AI in Public Administration: Efficiency and Safeguards

This page is a translation of the original Portuguese text, provided for convenience. It deals with Brazilian law. If there is any discrepancy, the Portuguese version prevails. Read the original in Portuguese

Artificial intelligence is already transforming Brazilian public administration, but adopting it requires a balance between operational efficiency and the protection of citizens’ fundamental rights.

The Advance of Artificial Intelligence in Brazil’s Public Sector

In recent years, we have seen a growing use of artificial intelligence tools in Brazilian government agencies and entities. Courts use automated systems to sort cases, city governments deploy chatbots to serve residents, and federal agencies use algorithms to review administrative requests. This digital transformation is not a passing trend. It is a structural reorganization of how the State deals with the public and carries out its functions.

The main motivation for this adoption is the promise of significant efficiency gains. Repetitive, high-volume tasks, such as the initial review of documents, the ranking of requests by urgency, and the verification of formal requirements in administrative proceedings, can be performed by automated systems in a fraction of the time a human civil servant would need. This frees up staff for work that calls for qualitative analysis, empathy, and judgment based on context, functions that machines still cannot perform satisfactorily.

Still, deploying these technologies in the public sector brings challenges that do not exist in the private sector, or are less intense there. Public administration is bound by constitutional principles such as legality, impersonality, morality, publicity, and efficiency. Any technological tool adopted by the government must operate within those limits, and that creates specific demands for transparency, auditability, and accountability that go far beyond what is normally required of a private company.

Practical Applications and Use Cases in Public Management

When we look at concrete applications of artificial intelligence in public administration, we find a broad range of possibilities already taking shape at different levels of government. In the Judiciary, AI systems help organize case files, identify repetitive claims, and suggest draft decisions for less complex cases. These tools do not replace the judge, but they reduce the time spent on bureaucratic tasks and let human attention focus on the cases that truly require in-depth analysis.

In the Executive Branch, artificial intelligence is already used to detect fraud in social programs, optimize public transportation routes, monitor the environment through satellite image analysis, and forecast demand for health services. Oversight bodies, such as courts of accounts and comptroller offices, use algorithms to cross-reference data and identify irregularities in bidding processes and administrative contracts, expanding their oversight capacity without a proportional increase in staff.

Automation of Administrative Proceedings

One of the most promising areas is the automation of high-volume administrative proceedings. Consider the number of requests that agencies such as the INSS (Brazil’s National Social Security Institute) receive every day. The preliminary review of documents, the checking of deadlines, and the verification of formal requirements are steps that consume a great deal of civil servants’ time. AI systems trained for these tasks can perform the initial screening, flagging cases with incomplete documentation, identifying legal priorities (such as benefits for older adults or people with disabilities), and organizing service queues more rationally.

However, we must stress that automating these early steps does not mean (and should not mean) eliminating human decision-making at critical moments. Granting or denying a benefit, imposing an administrative sanction, or making any decision that directly affects individual rights must remain the responsibility of an identifiable public official who can justify the decision and answer for it.

Citizen Service and Accessibility

Another important area is digital service for the public. Virtual assistants on government websites can provide information about services, explain what documents are needed, track the status of requests, and route inquiries to the appropriate departments. When well implemented, these systems expand the public’s access to government information, especially for people who live far from urban centers or have difficulty traveling.

The government’s use of artificial intelligence is justified only when it expands access to rights, not when it creates new barriers between citizens and the State.

But there is an important risk here: the digital divide. A significant share of Brazil’s population still lacks adequate internet access or familiarity with digital tools. AI-mediated digitization of public services cannot lead to the elimination of in-person or telephone service channels. Doing so would create a barrier to access for the most vulnerable populations, who are often the ones who depend most on public services.

The Legal Framework and the Principles Governing Public-Sector AI

Brazil’s legal system contains provisions that, although not written specifically for artificial intelligence, offer important guideposts for its use in public administration. Article 37 of the Federal Constitution sets out the principles that govern administrative action: legality, impersonality, morality, publicity, and efficiency. Each of these principles places specific limits and requirements on the State’s use of automated systems.

The principle of legality requires that every act of the State have a basis in law. This means that government bodies need legal authorization to adopt AI systems, especially when those systems take part in decision-making that affects individual rights. The principle of impersonality prohibits discriminatory treatment, which obliges the government to check that the algorithms it uses do not reproduce or amplify biases present in their training data.

Brazil’s General Data Protection Law (Law No. 13.709/2018) also applies directly to this subject. Article 20 of the LGPD guarantees data subjects the right to request review of decisions made solely on the basis of automated processing of personal data. This provision is especially relevant when public administration uses algorithms to make or inform decisions affecting citizens, such as granting benefits, classifying risk, or prioritizing services.

