1. Introduction

In recent years, the rapid development of artificial intelligence has given a somewhat new meaning to debates about the automation of administrative decision-making^[1]. Public administrations across Europe increasingly try to explore digital tools to process large volumes of cases, streamline administrative procedures, and improve efficiency. At the same time, the prospect of delegating administrative tasks to automated systems raises fundamental questions about legality, accountability, and the role of human judgment in public administration. Administrative discretion (understood primarily as the authority’s power to choose among legally permissible outcomes) represents one of the most frequently invoked limits to automation and use of artificial intelligence. It is commonly assumed that while routine and strictly rule-based tasks may be automated, decisions involving discretion must remain within the exclusive domain of human officials.

This assumption has recently been reflected in legislative developments in the Czech Republic, as in the course of discussions on the modernization of the Code of Administrative Procedure, a proposal was introduced to explicitly regulate the possibility of automated administrative acts, allowing certain procedural acts to be performed automatically without the participation of an official, while explicitly excluding the use of automation in situations where administrative discretion must be exercised^[2]. This model is, of course, not unique to the Czech Republic, as other jurisdictions have adopted similar approaches to the automation of administrative decision-making. Such is, for instance, the case of the German VwVfG, stipulating that administrative acts may only be issued automatically provided that neither discretion nor room for interpretation is present in the case (although VwVfG does not prevent the legislator from adopting special legislation (lex specialis) as well as subsequently enacted legislation (lex posterior) modifying this setup)^[3].

At first sight, such a limitation may appear intuitive. Administrative discretion is often perceived as a domain that necessarily requires human judgment, sensitivity to context, and the ability to weigh competing interests. From this perspective, excluding discretionary decisions from automated processing seems to be a natural safeguard against the risks associated with algorithmic decision-making^[4]. A closer examination, however, reveals a significant conceptual flaw underlying this approach: modern administrative law in continental European legal systems is structured in such a way that elements of discretion are present in a large number of administrative decisions, including those that are otherwise highly routine and standardized. Legislative provisions frequently authorize administrative authorities to impose sanctions, grant benefits, or determine obligations within a range of possible outcomes^[5]. Formally speaking, such provisions grant the administrative authority a wide range of possible choices. If the existence of such discretion were interpreted as an absolute barrier to automation, the practical result would be that even the most routine administrative decisions could not be processed automatically.

This observation exposes a deeper problem in the way administrative discretion is often conceptualized, especially in the context of automated decision-making. Much of the current debate implicitly assumes that discretion operates in a binary manner: either a decision involves discretion (in which case automation should be excluded), or it does not (in which case automation may be permissible). In the present article, I intend to challenge this assumption. I argue that administrative discretion is not a uniform or binary phenomenon, but rather a linear concept that manifests itself in different degrees and modalities. While some forms of discretion indeed involve complex value judgments that are difficult to formalize, other forms are already highly structured and constrained in administrative practice.

The gap between the theoretical conception of discretion and its practical application plays a central role in this discussion. This article proceeds from the hypothesis that in practice, administrative authorities rarely exercise discretion in a completely open-ended manner, which manifests itself in the fact that discretionary powers are often structured through internal guidelines, established administrative practice, and methodological frameworks that seek to ensure consistency and equality in decision-making. As a result, many discretionary decisions are in reality guided by predetermined criteria that significantly limit the range of acceptable outcomes. Recognizing this reality has important implications for the debate on automated decision-making in public administration. If certain forms of discretion are already exercised within a structured framework of criteria and established practice, the categorical exclusion of automated systems from all discretionary decisions becomes difficult to justify. Rather than treating discretion as an absolute barrier to automation, it may be more appropriate to examine the different forms that discretion can take and to consider whether some of them are compatible with carefully designed forms of automated decision-making.

The aim of this article is therefore to reconsider the relationship between administrative discretion and artificial intelligence, or automation of administrative decision-making in general. This objective can be translated into the research question of whether the presence of administrative discretion necessarily precludes the use of automated decision-making in public administration, or whether this incompatibility is based on an overly simplified binary understanding of discretion. I seek to demonstrate that the prevailing binary understanding of discretion obscures important distinctions within the concept itself and leads to an overly restrictive approach to automation in public administration. Thus, by examining the doctrinal foundations of administrative discretion, its practical application, and the different modalities in which it operates, this article aims to develop a framework for assessing the compatibility of artificial intelligence with certain forms of discretionary decision-making.

The analysis, based upon doctrinal methodology, proceeds in several steps. I first examine the concept of administrative discretion from a doctrinal perspective and explore the reasons why legal systems rely on discretionary powers in the first place. The research focus then turns to the practical operation of discretion in administrative decision-making, highlighting the extent to which discretionary powers are structured and constrained in practice. Building on these observations, the subsequent section develops a more nuanced understanding of discretion by distinguishing between different degrees and modalities of discretionary decision-making. Against this background, I finally examine the potential role of artificial intelligence in replicating certain forms of structured discretion and explore alternative models for integrating automated systems into administrative decision-making without abandoning the safeguards traditionally associated with discretionary power.

2. The concept of administrative discretion and its raison d’être

It would probably be fair to call administrative discretion one of the central concepts of modern administrative law. In the simplest terms, the concept of administrative discretion (also known as administrative consideration or discretionary power)^[6] refers to situations in which the law authorizes an administrative authority to choose among several legally permissible solutions when deciding a particular case^[7]. Rather than dictating a single predetermined outcome, legislation often leaves decision-makers with a range of permissible options within which they must exercise judgment. As a result, administrative authorities must interpret statutory provisions, evaluate factual circumstances, and select an appropriate course of action within legally defined boundaries^[8]. Hence, unlike strictly bound decision-making, where the legal norm prescribes a single outcome once the relevant facts are established, discretionary powers allow the administrative authority to determine which of several possible outcomes is most appropriate under the circumstances of the case^[9].

