1. Introduction
Public administrations have increasingly turned to artificial intelligence (AI) to support or fully automate their decision-making.Footnote 1 In 2024, 940 cases of AI use in the public sector of 30 European countries were identified (Brizuela et al., Reference Brizuela, Montino, Galasso, Polli, Bosch, De Vizio, Tangi, Combetto and Gori2024, p. 45), an increase of almost 30 percent compared to the numbers reported at the end of 2022 (Tangi et al., Reference Tangi, van Noordt, Combetto, Gattwinkel and Pignatelli2022, p. 35). The application of AI in the public sector spans from strengthening the connection with citizens and routine administrative tasks, such as supporting the enforcement of road traffic offences, to critical tasks that significantly impact people’s lives, like processing welfare benefits and taxes, predicting crime and assessing a person’s risk of committing it, and supporting decisions in migration and asylum procedures.
Despite the profound impact that the use of AI tools in public decision-making is already having – and will continue to have – on the lives of large segments of the population, one key party remains ostensibly absent from decisions about their use and development: the public. People do not know who decides to deploy AI in the public sector, when those decisions are made and how Public-Sector AI is being designed and developed. If citizens become aware of the deployment of AI systems, it is typically when an individual measure affecting them is taken, and the damage is done. As a result of this opacity, citizens are, in effect, prevented from being involved in any way in the process of Public-Sector AI development.
However, the deployment of AI systems by public administrations should not be opaque to citizens since the lack of public involvement may hamper the effective translation of public values in those machines. In a bid to prevent the institution of agnostic technological bureaucracies, public participation has been a feature constantly present in the discourse around AI development, especially in the public sector. In the latter context, participation was suggested as a potential silver bullet to enhance public values and address the democratic deficit of Public-Sector AI systems (Coglianese & Lampmann, Reference Coglianese and Lampmann2020; Corbett et al., Reference Corbett, Denton and Erete2023, p. 1; Gilman, Reference Gilman2023, p. 2; Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022; Groves, Reference Groves2023, pp. 36–38, 53; Sanchez-Graells, Reference Sanchez-Graells2024; Sloane et al., Reference Sloane, Moss, Awomolo and Forlano2022, p. 1; Sloane, Reference Sloane2024, p. 1; Wong et al., Reference Wong, Morgan, Straub, Hashem and Bright2022, p. 4). Enabling citizens to participate and signal possible issues with the use of a certain technology during its development process could make the utilization of technologies by public administrations more user-friendly and avoid future harms arising from that use. Given this prospect of public participation acting as a means to overcome acute transparency and accountability issues raised by AI systems (Coglianese & Lehr, Reference Coglianese and Lehr2019; Engstrom et al., Reference Engstrom, Ho, Sharkey and Cuéllar2020, pp. 75–78; Hickok, Reference Hickok2022, pp. 3–4), the topic of participation has sparked a sudden curiosity in scholarship dwelling on Public-Sector AI.
Notwithstanding the growing interest in exploring ways to incorporate participation into the development of Public-Sector AI, there is still no clear answer on how to achieve it. If citizen participation is indeed important to ensure that AI systems deployed in the public sector align with public values, it becomes evident that public administrations must begin involving individuals and organizations in these processes. The key question, however, is how. And what lessons can be drawn from participatory experiences in other administrative domains? Despite the considerable conceptualizations around the expectations and promises of public participation in this context, legal scholarship has yet to engage with its operationalization. Against this backdrop, this paper takes a first step in examining the legal avenues available to natural and legal persons to participate in the development of AI systems by the EU and member states’ public administrations.
Two disclaimers should be made, however. The scope of this paper is limited, on the one hand, to the design and development stageFootnote 2 of the Public-Sector AI systems and does not entail participation after the AI system has been effectively deployed. On the other hand, it provides a general doctrinal analysis of the possibilities to involve individuals and organizations in the development processes of AI systems by EU and member states’ institutions, bodies and agencies based solely on EU law and administrative law principles common to member states’ legal systems, such as the right to good administration. However, it does not go deep into the specificities of each administrative law jurisdiction.
This paper is organized into five parts. First, it will examine why citizen participation in the design and development of Public-Sector AI could matter. Second, it will explore public participation under traditional administrative law, examining the opportunities for citizen involvement in administrative rulemaking and decision-making. Third, it will examine whether there are avenues for citizen participation in public procurement procedures aimed at purchasing AI tools. Fourth, it will analyze the instances of stakeholder participation in the novel regulatory framework of AI in the EU area – the Artificial Intelligence Act (AIA).Footnote 3 Lastly, it will explore the possibility of applying public participation avenues foreseen in environmental law to Public-Sector AI development, considering its strengths and weaknesses and, more broadly, the effective added value of public participation in AI development by public administrations.