Algorithmic Transparency and the Right to an Explanation

One of the most important debates in the regulation of public-sector AI concerns algorithmic transparency. Unlike the private sector, where trade secrets may justify a degree of opacity, public administration operates under the principle of publicity. This means citizens have the right to understand how the systems that process their information and influence decisions about them actually work.

In practice, algorithmic transparency operates on several levels. At the most basic level, the government body must tell citizens that an AI system is being used in a given proceeding. At an intermediate level, it must be able to explain, in accessible language, the system’s general logic and the criteria it uses. At the most advanced level, there must be audit mechanisms to verify that the system works as expected and does not produce discriminatory or arbitrary results.

This right to an explanation becomes even more important when we consider administrative decisions that go against the citizen. If a request is denied on the basis of automated analysis, the person concerned must have access to an understandable statement of reasons that lets them exercise their right to appeal effectively. An opaque decision, based only on a “system analysis,” violates the principle of due process of law and undermines the right of defense.

Risks, Biases, and Control Mechanisms

Using artificial intelligence in public administration is not free of risk, and acknowledging those risks is a necessary condition for responsible implementation. The most frequently discussed risk is algorithmic bias: the tendency of AI systems to reproduce, and at times amplify, discriminatory patterns found in the historical data on which they were trained. If a system is fed data from administrative proceedings that historically treated certain groups unequally, the algorithm may learn and perpetuate that inequality.

This risk is especially serious in the public sector because State decisions are binding and affect fundamental rights. Bias in an entertainment platform’s recommendation system causes inconvenience. Bias in a system that helps grant social security benefits or prioritize health care appointments can mean the denial of essential rights to populations that are already vulnerable.

To mitigate these risks, it is necessary to establish robust control and oversight mechanisms. Periodic audits of the AI systems used by public administration, carried out by independent technical teams, are essential to identify and correct biases before they cause harm at scale. Creating AI ethics committees in the government bodies that adopt these technologies is also an important measure, as long as those committees are diverse in composition and genuinely independent.

Responsibility and the Question of Accountability

Another critical issue is responsibility for errors or harm caused by automated decisions. In the traditional model of public administration, there is always an identifiable public official who signs the decision and answers for it. When an AI system takes part in the decision-making process, the chain of responsibility can become blurred. Who is responsible when an algorithm gets it wrong: the manager who authorized its deployment, the civil servant who operated the system, or the company that developed the software?

We believe that the State’s liability for harm caused by AI systems should follow the logic of strict liability provided for in Article 37, paragraph 6, of the Federal Constitution. A citizen harmed by a mistaken automated decision cannot be required to prove fault on the part of the public official or a flaw in the algorithm. It is up to the State to show that it acted properly. This underscores the importance of government bodies keeping detailed records of automated decisions, including the input data, the criteria applied, and the results produced, so that the decision-making process can be reconstructed when necessary.

Prospects for Responsible Governance of Public-Sector AI

Given this picture, it is clear that the central question is not whether public administration should adopt artificial intelligence, but how to do so in a way that respects fundamental rights and the principles governing State action. Building responsible governance of public-sector AI rests on several key axes that we must consider together.

The first axis is regulatory: the need for specific regulatory frameworks that set minimum requirements for the procurement, deployment, and operation of AI systems by the government. These frameworks should include impact assessments before implementation, transparency obligations, human oversight mechanisms, and periodic audit procedures. Regulation must be flexible enough to keep pace with technological change, yet firm enough to protect fundamental rights.

The third axis is educational: training civil servants to understand, operate, and supervise artificial intelligence systems. The goal is not to turn every civil servant into a data scientist, but to ensure that the professionals who interact with these systems understand them well enough to spot anomalies, question results, and exercise the human oversight that the law requires.

Finally, the fourth axis is participatory: including civil society in decisions about the government’s adoption of AI. Public consultations, hearings, and mechanisms for public participation are not mere democratic formalities. They are tools that enrich the debate and help identify risks and impacts that a purely technical or governmental perspective may miss. Artificial intelligence in public administration will be only as good as the governance that supports it, and that governance will be legitimate only if it is built with the genuine participation of the people who will be directly affected by these decisions.

This subject is directly related to INSS chatbots, a topic we cover in a dedicated article.

This subject is directly related to AI in public surveillance, a topic we cover in a dedicated article.

As informações deste artigo são de caráter informativo e não substituem consulta jurídica individualizada. Cada caso possui particularidades que exigem análise específica.