In doctrinal terms, administrative discretion typically arises when the legal norm is formulated in a way that does not fully predetermine the content of the administrative decision. This may occur, for example, where the law authorizes an authority to impose a sanction “up to” a specified amount^[10], to grant or refuse a permit depending on an assessment of certain public interests, or to select one of several possible regulatory measures. In such situations, the law establishes the legal framework within which the authority must operate, while leaving a margin of choice regarding the final decision. The exercise of discretion therefore presupposes that several outcomes are legally permissible and that the authority must determine which of them should be applied in the specific case.^[11]

At this point, it is necessary to clarify that the concept of administrative discretion is not understood uniformly across legal systems. For instance, in German administrative law, a distinction is traditionally drawn between “Rechtsfolgenermessen”, referring to the authority’s discretion in selecting among several legally permissible legal consequences, and “Beurteilungsspielräume”, which concern the administration’s margin of appreciation in assessing whether the legal requirements for administrative action are fulfilled. While the former constitutes the purer form of administrative discretion, the latter occupies a more contested position. German doctrine generally proceeds from the assumption that vague legal terms do not, as a rule, confer autonomous margins of appreciation upon the administration but remain fully subject to judicial interpretation, save for certain exceptional categories recognized by legislation and case law. By contrast, many other legal systems treat both the determination of legal consequences and the assessment of open-ended legal concepts as manifestations of administrative discretion. The present article is primarily concerned with the former category, namely discretionary powers relating to the choice among legally permissible outcomes. Nevertheless, the broader comparative perspective illustrates that the concept of discretion itself admits different doctrinal configurations, reinforcing rather than weakening the argument that discretion should not be regarded as a single, homogeneous phenomenon^[12].

The existence of discretionary powers reflects certain structural characteristics of our administrative governance. Legislators are rarely able to foresee all possible factual situations that may arise in practice. If legal norms attempted to casuistically regulate every individual case in advance, they would either become excessively rigid or require an unrealistic level of detail. Discretion therefore functions (at least in theory) as an instrument that allows administrative authorities to adapt the application of the law to the particular circumstances of individual cases. By leaving a certain margin of choice to administrative authorities, the legal system acknowledges the limits of legislative foresight and enables the administration to respond flexibly to diverse and evolving situations.

Another important function of administrative discretion lies in the pursuit of fairness and proportionality in administrative decision-making. Strictly rule-based systems may produce outcomes that are formally correct but substantively inappropriate in particular circumstances^[13]. Discretion allows the administrative authority to take into account relevant contextual factors and to balance competing interests in a manner that better reflects the underlying purpose of the law. In this sense, discretion serves as a mechanism that helps to mitigate the rigidity that might otherwise result from the application of general legal rules to highly diverse factual situations^[14].

At the same time, the exercise of discretion is not unlimited. Even where the law grants a margin of choice, the administrative authority remains bound by the legal order and must exercise its discretion within the limits established by law^[15]. Administrative law doctrine therefore traditionally distinguishes between the existence of discretion and its lawful exercise. While the authority may choose among several possible outcomes, this choice must respect the purpose of the legal norm, the general principles of administrative law, and the requirements of equality and proportionality. Discretionary powers must not be exercised arbitrarily or in a manner that is inconsistent with the objectives pursued by the legislature^[16]. These limitations are reflected in the doctrinal concept often described as the prohibition of misuse of discretion. Administrative authorities must not base their decisions on irrelevant considerations, must take into account all relevant circumstances of the case, and must not exceed the limits of the discretion granted by law. In addition, discretionary decisions are subject to judicial review, although courts typically refrain from substituting their own assessment for that of the administrative authority and instead examine whether the authority has remained within the legal boundaries of its discretionary power^[17].

Although the concept of administrative discretion is described in doctrine by the term “power”, which might suggest that the administrative authority is free to decide whether it will exercise discretion at all, such an interpretation would be misleading. The fact that the law grants a discretionary margin does not mean that the authority may arbitrarily choose whether or not to use it. If the legal norm requires that certain factors be assessed through discretionary evaluation, the administrative authority is not entitled, but rather obliged to carry out that evaluation and to base its decision on it. The Czech Supreme Administrative Court has observed that «the counterpart of the discretionary power of an administrative authority is the duty to exercise such discretion, that is, to consider all factors that the law establishes as premises of such reasoning, to obtain the necessary evidence for that purpose, to derive factual and legal findings from that evidence, and subsequently, through discretionary reasoning, to reach an appropriate decision»^[18].

In other words, discretion is not merely a privilege granted to administrative authorities but also a procedural and substantive obligation. If an authority fails to exercise discretion in situations where the law requires it, its decision would be unlawful.

From a doctrinal perspective, administrative discretion is therefore commonly understood as a legally structured margin of choice that enables administrative authorities to adapt the application of the law to the specific circumstances of individual cases. It reflects the recognition that legal regulation cannot fully anticipate the complexity of social reality and that administrative decision-making requires a certain degree of flexibility in order to achieve fair and proportionate outcomes. This traditional understanding of administrative discretion plays an important role in contemporary debates on the automation of administrative decision-making. Discretion is frequently perceived as the domain of human judgment and as a safeguard against overly mechanical forms of governance. For this reason, many legislative proposals and scholarly contributions assume that discretionary decisions should remain outside the scope of automated processing. Before accepting such conclusions, however, it is necessary to examine more closely how discretionary powers are actually exercised in administrative practice. As the following section will demonstrate, the practical operation of discretion often differs significantly from the theoretical model that is commonly described in administrative law doctrine.

3. Administrative discretion in practice

The doctrinal account outlined above portrays administrative discretion as a legally structured margin of choice that enables administrative authorities to adapt the application of legal norms to the specific circumstances of individual cases. While this theoretical model accurately captures the legal framework within which discretion operates, the evidence suggests that the way discretionary powers are exercised in administrative practice often differs from this idealized model. While “discretion in the books” is based upon the abstract ideal in which every case is individually decided upon while carefully considering all the evidence and all the specific facts and circumstances of each case, “discretion in action” appears to be much more mundane. In reality, discretionary decision-making rarely takes the form of an entirely open-ended evaluation in which the administrative authority freely selects one of many possible outcomes. Instead, it appears that the exercise of discretion is typically shaped and constrained by a variety of practical mechanisms that aim to ensure consistency, predictability, and equal treatment in the best-case scenario, or realistically to simply increase efficiency^[19].

One important reason for this development lies in the institutional nature of administrative decision-making. Administrative authorities do not operate in isolation, but rather as organizations responsible for deciding large numbers of similar cases. In such a context, purely individualized and unconstrained exercises of discretion would inevitably lead to significant variations in outcomes. If different officials were to decide comparable cases in substantially different ways, the resulting disparities would raise concerns regarding equality before the law and the legitimacy of administrative action. For this reason, administrative bodies often develop internal mechanisms that guide the exercise of discretion and help to align individual decisions with a broader institutional practice^[20].