2. Public participation as a potential shaper of Public-Sector AI for the public good
Public participation is not a new idea in the block. Public authorities have long involved citizens in public governance to enhance the provision of public services and uphold the public interest. The involvement of the public in environmental decision-making is a good example of that (Bándi, Reference Bándi2014), but not the only one. Public participation is also used in urban planning (Adjei Mensah et al., Reference Adjei Mensah, Andres, Baidoo, Eshun and Antwi2017) and, in particular, is used considerably at the local level, such as to approve participatory local budgets (Cabannes, Reference Cabannes2004). Yet, in recent times, public participation has gained traction in current AI discourse and governance (Delgado et al., Reference Delgado, Yang, Madaio and Yang2023; Gilman, Reference Gilman2023; Groves, Reference Groves2023; Sloane, Reference Sloane2024).Footnote 4 Looking at the recently published AIA, there is an ambition to include stakeholders in the loop of AI development. Stakeholder participation is foreseen in the drafting of standardsFootnote 5 and codes of conductFootnote 6 that will establish AI systems requirements and the obligations incumbent on the providers of those systems. Moreover, particularly concerning AI systems to be deployed in the public sector, the AIA provides for the involvement of stakeholders in the performance of fundamental rights impact assessments (FRIA) for high-risk AI systems to be deployed in the public sector (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, pp. 1–2; Corbett et al., Reference Corbett, Denton and Erete2023, p. 1; Gilman, Reference Gilman2023, p. 2; Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236; Groves, Reference Groves2023, pp. 36–38; König & Wenzelburger, Reference König and Wenzelburger2021, p. 3; Sloane, Reference Sloane2024, p. 1; Wong et al., Reference Wong, Morgan, Straub, Hashem and Bright2022, p. 4).Footnote 7
Indeed, incorporating public participation into decision-making processes has consistently proven to hold multiple benefits in the past. Historically, the public’s involvement in decision-making has been identified with processes that are more transparent, accountable and responsive to the people’s needs, empowering individuals and communities and fostering the overall well-being of societies (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, p. 3). Furthermore, it provided an opportunity to learn from and include the perspectives of those affected by the measure under public debate (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, p. 3; Seger et al., Reference Seger, Ovadya, Garfinkel, Siddarth and Dafoe2023, p. 3), while also potentiating the mastering of new skills, the emergence of novel networks and the self-determination of the participants (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, p. 3). More specifically, in environmental decision-making, public participation has been associated with improved substantive outcomes and enhanced democratic legitimacy of decision-making (Lee, Reference Lee2014, pp. 176–180; Lee & Abbot, Reference Lee and Abbot2003, pp. 82–88). These benefits are typically demanded by decision-making processes in important fields that risk causing harm to individuals and communities and lack strong democratic legitimacy, such as the environmental field.
Considering that AI deployment in the public sector shares similar features with those fields (Smuha, Reference Smuha2021, pp. 9–13), the recent urge to involve citizens in Public-Sector AI development is attributed to industry practitioners and experts’ conviction that public participation can be a promising “fix” for the damages caused by the deployment of AI (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, p. 1; Corbett et al., Reference Corbett, Denton and Erete2023, p. 1; Gilman, Reference Gilman2023, p. 1; Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236; Morán et al., Reference Morán, Ross and de la Maza2024; Sloane, Reference Sloane2024, p. 1). Expanding evidence has shown that AI application, both in the public and private sectors, generates multiple harms, intensifying existing biases and discriminations, undermining democracy and producing injustices of every kind.Footnote 8 These harms affect individuals, communities and societies worldwide, hurting, in particular, marginalized communities and low-income and disadvantaged individuals.Footnote 9 To tackle them, scholars from diverse fields believe that these demonstrated benefits of public participation can be an effective countermeasure to AI societal harms (Wong et al., Reference Wong, Morgan, Straub, Hashem and Bright2022, pp. 2, 4).
Participation can contribute to the mitigation of AI societal harms, especially those resulting from its use by public authorities, by improving the quality of public services and fostering public values such as transparency, accountability and legitimacy of public activity (removed for peer-review purposes). Regarding the former, Gilman (Reference Gilman2023) contends that, in addition to preventing AI biases and injustices, seeking the public’s input when deciding to deploy an AI system may avert public authorities from going ahead with a project to develop an AI tool that is deemed to fail due to a lack of public approval (p. 3; see also Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236). Simultaneously, it allows for distinct societal values to be considered when making such a decision, precluding exclusive technocratic decision-making. Finally, by enabling the public to promptly highlight “(…) where there are sensitive issues and what the main concerns are” (Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236), participation can mitigate risks to the population, preventing poor performance in the provision of public services with the support of these tools.
In addition to enhancing public service provision and administrative performance, public involvement in public entities’ decisions to deploy AI makes them more legitimate and accountable (Coglianese & Lampmann, Reference Coglianese and Lampmann2020, p. 195; Gilman, Reference Gilman2023, p. 3; Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236; König & Wenzelburger, Reference König and Wenzelburger2021, pp. 3–4). König and Wenzelburger (Reference König and Wenzelburger2021) argue that, since the “epistemic basis” (p. 4) of AI systems is constructed during their design and development, involving individuals and organizations in the early design stages of AI systems can help ensure they reflect public values (pp. 3–4). This involvement precludes an expert-driven approach and fosters acceptance and legitimacy of public decision-making (König & Wenzelburger, Reference König and Wenzelburger2021, pp. 3–4). Thus, in a way, “(…) by participating in development, people (i.e., end-users, citizens, communities, the general public) can become empowered to determine the outcomes and consequences of AI systems” (Corbett et al., Reference Corbett, Denton and Erete2023, p. 1). Furthermore, citizen involvement, especially when AI tools are developed in cooperation with private parties, may prevent commercial takeover of that development and foster democratic legitimacy and control of public-private collaborations (Groves, Reference Groves2023, p. 38; Wong et al., Reference Wong, Morgan, Straub, Hashem and Bright2022).
More recently, this “participatory turn” (Birhane et al., Reference Birhane, Isaac, Prabhakaran, Díaz, Elish, Gabriel and Mohamed2022, p. 1) has arrived in the legal realm. Following this tendency to praise the potential benefits of public participation in the AI field, some legal scholars suggested that involving citizens in the development of these technologies, particularly if outsourced to private companies, “(…) can help ensure that AI tools better match the values and priority of the public and that they enjoy greater public acceptance when put into use” (Coglianese & Lampmann, Reference Coglianese and Lampmann2020, p. 195), as well as alleviate the adverse outcomes of automated administrative decisions on citizens (Coglianese & Lampmann, Reference Coglianese and Lampmann2020; Sanchez-Graells, Reference Sanchez-Graells2024). Smuha (Reference Smuha2024) goes even further to argue that “[g]iven the potential consequences linked thereto, it is fair to claim that there should be no algorithmisation without representation (…)” (p. 228).
Considering this optimistic outlook on public participation in AI development across various academic fields and its potential positive effects on AI use, pushing forward public participation in the design and development of Public-Sector AI seems not just beneficial but almost inevitable. Therefore, to test this hypothesis and reap the benefits of public participation, legal systems must create avenues for public involvement in the administrative procedures for AI development by public entities. However, if it seems that participation can play a significant role in promoting a more secure and fair use of AI in the public sector, neither the traditional legal frameworks applicable to EU and national public administrations’ activities nor the new digital regulatory framework abounds in avenues for public participation, as I will demonstrate below.
