by whether we build institutions that use intelligence to enlarge citizenship, or to make citizenship easier to process from a distance.