The tendency to structure discretionary decision-making is not merely a practical response of administrative bodies but is also reflected in broader European administrative law standards. Guidance on the exercise of discretionary powers has long emphasized that discretion must be exercised within a framework of objective and consistent criteria. This approach is reflected, for example, in Recommendation No. R (80) 2 of the Committee of Ministers of the Council of Europe concerning the exercise of discretionary powers by administrative authorities. The Recommendation establishes a number of principles designed to prevent arbitrariness and ensure consistent decision-making. Among other requirements, administrative authorities exercising discretion must observe objectivity and impartiality, take into account only relevant factors, respect the principle of equality before the law, and apply any general administrative guidelines in a consistent manner while still considering the particular circumstances of each case. These principles illustrate that discretion has long been understood in European administrative law not as an unfettered freedom of decision but as a legally structured form of decision-making that operates within a set of normative constraints^[21].

These mechanisms take a variety of forms. In many areas of administrative law, discretionary decisions are guided by internal guidelines, methodological instructions, established administrative practice^[22], but also constrained by the principle of proportionality, which operates as a fundamental limit on the exercise of discretionary powers and simultaneously structures judicial review^[23]. Such instruments typically specify criteria that should be taken into account when exercising discretion, identify relevant aggravating or mitigating circumstances, and sometimes even indicate approximate ranges of appropriate outcomes in typical cases. Although these instruments are typically only internally binding and thus neither have the same external effect as statutory provisions nor limit the legal obligation of the administrative authority to differentiate between specific cases and individualize its decisions, they nevertheless play a significant role in shaping the day-to-day operation of administrative decision-making and have significant de facto power over officials who are in principle obligated to adhere to internal instruments^[24].

A clear illustration of this phenomenon can be found in the field of administrative punishment. Legal provisions frequently authorize administrative authorities to impose sanctions within a relatively broad range. A statutory provision may, for example, allow the imposition of a fine “up to” a certain maximum amount. From a purely formal perspective, such a provision should be considered the highest form of discretion, since it grants the authority a wide range of possible outcomes. In practice, however, administrative authorities rarely treat each possible amount within this range as an equally plausible option. Instead, sanctioning practices tend to develop around relatively stable patterns, often supported by internal guidelines or methodological frameworks that link the severity of the sanction to specific factors such as the seriousness of the offence, the circumstances of its commission, or the conduct of the offender^[25]. These forms of structured decision-making reflect an effort by administrative authorities to reconcile the flexibility inherent in discretionary powers with the need for consistency and predictability. By articulating common criteria for the exercise of discretion, administrative bodies seek to reduce the risk of arbitrary differences between similar cases and to provide a degree of transparency regarding the reasoning underlying their decisions. In this way, discretionary powers that appear broad at the level of statutory formulation may in practice be exercised within a significantly narrower and more structured framework^[26].

Another factor that contributes to the structuring of discretion is judicial review. Courts reviewing discretionary decisions typically refrain from substituting their own assessment for that of the administrative authority, but they nevertheless examine whether the authority has exercised its discretion in a lawful and rational manner. Requirements such as the prohibition of arbitrariness, the obligation to consider relevant circumstances, and the principle of proportionality encourage administrative authorities to articulate and apply coherent criteria when exercising discretionary powers. Over time, these expectations may further consolidate administrative practices that promote consistency across cases^[27].

Taken together, these developments reveal an important feature of modern administrative governance: while discretion remains formally embedded in many legal norms, its practical exercise is often mediated by a variety of institutional and methodological constraints. Discretionary decision-making therefore frequently operates within a structured environment that limits the range of acceptable outcomes and channels the reasoning of administrative authorities through predefined criteria and established practices. This observation plays a crucial role in the context of automated decision-making. If discretionary powers are already exercised through structured frameworks designed to ensure consistency, the question arises whether certain aspects of such structured discretion might, at least in principle, be replicated or supported by automated systems. The following section therefore turns to a closer examination of the different degrees and modalities of discretion that can be identified within administrative decision-making.

4. Modalities of administrative discretion

The discussion in the preceding sections reveals an important tension between the doctrinal understanding of administrative discretion and the way in which discretionary powers are actually exercised in administrative practice. While doctrine often describes discretion as a margin of choice between several legally permissible outcomes, the practical operation of discretion frequently takes place within structured frameworks that guide the decision-making process and limit the range of acceptable outcomes. This observation raises a broader conceptual question. If discretion is routinely exercised within such constraints, is it appropriate to treat it as a uniform and indivisible category, or should it instead be understood as a phenomenon that manifests itself in different degrees and modalities?

Traditional doctrinal discussions tend to approach discretion as a binary concept. Either a legal norm leaves the administrative authority with a margin of choice, in which case the decision is discretionary, or the norm determines the outcome with sufficient precision, in which case the decision is bound^[28]. This binary distinction undoubtedly serves an important analytical function: it allows legal doctrine to distinguish between situations in which the administrative authority merely applies the law and those in which it must also evaluate and select among different permissible outcomes. At the same time, however, this distinction may obscure significant differences between various forms of discretionary decision-making.

A review of different instances of discretion across legislation suggests that not all discretionary powers are alike. Some involve complex value judgments, the balancing of competing interests, and the evaluation of open-ended legal concepts. In such cases, the administrative authority must consider a wide range of contextual factors and determine which solution best reflects the purpose of the legal norm in the circumstances of the particular case. Decisions of this kind often involve a high degree of responsibility and normative evaluation, as the authority must effectively determine how the legal framework should be concretely applied to a specific situation^[29].

Other forms of discretion, however, operate in a markedly different manner. For instance, in many areas of administrative law, discretionary powers take the form of selecting a value or outcome within a numerically defined range. The law may authorize the administrative authority to impose a fine “up to” a certain amount, to determine the duration of a measure within specified limits, or to select one of several regulatory options that differ primarily in degree rather than in kind. Although such provisions formally leave a wide margin of choice, the practical decision-making process often resembles the application of a structured methodology rather than an open-ended evaluative judgment^[30].