3. Public participation in traditional administrative law
General administrative law stands as the initial frontier when exploring the legal avenues for the public to be involved in AI development by public administrations. As the field of law that governs public administrations’ organization and activity, as well as the rights of natural and legal persons against administrative power (Ranchordas, Reference Ranchordas2024, pp. 16–17), administrative law contains the rules that oversee day-to-day administrative activity, including the administration’s digitalization. It is true that these rules vary from jurisdiction to jurisdiction, especially since administrative law is deeply connected with states’ sovereignty.Footnote 10 This diversity averts a detailed analysis of the specificities of each jurisdiction at this stage. Instead, this paper will focus on EU administrative law rules as a sample and the mutual repository of common principles, rules and doctrines of member states’ administrative law frameworks to explore the possibility of applying administrative rules and principles to allow for public participation in Public-Sector AI development.
In fact, EU administrative law comprises some of the administrative law features shared by member states, such as the right to good administration. This field of law contains the rules and principles governing the organization and activity of EU administration in implementing and applying Union law (Widdershoven, Reference Widdershoven and Seerden2012, p. 245). Accordingly, this Section will delve into EU primary and secondary law, particularly emphasizing the general rules and principles stemming from the EU Treaties, the Charter of Fundamental Rights of the European UnionFootnote 11 and the case law of the Court of Justice of the European Union (CJEU).
When considering public participation in administrative activity, one must address two separate fronts: on the one hand, the rulemaking power of public authorities and, on the other hand, their decision-making power. Given that these administrative powers entail distinct goals and result in different acts of power, the opportunities for public involvement in exercising them must be necessarily diverse.
3.1. Participatory rights in rulemaking
Public participation in rulemaking procedures has long been a topic of discussion in EU administrative law, and with the advent of digitalized administration, this debate may experience a revival. Individuals’ involvement in procedures for enacting general rules that implement or develop legislative acts can have far-reaching and significant effects, as it allows citizens to directly influence the creation of the rules and indirectly affect the individual cases to which these rules will apply. In the digital age, the significance of public participation in rulemaking procedures is heightened, as it can serve as a pivotal pathway for both individuals and organizations to engage in the development of Public-Sector AI.
In fact, some of the decisions that administrative authorities must make during AI development may qualify as regulations. For instance, this may be the case of the initial decision authorizing the implementation of AI systems, which pertains to the organization of the functioning of public authorities.Footnote 12 Assuming that a public authority must adopt a regulation sanctioning the deployment of AI systems to support or automate the performance of administrative tasks, the first opportunity for public participation appears in the adoption process of this regulation.
Recent EU practice demonstrates that public participation is systematically embedded in regulatory processes across policy areas. The representative democracy and participatory democracy principles enshrine the right of every citizen to participate in the democratic life of the Union.Footnote 13 This openness to participatory democracy in rulemaking procedures is reflected in the Commission’s practice of conducting consultations with the parties concerned.Footnote 14 For instance, the Commission recently launched a public consultation on the revision of the current EU public procurement directives.Footnote 15 Public consultations were also held regarding the AIA, including on the requirements for high-risk AI systems.Footnote 16 Moreover, the Commission facilitated the drafting of a General-Purpose AI Code of Practice involving more than 1,000 stakeholders.Footnote 17 Thus, at least some avenues for participation in rulemaking exist in the EU.
However, public participation in rulemaking procedures is not the norm, and its accessibility to the general public varies significantly. In EU law, the general rule remains the absence of participation in rulemaking (Craig, Reference Craig2018, pp. 316, 324; Curtin & Mendes, Reference Curtin and Mendes2011, p. 12; Mendes, Reference Mendes2011, pp. 142–161, 193–229). Even where public consultations are conducted, they primarily target representative groups and civil society organizations rather than individual participants (Lee, Reference Lee2014, p. 185).Footnote 18 Furthermore, individual participation lacks a compulsory character, as confirmed by longstanding CJEU case law denying the existence of an individual right to participate in rulemaking procedures.Footnote 19 Consequently, besides a general duty of the Commission to conduct consultations when proposing regulations, participation in rulemaking procedures in the EU is dependent on informal practices, such as stakeholders’ consultations, expert seminars, contacts with civil society organizations or online consultations with the public (Mendes, Reference Mendes2011, p. 12). Beyond this realm, participation is confined to the right to be heard in individual decision-making procedures.Footnote 20
3.2. Participatory rights in decision-making
Contrary to rulemaking procedures, participation is effectively integrated into administrative decision-making through the right to access documentsFootnote 21 and the right to be heard before an individual measure adversely affecting individuals is taken.Footnote 22 These rights are part of the European culture of rights granted to citizens to guard them vis-à-vis administrative power and are recognized by EU administrative law through the right to good administrationFootnote 23 and by member states’ national legal systems more generally. By enabling a peek into administrative action and giving voice to citizens, these two rights of defense are especially relevant in permitting public participation in European administrative activity. Jointly, they compose the participatory rights granted in administrative decision-making procedures. However, rather than granting genuine avenues for public participation in European public administrations’ AI development procedures, these participatory rights, if anything, merely offer potential access to information.
Regarding the right to be heard, this right is an ineffective avenue to empower individuals and organizations to participate in the procedures for AI development by EU and member states’ public administrations. Despite giving voice to people vis-à-vis the administration when it takes an individual measure that negatively affects them, the right to be heard, as it is envisioned, is not suitable for being operationalized in relation to the type of decisions involved in AI development. This inapplicability is primarily due to the fact that the right to be heard requires a bilateral relationship between a natural or legal person addressee of an individual decisionFootnote 24 and an administrative entity to be actionable (Mendes, Reference Mendes2011, pp. 161–162, 170, 190). The required bilateral relationship entails a direct connection between the act of the administrative entity and the potential adverse effects on the individual’s legal sphere stemming from that act,Footnote 25 which is typically found in administrative decision-making procedures leading to individual measures with one or more individualized addressees.
