The bridge applied to the public sector

In the State, what is at stake stops being a profit-and-loss statement. Behind every procedure there is a citizen waiting, and behind every automated decision there is someone whose health, whose school or whose right depends on the system having been well thought out. That is why the bridge, taken to the public realm, becomes at once more difficult and more necessary. This chapter sets out how it operates in civic modernization without losing sight of the person on the other side of the counter.

Why the State is a case apart

The difference from the SME is not one of degree, it is one of nature. In the SME, poorly applied AI costs money, and money is recovered. In the State, an opaque algorithmic decision about health, education, security or the allocation of resources costs rights, and rights are not recovered with the following quarter’s balance sheet. Moreover, the citizen did not choose that system and cannot go to another vendor: they are captive to the quality with which the State designed what decides about them. That asymmetry is what raises the bar —what in the SME is good practice, in the State is a requirement: auditability, fairness, transparency, accountability—.

The risk, moreover, is not theoretical, and research named it before generative AI existed: opaque models that decide at scale about people’s lives tend to hide their biases precisely where they would most need to be audited, and to produce systematic harm under an appearance of mathematical objectivity.¹ The State is the terrain where that risk is greatest, because it is where an automated rule can be repeated over millions without anyone looking at it. The warning the method brings from its foundations —that in the agentic layer a process no one understood is not executed more slowly, it is executed on its own, at scale and without a witness— becomes, in the public realm, its gravest version: an agent that prioritizes case files, allocates a subsidy or routes a case according to a criterion that no one stopped to audit. → The bridge thesis.

The State concentrates, on the other hand, the three classic ills that AI can ease or aggravate depending on how it is applied: the bureaucracy and waiting times that wear down the civic relationship; the data silos between agencies that prevent comprehensive policies; and the distrust that grows every time an algorithm decides in a way no one can explain. The three are, at bottom, the same sociotechnical problem: systems thought out without reading the people —citizens and officials— they were meant to serve.

What the method refuses to propose

As in the SME, the bridge in the State is defined as much by what it refuses as by what it offers, and each refusal disarms a habitual way of failing in the public realm. It does not propose to “install systems”: it proposes to redesign state action from the citizen’s standpoint, because a tidy system mounted on a procedure that is unnecessary only digitizes the waste. It does not propose to replace the public employee: it proposes to free up their capacity to manage and serve, with AI as an extension of their judgment, not a substitute. It does not propose to buy software: it proposes to strengthen the institutional capabilities that allow any software to be used well, because without them the best tool is left with no one to sustain it. And it does not propose black boxes: in the State, a model that cannot be audited cannot be used, and this is not a design preference but the line that separates a legitimate policy from an automated arbitrariness.

The method, in public terms

The four links, brought down to the State, keep their logic and change in demand: each stretch now carries the weight of the rights it touches.

Three principles in public terms

That practice is ordered into three principles that, in the public realm, stop being good intentions and become conditions of legitimacy. The first puts the citizen at the center: the procedure is redesigned from the experience of whoever suffers it, not from the org chart of whoever administers it. The second treats AI as an extension of the official: public judgment is not replaced, it is amplified with evidence, so that the employee decides better and faster, not less. It is the direct translation of the research in human-centered AI —high automation and high human control are not a trade-off, but the quadrant that truly works—. → Authors and currents.

The third makes auditability a condition, not an ornament: in decisions about rights, what cannot be audited cannot be used. And this, far from braking modernization, is its engine. Governance —traceability, human control, explainability— is what makes an official and a citizen trust a system enough to use it. A system that cannot be explained is not adopted: it is tolerated, and at the first doubt it is abandoned or worked around on the sly.

The governance framework

That demand for governance is neither rhetoric nor a local invention: it rests on the frameworks that today structure responsible AI in the world. The NIST AI Risk Management Framework orders the work into four functions —Govern, Map, Measure, Manage— and places Govern as cross-cutting by design, running through the other three.² The EU AI Act, already binding and with staggered application, classifies systems by risk level and requires traceability and human oversight in high-risk ones —which is, precisely, the category into which public decisions about people fall—.³ The bridge does not invent its demand for auditability: it takes it from the architecture of the global standards and brings it down to concrete measurement through IMIA, the instrument that measures whether an agency is ready for AI to add value before installing anything.

The State is, moreover, where the arrival of agents —software that no longer suggests but decides and executes on its own— makes this bar unavoidable. An agent that prioritizes a case file, routes a social case or assigns an appointment on its own does not allow an official to review case by case; human control is exercised over the design and auditing of its rules —what it resolves on its own, when it escalates to a person, what it leaves on record—. In decisions about rights, a rule that the agent applies on its own and that no one can audit or revoke is not administrative efficiency: it is delegation in the dark of public power. That is why, the more autonomous the AI, the higher the demand for governance, not the lower —which is, exactly, what the IMIA governance gate turns into a condition and not advice—.

The underlying thesis

There is an assertion that orders all of the above: the modernization of the State in Latin America is not bought, it is built. It is not a software package that is tendered and installed, but an institutional capability that is strengthened —processes that are understood, data that is governed, officials whose judgment is amplified instead of retired—, with the citizen at the center of every design decision.⁴ It is the bridge thesis without a single concession: a problem treated as technical —“let’s buy the system”— when it is, end to end, sociotechnical —“let’s redesign state action and whom it serves”—. The difference between the two readings, in the public realm, is measured in rights.

See also: The bridge applied to SMEs · IMIA, the maturity instrument · The bridge thesis · Authors and currents · Bibliography