A further form of discretion can be observed in situations where the law formally confers a discretionary power on the administrative authority, yet the practical outcome of the decision depends almost entirely on whether the applicant fulfils a set of predetermined criteria. This situation frequently arises in licensing or authorization procedures^[31]. In many jurisdictions and in certain categories of licensing procedures, legal provisions may state that the authority “may” grant a license if certain conditions are met, thereby formally framing the decision as discretionary. In practice, however, the decision-making process often revolves around the verification of whether the applicant satisfies the statutory requirements, such as professional qualifications, technical standards, or other objective conditions laid down by law. Once these criteria are fulfilled, the authority will normally grant the license, whereas their absence typically leads to refusal. Although such decisions may formally fall within the category of discretionary powers, the actual scope for evaluative judgment is often very limited. The task of the administrative authority primarily consists in verifying compliance with predefined conditions rather than weighing a broad range of policy considerations or selecting among qualitatively different outcomes. In this sense, the decision-making process resembles the application of a structured checklist more than the exercise of open-ended discretionary reasoning. The authority’s role is not so much to determine which of several possible outcomes would be the most appropriate, but rather to ascertain whether the legally prescribed criteria have been satisfied.

The difference between these forms of discretion becomes particularly visible when considering the structure of the choices available to the administrative authority. In some cases, the authority is effectively faced with a limited number of qualitatively distinct options. A decision whether to grant or refuse a particular status or authorization may involve only two possible outcomes, each carrying significant legal and social consequences. In other situations, however, the authority may formally be able to choose from a large number of possible outcomes that differ only incrementally. For example, where the law authorizes the imposition of a fine of up to a specified maximum amount, the administrative authority may theoretically select from a wide spectrum of possible amounts. Yet this multiplicity of potential outcomes does not necessarily translate into a correspondingly broad exercise of evaluative judgment.

Paradoxically, the existence of a large number of formally available options may actually encourage the development of structured decision-making frameworks that reduce the effective scope of discretion. Faced with the possibility of choosing among hundreds or thousands of potential outcomes, administrative authorities often seek to ensure consistency by establishing methodological criteria that link particular factual circumstances to typical ranges of sanctions or measures. In such cases, the practical exercise of discretion increasingly resembles the application of a structured decision-making model rather than the exercise of entirely open-ended judgment, thus resembling what might be described as the “paradox of choice” of a kind^[32].

These observations suggest that administrative discretion is better understood as a spectrum rather than as a binary category. At one end of this spectrum lie forms of discretion that require complex normative evaluation and the balancing of competing considerations. At the other end are forms of discretion that are primarily technical or quantitative in nature and that are frequently exercised through structured methodologies designed to ensure consistency and predictability. Between these poles lies a variety of intermediate situations in which discretion is present but significantly shaped by institutional practices, guidelines, and methodological frameworks^[33].

Recognizing these different modalities of discretion has important implications for the debate on automated decision-making in public administration. If discretion is treated as a single, undifferentiated concept, it may appear natural to exclude all discretionary decisions from automated processing. If, however, discretion is understood as a spectrum encompassing different forms of decision-making, the relationship between discretion and automation becomes more nuanced. Certain forms of discretion that involve complex normative judgments may indeed require the involvement of human decision-makers. Other forms, particularly those that are already exercised through structured criteria and methodological frameworks, may be more compatible with carefully designed forms of automated decision support or decision-making. The purpose of distinguishing between different degrees and modalities of discretion is therefore not to deny the importance of discretionary powers in administrative law, but rather to refine the conceptual framework within which discretion is understood. By recognizing that discretion operates in different ways across different types of decisions, it becomes possible to move beyond the simplistic assumption that the mere presence of discretion necessarily precludes the use of automated systems. This refined understanding provides the conceptual foundation for the analysis examining whether and to what extent artificial intelligence may be capable of replicating certain forms of structured discretion in administrative decision-making.

5. Artificial intelligence and the replication of structured discretion

Much of the existing scholarly debate on artificial intelligence and administrative decision-making is framed around the question of whether automated systems should be allowed to exercise discretionary powers^[34], or prohibited from doing so^[35]. This formulation of the problem, however, may itself be based on a questionable premise. From my perspective, the crucial issue is not simply whether artificial intelligence should or should not be allowed to use discretion, but rather what is meant by discretion in the first place. If discretion is understood as the process of deciding cases that require genuinely individualized assessment, complex contextual judgment, and the balancing of competing values, the argument that such decisions should remain within the domain of human officials is both intuitive and convincing. In such situations, the exercise of discretion indeed presupposes forms of reasoning and responsibility that are difficult to translate into purely technical processes and should be kept in the human domain. If, however, discretion refers to the application of structured and standardized decision-making frameworks that are designed to navigate the wide range of options formally provided by legislation, the picture appears considerably different.

Professor Juli Ponce Solé has famously remarked^[36] that artificial intelligence is “psychopathic”, in the sense that it lacks the capacity for empathy and moral sensitivity that may be required when exercising administrative discretion. From this perspective, automated systems appear inherently unsuitable for decisions that require individualized assessment and the weighing of competing human interests. Modern administrative systems are not designed to encourage purely subjective or emotionally driven decisions. On the contrary, administrative law traditionally seeks to limit the influence of personal preferences and individual sensibilities by requiring officials to base their decisions on legally relevant criteria, to apply general guidelines consistently, and to treat comparable cases alike^[37]. In this sense, administrative institutions deliberately discipline the decision-making process in order to reduce variability between individual officials, thus rendering administrative decision-making similarly detached from individual emotional judgment. The exercise of discretion is therefore often structured in a way that minimizes the role of purely personal judgment and channels decision-making through predefined methodological frameworks. When discretion operates within such structured systems, the decision-making process may already resemble the application of standardized rules rather than the expression of individual moral intuition.

In many areas of administrative practice, the exercise of discretion already involves the systematic application of predefined criteria, internal guidelines, and methodological frameworks that substantially limit the freedom of the individual decision-maker. In this sense, the institutional design of discretionary decision-making often seeks to reduce variability and align individual decisions with established patterns of administrative practice. Where discretion operates in this highly structured manner, it becomes difficult to maintain that such decision-making is intrinsically human in nature. Rather, it may increasingly resemble the application of complex but formalizable decision rules, which raises the possibility that at least some forms of such structured discretion could, in principle, be replicated by technical means.