However, a comparable link cannot be found in the procedures for designing and developing AI systems for the public sector, as the acts performed by administrative entities in those procedures fall outside particular administrative decision-making procedures. In fact, AI systems created by public entities are not tailored for a specific individual decision-making procedure but rather for a multitude of administrative decision-making procedures concerning a particular task (whether, for example, the enforcement of a road traffic offense or the assessment and collection of taxes). As a result, the acts carried out by those entities when developing AI systems will, if anything, have a widespread impact. They will affect a vast number of people, albeit indirectly, because AI systems will be used in a variety of decision-making procedures. Thus, in its pure form, the right to be heard is unsuitable for enabling participation in Public-Sector AI development procedures since those procedures autonomously pre-exist administrative decision-making procedures where the right is intended to be exercised (Craig, Reference Craig, Peers, Hervey, Kenner and Ward2021, p. 1129).
Additionally, even though the CJEU’s Al-Jubail caseFootnote 26 expanded the right to be heard to “hybrid”Footnote 27 regulations, this protection still does not extend to acts involved in developing Public-Sector AI. The reasoning behind this expansion requires that acts of a general nature produce direct and individual effects on identifiable persons, akin to those produced by individual administrative decisions.Footnote 28 For this ruling to apply to AI development, prospective participants would need to experience direct and individual effects on their legal spheres arising from the acts undertaken during AI development.Footnote 29 However, even when an AI system is designed for precise usage within a procedure affecting a relatively delimited group of people, the acts involved in its development do not directly and individually impact those people. AI development logically precedes decision-making procedures that directly affect individuals. As a result, the link between an act authorizing the use of AI or specifying the parameters of an AI system and those ultimately impacted is too indirect to confer a participatory right, rendering the right to be heard inapplicable to AI development procedures.
In contrast, if individuals lack the right to be heard in the procedures for public entities to implement AI systems, the same cannot be said for its analogous participatory right. As a key tool that allows the public to participate in administrative decision-making, the right of access to documents empowers natural and legal persons with the information required for operative participation. In its broader version, the right to access documents entails access to all records and information of EU institutionsFootnote 30 by any person residing or having a registered office in the EU, regardless of having any specific legal or economic interest in that access.Footnote 31 Accordingly, in this extended version, the right to access documents could act as a facilitator of information regarding AI development by public authorities, even though the procedures involved in Public-Sector AI development do not comprehend the existence of individualized addressees. In fact, given its broader scope, the right of access to documents could clearly encompass access to all documents related to the administration’s design and development of AI systems by any requesting party. In theory, it should even comprise third-party documents in cases where the development has been outsourced (Lazarov, Reference Lazarov, Peers, Hervey, Kenner and Ward2021, p. 1159).
However, given both the public and private stakes involved in the development of AI systems for public administrations, two key exceptions could be invoked to restrict access to documents related to the development of AI by public entities.Footnote 32 The commercial interest exception protects proprietary data when private entities are involved in Public-Sector AI development, particularly when governments outsource these projects (Lazarov, Reference Lazarov, Peers, Hervey, Kenner and Ward2021, p. 1167).Footnote 33 Additionally, even if the AI system is built in-house by public officials, the internal documents exception allows public entities to withhold documents if their disclosure could impair the decision-making process – though this must be justified with substantial and specific reasons (Lazarov, Reference Lazarov, Peers, Hervey, Kenner and Ward2021, p. 1169).Footnote 34 These exceptions suggest that access to documents related to Public-Sector AI development may be limited despite its broader and more promising scope.
Against this backdrop of participatory rights, EU administrative law seems to be ineffective in providing avenues for natural and legal persons to participate in AI development by public administrations. On the one hand, the general absence of direct public participation in administrative rulemaking indicates that what seemed or should have been a preliminary doorway for individuals and organizations to enter and become involved in the development process of Public-Sector AI is, indeed, closed. On the other hand, even if the right of access to documents can grant some degree of public participation in the implementation process of Public-Sector AI, as exceptions to this access are circumscribed, this instrument will largely be useless without a mechanism to incorporate the views of natural and legal persons that the consultation of the disclosed documents will unleash – as the right to be heard could be. Even if relying on existing extra-administrative mechanisms, the restricted, lengthy and costly proceedings before the CJEUFootnote 35 or the unenforceable EO action against maladministrationFootnote 36 appear to be unsatisfactory in attaining the goal of citizen participation.
4. Public participation in public procurement
Public procurement offers another potential channel through which citizens could be involved in the development of Public-Sector AI systems, given that public authorities often outsource their design and maintenance to private contractors (Adopt AI Study, 2024, p. 26; Focus on AI in the Dutch central government, 2024, p. 30; Engstrom et al., Reference Engstrom, Ho, Sharkey and Cuéllar2020, p. 18; Wan & Sieber, Reference Wan and Sieber2025, pp. 10–11).Footnote 37 Collaboration with private parties may entail purchasing “off-the-shelf” solutions or outsourcing the creation of tailor-made AI systems to private entities. Yet, despite the centrality of procurement when outsourcing AI development, EU public procurement law offers citizens only limited and indirect avenues to participate in such procedures.
When contracting out AI, public authorities will often launch a public procurement procedure to select the economic operator that will supply or develop the system. At least in the EU area, EU institutions or agenciesFootnote 38 and member states’ central, regional and local authorities, bodies governed by public law and their associationsFootnote 39 must follow a public procurement procedure if they plan to award a contract for the supply of goods, execution of works or provision of services, in exchange for a consideration.Footnote 40 This procedure is governed by a set of public procurement laws,Footnote 41 which prescribe procedures and requirements to be followed when public authorities buy goods, works and services from private entities.Footnote 42 This legal framework is designed to regulate competition for public contracts, ensuring transparency, equal treatment and efficiency of the internal market.Footnote 43
Considering the common use of public procurement procedures to buy AI tools, coupled with the highly proceduralized character of procurement processes, these procedures would be a prime moment to introduce public participation. In general, integrating civil society in public procurement procedures was demonstrated to potentially decrease transaction costs and duration of the procedures, as well as enhance the quality of the final contract (Cravero, Reference Cravero2019, p. 42). Particularly in relation to AI development in the public sector, as Coglianese and Lampmann (Reference Coglianese and Lampmann2020) recognized, leveraging participatory mechanisms when procuring AI systems for the public sector “(…) could assist both procurement officers tasked with navigating the field of Al as well as help agency officials and vendors improve their design and use of Al tools” (p. 194). Furthermore, they assert that public participation could be valuable both at the early design stage – by creating opportunities to consult citizens and allowing them to comment on the main design choices of the AI systems – and later, by requiring private contractors to engage interested parties in discussions about the use of AI tools (Coglianese & Lampmann, Reference Coglianese and Lampmann2020, pp. 194–195).
Notwithstanding these potential benefits of public participation in procurement procedures, current EU public procurement rules do not contemplate any avenues for the involvement of individuals and organizations in those procedures or in the performance of public contracts. Not only is public participation in procurement processes fairly rare in procurement frameworks worldwide (Cravero, Reference Cravero2019, p. 42), but also this participation is not envisaged in the rules that govern procurement by EU institutions and agencies,Footnote 44 nor the directives applicable to procurement by member states.Footnote 45 On the contrary, “participation” is envisioned in EU public procurement law as the ability of economic operators to take part and compete in the procurement procedures.Footnote 46 Consequently, EU public procurement law is primarily structured around contracting authorities and economic operators, leaving little formal space for citizens to directly participate in procurement procedures as monitoring third parties.
Nevertheless, opportunities for indirect involvement do exist. Firstly, EU public procurement law establishes transparency obligations for contracting authorities, such as the publication of contract notices and award results in the Official Journal of the EUFootnote 47 and the electronic provision of access to procurement documents.Footnote 48 Citizens also have an option to survey the general trends of awarding procurement procedures by monitoring information disclosed by public institutions on budget implementation.Footnote 49 Secondly, individuals and organizations may also exert influence through preliminary market consultations.Footnote 50 These consultations enable contracting authorities to access market operators or independent experts before launching a procurement procedure, permitting authorities to gather information useful to prepare the procurement procedure. When undertaking these consultations, contracting authorities are allowed to consult experts, which may include not only individuals, but also civil society organizations with expertise in the field of procurement. In doing so, these experts may be capable of influencing the technical features of the procured solutions, as well as the design of the procurement process.
More indirectly, individuals and organizations may play a role in shaping public procurement procedures at early and later stages. At the outset, citizens and civil society organizations may participate in public consultations for the amendment of procurement laws or be consulted before the enactment of the Commission’s delegated acts in matters of public procurement.Footnote 51 They may also intervene in the setting of standards, such as technical standards or for social and green procurement, which will later be used when setting the requirements for the works, goods or services to be procured. Finally, they may have a lateral role – for example, by participating in environmental impact assessments linked to procurement procedures of infrastructure projects. On the other hand, legal challenges also provide a form of indirect participation. Although remedies are primarily conferred to economic operators,Footnote 52 member states’ procedural rules may extend rights of action to other interested parties, such as civil society organizations.
At the same time, member states retain a degree of discretion when transposing public procurement directives, which may be used to create participatory channels at the national level. Indeed, EU public procurement directives establish minimum requirements, leaving scope for additional regulatory detail, provided that national rules remain compatible with the principles of non-discrimination, transparency and proportionality. Within this margin, member states could, for instance, broaden the scope of the actors involved in preliminary market consultations to explicitly include citizen groups or civil society organizations or expand transparency obligations by requiring the publication of public contracts. This discretion, however, is not unlimited. Member states cannot introduce participatory mechanisms that would distort competition or undermine the equal treatment of bidders. Participation must therefore take forms that enhance transparency and accountability without interfering with the competitive structure of the procedure itself.
Against this context, while EU public procurement law does not formally provide for citizen participation, there is still some leeway for citizens and civil society organizations to shape public procurement procedures – albeit indirectly. However, these indirect avenues for public participation do not allow individuals and organizations to have a meaningful impact on the design and development of the AI systems purchased through a procurement procedure. It is true that, arguably, they might have a limited influence on the technical specifications of the system, for instance by intervening in the standard-setting of those technical specifications or by being consulted in preliminary market consultations. Nonetheless, these avenues are too indirect and depend on the adoption of co-created technical standards or the recommendations of civil society when consulted by the contracting authority. More importantly, none of the potential avenues for citizen participation in EU procurement of AI tools involves direct intervention and collaboration with contracting authorities and economic operators in the negotiation and execution of contracts. In other words, it is not possible for individuals and organizations to collaborate in the development of procured AI systems. Therefore, while public procurement processes may allow imperfect opportunities for public participation at the outset, the public is largely excluded from their core stages.