A categorical exclusion of discretionary decisions from automated processing would have far-reaching consequences for the practical deployment of artificial intelligence in public administration. As noted above, elements of discretion are present in a large number of administrative decisions, including those that are otherwise highly routine and standardized. Legal provisions frequently authorize administrative authorities to determine outcomes within a certain range, to assess compliance with various criteria, or to select among several permissible measures. If the mere presence of such discretionary elements were interpreted as an absolute barrier to automation, the result would be that artificial intelligence could be used only in a very limited number of administrative tasks. Many areas of decision-making that are today considered among the most suitable candidates for automation would remain outside the scope of automated processing^[38].

This consequence appears difficult to reconcile with the practical realities of modern public administration. Administrative authorities are often required to process large volumes of relatively similar cases in a consistent and timely manner. In many such situations, the exercise of discretion already takes place within structured frameworks that guide the decision-making process. Internal guidelines, methodological instructions, and established administrative practices frequently provide detailed criteria that link specific factual circumstances to particular types or ranges of outcomes. As discussed in the previous sections, the practical exercise of discretion in these contexts often resembles the application of a structured methodology rather than the exercise of entirely open-ended evaluative judgment.

I believe that the debate on artificial intelligence and administrative discretion should not be framed as a choice between two extreme positions. On the one hand, it would be difficult to justify granting automated systems unrestricted authority to exercise discretionary powers in the same manner as human officials^[39]. Decisions involving complex value judgments, the balancing of competing interests, or the assessment of open-ended legal concepts raise significant concerns regarding accountability, transparency, and the protection of fundamental rights. On the other hand, a categorical exclusion of all discretionary decisions from automated processing ignores the structured and methodologically guided nature of many forms of discretionary decision-making in contemporary administrative practice. A more nuanced approach would therefore recognize that certain forms of discretion may be compatible with carefully designed forms of automated decision-making.

In particular, artificial intelligence may be capable of replicating forms of structured discretion that are already exercised through predefined criteria, methodological frameworks, or established administrative practice. In such situations, the role of the automated system would not be to invent new standards of decision-making but rather to apply existing criteria in a consistent and predictable manner. One possible model involves the replication of structured sanctioning practices^[40]. In areas such as administrative punishment, internal guidelines often link specific types of violations and aggravating or mitigating circumstances to particular ranges of sanctions. An automated system could, in principle, apply these criteria in a manner similar to that used by human officials, thereby ensuring consistency in the treatment of comparable cases. Rather than replacing the underlying legal framework, such a system would operate within the same methodological structure that already guides discretionary decision-making in practice^[41].

A second model concerns situations in which discretionary decisions are effectively determined by the fulfilment of predefined criteria. As discussed above, licensing procedures frequently involve the verification of whether the applicant satisfies a set of statutory conditions. Although the legal norm may formally grant the administrative authority a discretionary power to grant or refuse the license, the practical outcome of the decision often depends primarily on the objective assessment of whether these conditions have been met. In such cases, automated systems may assist or even perform the decision-making process by verifying compliance with the relevant criteria^[42].

Another possible approach involves the restructuring of administrative procedures in a manner that separates standardized and discretionary elements of decision-making. In such a model, the initial stage of the procedure could involve the automated application of objective criteria and structured methodologies^[43]. Human decision-makers would then retain the possibility of intervening in cases where the circumstances of the individual case justify a departure from the standard outcome. Discretion would thus function as a corrective mechanism rather than as the primary mode of decision-making. Instead of being exercised in every individual case, discretionary judgment would be reserved for situations in which the standardized decision does not adequately reflect the particular circumstances of the case or where exceptional factors call for a different outcome. In this sense, the role of the human official would shift from routinely determining outcomes in every case to supervising the functioning of the standardized system and intervening when necessary to ensure that the resulting decisions remain fair and proportionate.

Such a layered structure may offer a way to reconcile two objectives that are often perceived as being in tension. On the one hand, administrative systems require consistency, predictability, and the ability to process large volumes of cases efficiently. On the other hand, the legal order seeks to preserve the possibility of individualized consideration and to avoid overly rigid decision-making that fails to account for the particularities of individual cases^[44]. By combining automated standardization with the possibility of discretionary correction, administrative procedures could preserve the advantages of both approaches. Human discretion would be exercised primarily in atypical situations, where the circumstances of the case justify deviating from the standard approach. In this respect, the introduction of automated decision-making would not necessarily represent a radical transformation of administrative reasoning, but rather a technological extension of an existing institutional logic^[45].

This layered approach reflects an important insight into the function of discretion in administrative law. Discretion does not necessarily have to operate as an unrestricted margin of choice at every stage of the decision-making process. In some contexts, it may instead serve as a safeguard that allows administrative authorities to depart from standardized solutions when the particular circumstances of a case so require. By combining structured decision-making with the possibility of discretionary intervention, administrative systems may be able to reconcile the benefits of consistency and efficiency with the need to ensure fairness and individual consideration.

The proposed two-stage model naturally raises a further question: who should determine whether a particular case is sufficiently standardized to be decided automatically or whether it requires human assessment? This preliminary classification should not itself become an unconstrained algorithmic decision. As Timo Rademacher has convincingly observed^[46], contemporary AI systems operate primarily by identifying statistical patterns and may fail to recognize legally relevant circumstances that fall outside those patterns or have not been incorporated into the model’s informational framework. They may therefore lack the ability to identify their own informational blind spots or to appreciate that a particular case calls for an individualized assessment rather than the application of an established pattern. Rather than entrusting this gateway decision exclusively to the automated system, a hybrid allocation mechanism appears preferable. The automated system could initially process only those cases that satisfy predefined objective criteria indicating a sufficiently standardized factual and legal situation, while referring cases presenting predefined risk indicators, inconsistencies, missing information or low confidence levels to a human decision-maker. At the same time, the affected party should retain the procedural right (or even responsibility) to contest the automated classification and request human review where it considers that the case involves circumstances insufficiently reflected in the automated assessment.