5. Public participation in the AIA
Navigating administrative activity in the digital age requires supplementing traditional administrative law provisions with the contemporary regulatory framework governing the digital landscape. Particularly regarding the design and development of Public-Sector AI, this procedure will be governed by the novel AIA,Footnote 53 which contains the main regulatory framework for designing and developing AI in the EU. The AIA establishes harmonized rules for the placing on the market, the putting into service and the use of AI systems and general-purpose AI models in the Union area.Footnote 54 These harmonized rules regarding AI systems include prohibitions of certain AI uses, specific requirements for high-risk AI systems and transparency rules for other AI systems. The AIA’s goal is to “improve the functioning of the internal market (…),” while simultaneously “ensuring a high level of protection of health, safety, [and] fundamental rights (…) and supporting innovation.”Footnote 55
5.1. The AIA’s scope of application tailored to Public-Sector AI
Aimed to be the principal regulatory framework of AI in the EU zone, the AIA has a broad personal and material scope of application, comprising, among others, Public-Sector AI systems. The AIA regulates all elements of the AI systems life cycle, applying to every entity that develops, imports, distributes or uses AI systems producing outputs within the EU,Footnote 56 including public authorities. Indeed, under the AIA, EU or member states’ public authorities, agencies or bodies may be qualified as providers or deployers of AI, depending on whether they develop or use an AI system or a general-purpose AI model (Puigpelat, Reference Puigpelat2023, p. 240).Footnote 57 The regulatory framework extends to any EU or member states’ public authority, agency or body that either develops an AI system or a general-purpose AI model or possesses such a system and directly employs it for the first time within the Union’s jurisdiction.Footnote 58 Moreover, the AIA applies to any EU or member states’ public authority, agency or body that uses an AI system under its authority.Footnote 59 Accordingly, almost any attempt by public administrations to automate administrative tasks will trigger the application of the AIA.Footnote 60
This nearly all-encompassing scope of the AIA is particularly evident considering that it appears to cover any endeavor by public authorities to implement Public-Sector AI, irrespective of whether the system is developed internally by governments or outsourced to third parties (Puigpelat, Reference Puigpelat2023, p. 240). In fact, under the AIA, a public authority, agency or body is deemed an AI provider whether it independently develops the AI system or a general-purpose AI modelFootnote 61 or if it uses a system supplied to it for first use under its name or trademark.Footnote 62 Consequently, this definition covers both the internal development of AI systems by public entities and the procurement of such development to private actors.
In light of this comprehensive qualification of public authorities as providers, the AIA will provide the primary architecture of the procedures for developing AI systems in the public sector. As EU and member states’ public administrations must comply with the requirements emerging from the AIA when developing AI systems for public use, whether in-house or through contracting out to private entities, the AIA, alongside administrative law, will undoubtedly become a key framework for the digitalization efforts of public administrations.Footnote 63 As a result, as soon as it takes effect, the AIA will be a centerpiece in assessing the legal avenues for public involvement in AI development by public administrations.
5.2. Avenues for participation in Public-Sector AI development under the AIA
Contrary to traditional administrative law, the AIA adopts mechanisms for public participation in the design and development of AI systems. The underlying intention of the European legislator in the AIA was to incorporate stakeholder participation as a feature of the AI governance framework outlined by the regulation.Footnote 64 Consequently, it foresees avenues for the participation of stakeholders in the distinct moments of the AI life cycle:
1. First, in a “pre-design and development” moment, encompassing the drafting of standards and rules of conduct to apply to providers and deployers of AI systems and general-purpose AI models;
2. Second, in the development stage of AI systems; and
3. Third, more generally, across the AI life cycle, in the governance practice and application of the AIA.
First, stakeholders may participate in formulating the standards and rules that will govern AI development and use. In this stage, prior to AI development, the public will have an opportunity to be involved in the process of (i) drafting harmonized standards concerning the requirements and obligations of high-risk AI systems,Footnote 65 (ii) formulating codes of practice to develop the obligations of providers and assess the systemic risks raised by general-purpose AI models,Footnote 66 (iii) preparing voluntary codes of conduct to apply to any provider or deployer of AI systems (Kaminski & Malgieri, Reference Kaminski and Malgieri2025, p. 265),Footnote 67 and (iv) developing, together with metrology and benchmarking authorities, measurement methodologies and benchmarks to measure levels of accuracy and robustness of high-risk AI systems.Footnote 68 Various participants are expected to be involved in these processes. They include small and medium enterprises (SME), consumer organizations, environmental and social stakeholders, notably European stakeholders’ organizations receiving EU financing, public authorities, AI providers, civil society organizations, industry, academia, research organizations and trade unions.Footnote 69
Secondly, the AIA also entails several cues of public participation in AI development processes. The participation of relevant stakeholders is legally provided during the operation of AI regulatory sandboxes.Footnote 70 Furthermore, the recitals of the AIA signal the intention – although not mandatory – to allow stakeholders’ participation in the preparation of the risk management systems for high-risk AI systemsFootnote 71 and in the execution of FRIA conducted by public entities and private entities providing public services (Kaminski & Malgieri, Reference Kaminski and Malgieri2025, p. 263–264).Footnote 72
Particularly targeting the public sector, public authorities and private entities providing public services must conduct an FRIA when using standalone high-risk AI systemsFootnote 73 for the first time.Footnote 74 The FRIA shall include, among others, information on the categories of natural persons and groups likely to be affected by the specific context of the high-risk AI system’s intended use, the detailed risks of harm likely to have an impact on those categories of natural persons and the measures to be taken in the case of the materialization of those risks.Footnote 75 In order to collect this information, public authorities may, if necessary, involve the relevant public, including representatives of the groups of individuals likely to be affected by the AI system’s intended use, experts and civil society organizations, in the performance of the impact assessment.Footnote 76
Thirdly, the AIA governance system also includes a place for stakeholders (Kaminski & Malgieri, Reference Kaminski and Malgieri2025, pp. 263–264). According to this system, stakeholders from the industry, start-ups, Small and Medium Enterprises (SME) civil society and academia with recognized expertise in the AI field shall form part of an advisory board (Kaminski & Malgieri, Reference Kaminski and Malgieri2025, p. 264).Footnote 77 The Commission will establish this board to advise and provide technical expertise to the Commission and the European AI Board in applying the AIA.Footnote 78
Given these specific provisions, one cannot say that the AIA ignores participation. On the contrary, it envisages stakeholder participation at different stages of AI development and deployment. However, this framework of participation certainly has its flaws. With few exceptions, most of the avenues provided for participation in the AIA do not relate to the AI development process itself. Instead, they are anticipated at the earlier stage of preparation of the standards and rules that will guide the process of AI development, or they occur in the overall governance system of AI systems. Furthermore, even if the AIA intends to offer avenues for participation during the development of AI systems, those avenues are not as significant as those available for setting standards and rules. Participation in AI development, especially in the crucial step of the FRIA’s execution, lacks a stringent character as it is not foreseen in the AIA provisions but in recitals.Footnote 79 Finally, the AIA prescribes a form of “stakeholder” participation that is mainly qualified and plural, rather than open to the ordinary and lay public. In general, only categories of stakeholders, such as SMEs, consumers, environmental and social stakeholders, AI providers, industry or academia, are allowed to be involved in drafting the standards and codes of practice and participating in AI regulatory sandboxes. As the name suggests, the stakeholder participation foreseen in the AIA implies that the people involved have some stake, interest or expertise in a specific undertaking. Consequently, a crucial form of participation is sidelined: that of individuals who, despite lacking a direct stake or formal institutional ties, may be directly affected or concerned with the development of AI systems in the public sector.
In sum, the AIA not only falls short of effectively establishing avenues for participation in AI development but also confines any established avenues to a narrow scope of institutional affiliation, failing to embrace all “publics” in Public-Sector AI development. For these reasons, in my view, the AIA does not provide viable avenues for citizen participation in AI development by public administrations.
6. Public participation in environmental decision-making
As illustrated in the previous Sections, neither classical administrative law provisions nor the brand-new AIA provide satisfactory avenues for a broad public engagement of lay individuals and interest groups in AI development by public administrations. However, participation appears essential for ensuring legitimate and accountable administrative action in the era of digitalized administration.Footnote 80 Thus, in searching for inspiration to enforce effective public participation in Public-Sector AI development, one should turn to EU environmental law (Smuha, Reference Smuha2021, pp. 12–16).
Indeed, EU environmental law is particularly targeted at acting on societal harms. Given the parallels between environmental harms and the types of societal risks posed by AI,Footnote 81 participation in environmental law can serve as a paragon for public involvement in Public-Sector AI development. Nonetheless, their commonalities might also give us clues that, as in environmental matters, participation in Public-Sector AI development may fall short of its goals. In this regard, the following three subsections will (i) explain why participation in environmental law provides the most suitable reference framework for participation in Public-Sector AI development, (ii) concisely review public participation in environmental law and (iii) draw conclusions on how the insights from environmental participation can inform and shape participation in Public-Sector AI development.
6.1. The case for environmental law as a model for participation in Public-Sector AI development
Environmental law stands out as one of the few fields of public law with a long-established tradition of actively involving both the general public and expert stakeholders in environmental decision-making. Public participation in environmental matters is regulated at the international,Footnote 82 EU,Footnote 83 and national levels. Accordingly, with public participation foreseen across most levels of governance, environmental law is one of the most developed test cases for public participation in administrative procedures (Peters, Reference Peters, Fraenkel-haeberle, Kropp, Palermo and Sommermann2015, p. 311; Smuha, Reference Smuha2021, pp. 12–14).
Examples of public participation in environmental procedures abound both internationally and in the EU. More importantly, many of them represent successful cases in which the involvement of citizens and interest groups in the early stages of development and approval of environmental projects was effective in preventing subsequent harms to the environment, health, and cultural and natural heritage. For instance, the broad three-level public consultations in the process of decommissioning of the Trawsfynydd Nuclear Power Plant in North Wales, United Kingdom, guaranteed its success, as the concerns raised by the public regarding local employment, long-term visual impact on the site and radiation doses during dismantling prompted amendments to the final project of decommissioning (Bond et al., Reference Bond, Palerm and Haigh2004, pp. 625–628). Similarly, public participation at the early stage of the Amsterdam Schiphol Airport’s expansion plans between 2008 and 2020 also led to agreements to limit the airport’s growth and environmental and health impacts, including noise levels and emissions (Suman, Reference Suman2024). From infrastructure projects, such as the construction of airports or high-speed railway lines, to the approval of genetically modified crops, the disposal of toxic waste or the emission of greenhouse gas emission permits, public participation has demonstrated to be a useful mechanism to safeguard societal harms (Gilman, Reference Gilman2022, p. 535; Lee & Abbot, Reference Lee and Abbot2003, pp. 94–97; Voss, Reference Voss2014, pp. 32–35).
Environmental and Public-Sector AI decision-making procedures share many similarities. First, both are administrative decision-making procedures carried out by the EU or member states’ administrative bodies. Second, both deal with complex and specific matters that pose societal risks (Smuha, Reference Smuha2021, pp. 6–16) and require extensive technical knowledge, whether in environmental topics or AI (Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 235; Lee, Reference Lee2014, pp. 169, 193; Lee & Abbot, Reference Lee and Abbot2003, pp. 82–84; Sloane, Reference Sloane2024, p. 2). Third, as these procedures involve holding technical expertise, they are usually led or carried out by experts – environmental engineers in one case and computer scientists and developers in the other (Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, pp. 235–236; Lee, Reference Lee2014, p. 178; Lee & Abbot, Reference Lee and Abbot2003, pp. 83–84, 107). Fourth, because these administrative procedures are expert-based and rely on technical knowledge, the decisions they produce often lack democratic legitimacy (Calo & Citron, Reference Calo and Citron2021, p. 844; Grimmelikhuijsen & Meijer, Reference Grimmelikhuijsen and Meijer2022, p. 236; Lee, Reference Lee2014, pp. 177–178, 182–183; Lee & Abbot, Reference Lee and Abbot2003, pp. 84, 107), potentially leading to disastrous consequences for citizens. Together, these similarities make a strong case for applying the same participation standards of environmental decision-making to AI development decisions.
Crucially, environmental law allows public participation not just in administrative rulemaking but directly in decision-making procedures. When establishing public involvement in the approval of specific environmental plans, programs and projects, environmental law provides for participation in an ex ante stage of decision-making, which ultimately influences individual administrative decisions. In this context, EU environmental law offers not only a normative example but also a structurally comparable framework for embedding public participation in the administrative development processes of AI.
Nonetheless, what fundamentally links environmental law and Public-Sector AI development is their shared potential to generate societal harms. As previously mentioned, both the environmental harms and the injuries generated by AI use are particularly hazardous because they accumulate over time and cross state borders.Footnote 84 More than affecting individuals, detrimental AI and environmental outcomes have social costs. For example, the air pollution from a factory can harm the health of nearby residents and even further damage ecosystems and the atmosphere beyond its immediate surroundings, thereby generating social costs (Hirsch, Reference Hirsch, Strandburg and Raicu2006, p. 243). These social costs are negative externalities borne not by the company that benefits from the polluting activity but by society as a whole (Hirsch, Reference Hirsch, Strandburg and Raicu2006, p. 243). It was to address these negative externalities that environmental law adopted a societal rather than an individualist approach (Smuha, Reference Smuha2021, p. 13).