Such an arrangement recognizes that the individual concerned will frequently be in the best position to identify factual or contextual features that distinguish the case from the ordinary pattern on which the automated decision relies. Instead of requiring the administration or the AI system to anticipate every possible exceptional circumstance, the procedural burden of identifying relevant atypical features could, to a limited extent, shift to the party to the proceedings. Where the party substantiates that the case involves legally relevant circumstances requiring individualized consideration, the matter would proceed to a human official exercising the full scope of administrative discretion. The automated decision would thus operate as the default outcome only for cases that remain uncontested and genuinely correspond to the standardized factual pattern for which the automated procedure was designed. From this perspective, the proposed model does not eliminate individualized administrative decision-making but reallocates responsibility for identifying the need for individualization. Routine cases may be resolved through automated procedures, whereas individualized assessment becomes available whenever either the administration or the affected party demonstrates that the predefined standardized framework is insufficient.

It must also be acknowledged that the integration of artificial intelligence into discretionary decision-making raises a number of important legal and institutional questions. Issues such as the transparency of algorithmic decision-making, the risk of reinforcing problematic administrative practices, the potential influence of automated outputs on human reviewers, and the broader implications for procedural fairness require careful examination^[47]. These questions extend beyond the scope of the present article and will therefore be addressed in a separate study focusing specifically on the risks and institutional challenges associated with the automation of administrative decision-making. The analysis presented in this section nevertheless demonstrates that the relationship between artificial intelligence and administrative discretion is more complex than is often assumed. Once discretion is understood as a spectrum encompassing different modalities of decision-making, it becomes possible to identify situations in which automated systems may operate within existing frameworks of structured discretion without undermining the fundamental principles that govern the exercise of public authority.

6. Conclusions

The debate on the use of artificial intelligence in public administration is often framed around the question of whether automated systems can or cannot be used in situations involving administrative discretion (and predominantly on the assumption of the latter). I argue that such a question per se rests on a problematic premise, and that the perceived incompatibility between discretion and automation is largely a consequence of treating discretion as a binary concept. Once a decision is classified as discretionary, automation is assumed to be excluded, and vice versa. The analysis presented in this article suggests that this premise might be conceptually flawed.

I believe that administrative discretion in the age of automation should not be perceived as a uniform phenomenon that operates in an all-or-nothing manner. As the preceding sections have shown, discretionary powers manifest themselves in different degrees and modalities. While some forms of discretion involve complex value judgments, individualized assessments, and the balancing of competing interests, others operate within highly structured frameworks defined by predetermined criteria, internal guidelines, and established administrative practice. In many areas of administrative decision-making, discretion already functions through methodological frameworks designed to ensure consistency and predictability across large numbers of similar cases.

This phenomenon even gives rise to a somewhat paradoxical situation in which the broader the formal margin of choice granted to administrative authorities by legislation, the stronger the institutional incentives to limit that freedom through standardization, internal guidelines, and consistent administrative practice. Faced with a wide spectrum of formally permissible outcomes, administrative bodies frequently seek to reduce variability by developing structured decision-making methodologies. As a result, some forms of discretion operate in practice as what may be described as pseudo-discretion: although the legal framework formally leaves several options open, the decision-making process is effectively guided by predetermined criteria that significantly narrow the range of outcomes that can realistically be chosen.

This observation bears important implications for the relationship between administrative discretion and automated decision-making. If discretion is understood as a spectrum rather than a binary category, the question of automation cannot be answered by simply asking whether discretion is present. The more relevant issue concerns the nature and structure of the discretionary decision-making involved in a particular case. Decisions requiring genuine individualized judgment and the balancing of complex human considerations may remain inherently human tasks. By contrast, forms of discretion that primarily involve the application of structured criteria or standardized methodologies may in principle be capable of replication through technical means. From this perspective, the introduction of artificial intelligence into administrative decision-making does not necessarily require abandoning the concept of discretion. Instead, it invites a reconsideration of the role that discretion plays within administrative procedures. In many areas of administrative practice, discretion currently functions as a permanent element of decision-making, meaning that officials are expected to exercise evaluative judgment in every individual case.

The analysis developed in this article suggests that an alternative institutional design may be possible. Rather than relying on discretion at every stage of the decision-making process, administrative procedures could combine standardized decision-making with the possibility of discretionary correction. In such a model, automated systems could apply predefined criteria and established administrative methodologies in routine cases, while human officials would retain the authority to depart from the standardized outcome where the particular circumstances of a case justify such intervention. Discretion would thus operate primarily as a corrective mechanism rather than as the default mode of decision-making. This approach could preserve the flexibility traditionally associated with discretionary power while simultaneously allowing administrative systems to benefit from the consistency and efficiency of structured or automated decision-making.

The central challenge in the age of artificial intelligence is therefore not whether discretion should exist in administrative law, but how different forms of discretion should be understood and organized within increasingly digital systems of governance. Moving beyond the simplistic assumption that discretion and automation are inherently incompatible may open the way for more nuanced approaches to the integration of artificial intelligence into public administration.

1. This article was prepared within the framework of the research project No. MA/25/F3/0/017 titled “Europe: a Laboratory of a Digital State” supported by the 4 EU+ European University Alliance, and builds on the author’s presentation delivered at the international conference “Swan song of the anthropocentric administration in Europe” held in Prague on 27-28 November 2025. The manuscript was prepared with the assistance of artificial intelligence tools, which were used in full alignment with current standards for academic writing and applicable ethical guidelines. Specifically, GPT-5.2 (OpenAI) was used for language refinement, proofreading and citation, whereas Gemini (Google) and GPT Deep Research Mode were used to facilitate research and literature review. I would like to thank Jakub Handrlica for his valuable support and guidance throughout the entire project. I also find it fit to gratefully acknowledges the intellectual contribution of Juli Ponce Solé, whose insightful discussions with me helped me refine my own ideas and approach the presented research from new perspectives. Last but not least, I would like to also thank Alessia Monica and Leonardo Scuto for their help with the publication process and their patience. ↑
2. See Parliamentary Print No. 845/0: Proposal by Members of Parliament Tomáš Dubský, Milada Voborská, Martina Ochodnická, Jiří Havránek, and Jiří Carbol for the enactment of a law amending Act No. 128/2000 Coll., on Municipalities (Municipal Establishment), as amended, and other laws in connection with supporting cooperation among municipalities, proposing the introduction of a new Section 15a of the Administrative Procedure Code.

   The wording of the proposed Section 15a (1) reads as follows (translated by the author): «Where the nature of the matter under consideration, the protection of the rights of the persons concerned, or the protection of the public interest does not require an act in the proceedings to be performed by an authorized official, the act may be performed automatically without the participation of an authorized official. An act may not be performed in this manner, in particular, where it requires the exercise of administrative discretion or concerns a decision on a remedy». ↑

3. See Section 35a of the German Verwaltungsverfahrensgesetz (VwVfG) (Administrative Procedures Act), which reads as follows (translated by the author): «An administrative act may be issued entirely by automated means, provided that this is allowed by law and that neither administrative discretion nor any margin of assessment is involved». ↑
4. See e.g. M. Martini, D. Nink, Subsumtionsautomaten ante portas? – Zu den Grenzen der Automatisierung in verwaltungsrechtlichen (Rechtsbehelfs-) Verfahren, in Deutsches Verwaltungsblatt, 133, 17, 2018, pp. 1128-1137. ↑
5. A typical example can be found in the field of administrative punishment, where the law may authorize the imposition of a fine “up to” a certain amount, but also in many other cases. See further. ↑
6. Many continental European systems share a similar doctrinal concept developed mainly in German administrative law, which is represented by different terms of the same normative meaning in different European languages, e.g. správní uvážení in Czech, Ermessen or Verwaltungsermessen in German, pouvoir discrétionnaire in French or discrezionalità amministrativa in Italian. ↑
7. See e.g. H. Maurer, C. Waldhoff, Allgemeines Verwaltungsrecht (General Administrative Law), XXI Ed., C.H. Beck, München, 2024, pp. 147 et seq. Similarly e.g. in D. Hendrych et al., Správní právo. Obecná část (Administrative Law. General Part), IX Ed., C.H. Beck, Prague, 2016, p. 55, G. Cartier, Administrative Discretion: Between Exercising Power and Conducting Dialogue, in C.M. Flood, L. Sossin, Administrative Law in Context, II Ed., Emond Montgomery Publications, Toronto, 2013, pp. 381 et seq., P. Craig, EU Administrative Law, II Ed., Oxford University Press, Oxford, 2012, pp. 401 et seq., J.H. Grey, Discretion in Administrative Law, in Osgoode Hall Law Journal, 17, 1, 1979, pp. 107-132, etc. ↑
8. K.F. Warren, Administrative Discretion, in E. Berman (ed.), Encyclopedia of Public Administration and Public Policy, II Ed., Routledge, New York, 2007, pp. 58-62. ↑
9. Ibidem. ↑
10. While examples of this type of discretion can be found in any administrative sentencing, the discretionary aspects are put together nicely e.g. in M.N. Lintvedt, Putting a price on data protection infringement, in International Data Privacy Law, 12, 1, 2022, pp. 1-15. ↑
11. See also F.A. Hessick, C.B. Hessick, Facts, Policy, and Discretion, in University of California, Davis Law Review, 59, 2026, pp. 1581-1633. ↑
12. See e.g. H. Maurer, C. Waldhoff, Allgemeines Verwaltungsrecht, 21. ed., C. H. Beck, 2024, pp. 150-172, or also C. Hufen, Ermessen und unbestimmter Rechtsbegriff, in Zeitschrift für das Juristische Studium, 2010, pp. 603–607. ↑
13. See J.P. Solé, Artificial intelligence and interaction between machines and humans: discretionary powers, human reserve in the public sector and the EU Regulation of 2024, in CERIDAP, 1, 2026. ↑
14. See K.C. Davis, Discretionary Justice: A Preliminary Inquiry, LSU Press, Baton Rouge, 1969. ↑
15. This idea is pointed out in the case law of administrative judiciary repeatedly, see e.g. Supreme Administrative Court of the Czech Republic, decision of 23 March 2005, No. 6 A 25/2002-42. ↑
16. See J.H. Grey, Discretion in Administrative Law, in Osgoode Hall Law Journal, 17, 1, 1979, pp. 107-132. ↑
17. Ibidem. ↑
18. See Supreme Administrative Court of the Czech Republic, judgment of 13 March 2008, No. 5 As 51/2007-110, or also Supreme Administrative Court of the Czech Republic, judgment of 22 May 2008, No. 5 As 9/2008-106. ↑
19. See M. Lipsky, Street-Level Bureaucracy: Dilemmas of the Individual in Public Services, Russell Sage Foundation, New York, 1980, pp. 81 et seq. ↑
20. See e.g. K.C. Davis, Discretionary Justice: A Preliminary Inquiry, LSU Press, Baton Rouge, 1969, p. 126. See also A. McHarg, Administrative Discretion, Administrative Rule-Making and Judicial Review, SSRN, 2017. ↑
21. See Committee of Ministers of the Council of Europe, Recommendation No. R (80) 2 of the Committee of Ministers to Member States concerning the Exercise of Discretionary Powers by Administrative Authorities, 11 March 1980. ↑
22. See G.E. Metzger, K.M. Stack, Internal Administrative Law, in Michigan Law Review, 115, 2017, pp. 1239-1307. ↑
23. See N. Marsch, V.M. Tünsmeyer, The Principle of Proportionality in German Administrative Law, in S. Ranchordás, B. de Waard (eds.), The Judge and the Proportionate Use of Discretion: A Comparative Administrative Law Study, Routledge, 2015, pp. 13–42. ↑
24. Ibidem. See also W.F. West, Structuring Administrative Discretion: The Pursuit of Rationality and Responsiveness, in American Journal of Political Science, 28, 2, 1984, pp. 340-360. This structuring effect should not be understood as permitting administrative guidelines to replace or exhaust the discretion conferred by legislation. In German administrative law, such guidelines are generally binding only within the administrative hierarchy and do not alter the statutory obligation to exercise discretion in accordance with the purpose of the empowering provision and the circumstances of the individual case. Although consistent application of a guideline may produce an indirect external effect through the principle of equality and the administration’s self-binding practice (Selbstbindung der Verwaltung), the authority must remain capable of departing from the general approach where relevant particularities render the case atypical or otherwise require differentiated treatment. Administrative guidelines may therefore channel and standardize discretion, but they cannot lawfully convert it into an inflexible rule that excludes individualized assessment. ↑
25. See S. Easton, C. Piper, Sentencing and Punishment, V Ed., Oxford University Press, Oxford, 2022, pp. 41-72. Also, numerous instances of various fining guidelines can be found, such as European Data Protection Board, Guidelines 04/2022 on the Calculation of Administrative Fines under the GDPR, Version 2.1, 24 May 2023, or European Commission, Guidelines on the method of setting fines imposed pursuant to Article 23(2)(a) of Regulation No. 1/2003, OJ C 210, 1 September 2006. ↑
26. See S. Easton, C. Piper, Sentencing and Punishment, V Ed., Oxford University Press, Oxford, 2022, pp. 41-72. ↑
27. See G.E. Metzger, K.M. Stack, Internal Administrative Law, in Michigan Law Review, 115, 2017, pp. 1239-1307. ↑
28. Such is, after all, also the logic behind Section 35a of the German VwVfG, or proposed Section 15a of the Czech Code of Administrative Procedure. ↑
29. Such cases might include competition regulators deciding whether to approve corporate mergers by applying broad economic concepts and complex assessment, municipal planning authorities exercising broad discretion when deciding whether to allow construction projects, environmental permits for industrial projects, or some types of politically, socially or otherwise sensitive decisions such as expropriation etc. ↑
30. See G.E. Metzger, K.M. Stack, Internal Administrative Law, in Michigan Law Review, 115, 2017, pp. 1239-1307. ↑
31. This observation, of course, is not necessarily universally applicable. For example, German administrative law traditionally distinguishes between licensing regimes that constitute bound administrative decisions (“gebundene Genehmigungen”), such as many building permits or permits under the “Bundes-Immissionsschutzgesetz”, and genuinely discretionary licensing regimes (“Ermessensgenehmigungen”). In the former category, once the statutory requirements are fulfilled, the authority is generally obliged to issue the license, meaning that § 35a VwVfG does not preclude the use of fully automated administrative acts. By contrast, where licensing powers are genuinely discretionary, the discretion often remains substantively relevant in practice, for example with regard to the imposition of ancillary provisions (“Nebenbestimmungen”) attached to the license. The present example is therefore intended to illustrate a phenomenon observable in a number of legal systems rather than to describe a universal feature of licensing procedures. ↑
32. See e.g. B. Schwartz, The Paradox of Choice, in S. Joseph (ed.), Positive Psychology in Practice, Wiley-Blackwell, Chichester, 2015, pp. 121-138. ↑
33. G.E. Metzger, K.M. Stack, Internal Administrative Law, in Michigan Law Review, 115, 2017, pp. 1239-1307. ↑
34. See J.C. Covilla, Artificial Intelligence and Administrative Discretion: Exploring Adaptations and Boundaries, in European Journal of Risk Regulation, 16, 1, 2025, pp. 36–50. ↑
35. See e.g. J.P. Solé, Artificial intelligence and interaction between machines and humans: discretionary powers, human reserve in the public sector and the EU Regulation of 2024, in CERIDAP, 1, 2026, or M. Martini, D. Nink, Subsumtionsautomaten ante portas? – Zu den Grenzen der Automatisierung in verwaltungsrechtlichen (Rechtsbehelfs-) Verfahren, in Deutsches Verwaltungsblatt, 133, 17, 2018, pp. 1128-1137. ↑
36. Both during our conversation, but also in his publications, see e.g. J.P. Solé, Artificial intelligence and interaction between machines and humans: discretionary powers, human reserve in the public sector and the EU Regulation of 2024, in CERIDAP, 1, 2026. ↑
37. See also C. Harlow, Global Administrative Law: The Quest for Principles and Values, in European Journal of International Law, 17, 1, 2006, pp. 187-214. ↑
38. Such is the case of the Czech Republic, but also other European administrative systems traditionally relying on administrative discretion in almost every aspect of decision-making. ↑
39. See also O. Butler, Algorithmic Decision-Making, Delegation and the Modern Machinery of Government, in Oxford Journal of Legal Studies, 45, 3, 2025, pp. 727-752. ↑
40. See also A. Lengyel, Automated Decision-Making in Administrative Penal Matters, in Studia Iuridica Cassoviensia, 13, 2025, pp. 107-119. ↑
41. See further Y. Li, The Applicability of Automated Administrative Penalties: Legal Challenges and Regulatory Responses, in Journal of Humanities, Arts and Social Science, 8, 8, 2024, pp. 1960-1964. ↑
42. In the light of preceding references, it has to be noted that Sec. 35a VwVfG mentioned earlier is only applicable to fully automated decision-making, meaning that discussion should strictly speaking distinguish between fully automated administrative acts and decision-support systems, where the final administrative decision remains attributable to a human official. This article deliberately remains technologically neutral, whereas additional questions fall outside the scope of this article. ↑
43. E.g. in a form of a simplified decision such as payment orders, summary judgments etc. ↑
44. As is pointed out and argued by Timo Rademacher, who explains that machine-learning systems operate through statistical patterns and therefore face a problem of Einzelfalladäquanz, meaning adequacy to the specific case. See T. Rademacher, Recht ohne Zufall? Algorithmische Rechtsverwirklichung in der freiheitlichen Demokratie, Nomos Verlag, 2025, pp. 291-294. ↑
45. At the same time, such a model would require careful institutional design. Mechanisms would need to ensure that human review remains meaningful and that officials retain the authority and willingness to depart from automated outcomes when justified by the circumstances of the case. Without such safeguards, there is a risk that automated decisions might become de facto final in practice, with human oversight remaining merely formal. These concerns, however, relate primarily to the institutional design and governance of automated decision-making systems and will therefore be addressed in greater detail in a separate study focusing specifically on the risks associated with automation in public administration. ↑
46. See T. Rademacher, Recht ohne Zufall? Algorithmische Rechtsverwirklichung in der freiheitlichen Demokratie, Nomos Verlag, 2025, pp. 291-323. ↑
47. In this regard, see e.g. M. Suksi, Administrative Due Process when Using Automated Decision-Making in Public Administration: Some Notes from a Finnish Perspective, in Artificial Intelligence and Law, 29, 2021, pp. 87-110, U.B.U. Roehl, Automated Decision-Making and Good Administration: Views from Inside the Government Machinery, in Government Information Quarterly, 40, 4, 2023, or J.C. Covilla, Artificial Intelligence and Administrative Discretion: Exploring Adaptations and Boundaries, in European Journal of Risk Regulation, 16, 1, 2025, pp. 36-50. ↑