Likewise, as AI can cause societal harm, its regulation should also take a societal approach. The use of AI by public authorities can also generate “negative externalities.” While public authorities benefit from using AI tools in their decision-making processes, making them more efficient, this use may negatively impact the quality of administrative decisions and people’s trust in public decision-making, especially in the case of mass-disaster effects from AI use, thereby creating societal harm. For this reason, as in environmental law governance, the regulation of Public-Sector AI deployment should account for societal harm and include mechanisms to prevent it – notably by incorporating mechanisms of public participation.
6.2. Overview of public participation in environmental matters at the EU and national levels
Public participation in environmental matters is complex, taking distinct forms and engaging different actors at various stages of decision-making. The Aarhus Convention, the foundational international framework for public participation in environmental procedures at both the EU and national levels, structures participation in the environmental domain around “three pillars” (Lee, Reference Lee2014, p. 160), one of which is participation stricto sensu. Moreover, the Aarhus Convention specifically mandates public participation in three stages of decision-making: (i) authorization of specific environmental projectsFootnote 85; (ii) preparation of environmental plans, programs and policiesFootnote 86 and (iii) environmental rulemaking (Jendroska, Reference Jendroska2009, pp. 499–501; Lee, Reference Lee2014, p. 161; Lee & Abbot, Reference Lee and Abbot2003, p. 94).Footnote 87 Apart from participation in the elaboration of environmental regulation, these obligations are further implemented through EU legal instruments at different levels of decision-making.
At the EU institutions’ level, participation is limited in both material and personal range. As most environmental decision-making by EU institutions occurs at the macro-level planning, environmental participation is only foreseen in the preparation or amendments of legally environmental-related plans and programs (Lee, Reference Lee2014, pp. 190–191).Footnote 88 In addition, although formally open to every individual, association, organization or group,Footnote 89 participation is largely dominated by organized interest groups, with individual involvement remaining scarce and ineffective (Lee, Reference Lee2014, p. 192).
By contrast, public participation in environmental decision-making by EU member states is harmonized at a deeper level.Footnote 90 Member states must undertake environmental assessments when adopting environmental plans or programmesFootnote 91 or authorizing projects (Lee, Reference Lee2014, p. 165).Footnote 92 In drafting national environmental plans and programs, authorities with specific environmental responsibilities for their implementation, as well as the general public, must be consulted (Lee, Reference Lee2014, p. 172).Footnote 93 The public allowed to participate in that consultation must be identified by the public authority responsible for the decision-making, which, as a result, adds a burden on the competent authority (Jendroska, Reference Jendroska2009, pp. 514–515). Nonetheless, this positive obligation seems to expand rather than restrict participation (Jendroska, Reference Jendroska2009, pp. 514–515).
On the other hand, for specific projects, member states shall engage both the authorities with special environmental responsibilities connected with the project being assessedFootnote 94 and the public concerned.Footnote 95 In doing so, they shall provide information about the project to be developed (Lee, Reference Lee2014, p. 168). However, although the public at large has access to information related to the environmental project under analysis, when it comes to effective participation, that is reserved for the particular segment of individuals affected or likely to be affected or having an interest in the environmental project and to environmental interest groups, whose interest is presumed (Lee, Reference Lee2014, p. 168).Footnote 96
Despite these limitations, environmental decision-making offers real opportunities for public participation that could be adapted for Public-Sector AI development, especially given the similarities in decision types between the two domains. The parallels between environmental decision-making and Public-Sector AI development are particularly evident in the procedures for adopting environmental plans and programs. In fact, the several decisions entailed in the process of authorizing the use and developing AI tools for the public sector are prior steps to individual decision-making, where those technologies will be used in either a foundational or an ancillary role. Similarly, the decisions comprised in elaborating and adopting environmental plans and programs precede and significantly shape the individual authorizations of specific environmental projects. Accordingly, as both decisions occur in a preceding stage to individual administrative decision-making (Lee, Reference Lee2014, p. 176), one could assume that the participation avenues designed for elaborating environmental plans and programmesFootnote 97 would be the most tailored to AI development.Footnote 98
In fact, the participation opportunities established in the FRIA appear to draw inspiration from the procedures used to approve environmental plans and programs. Like those procedures, the FRIA is performed at a stage prior to individual decision-making and allows public authorities to involve representatives of groups likely to be affected by the intended use of an AI system, as well as experts and civil society organisations.Footnote 99 However, there are notable differences between participation in FRIAs and in the elaboration of environmental plans and programs. The key distinction is that the FRIA does not open participation to the general public, unlike the participatory mechanisms established for environmental planning.Footnote 100 As a result, this difference hinders FRIAs’ potential to democratize AI development processes in the public sector, as the avenues for public participation under the FRIA remain very narrow. For this reason, public involvement in the FRIA and, more generally, in Public-Sector AI development could be improved by adopting solutions from environmental planning procedures.
Nonetheless, as will be assessed, one should weigh both the benefits and the shortcomings of those participation solutions before replicating them.
6.3. Assessing the transposition of environmental participation to Public-Sector AI development
Given their common rationale and the absence of suitable participation avenues in traditional administrative law and the AIA, transposing participatory mechanisms from environmental law to Public-Sector AI development could prove advantageous, particularly in light of the recognized benefits of public involvement in environmental decision-making. One of the main benefits of public participation in environmental decision-making is enhancing the quality of environmental decisions. By allowing the public to express their views on environmental decisions and to voice their fears and values, it fosters public acceptance of those decisions (Lee, Reference Lee2014, pp. 178–179; Lee & Abbot, Reference Lee and Abbot2003, pp. 82–83). Furthermore, by providing environmental information, participation contributes to educating the population about the environment, while keeping a tight rein on the regulators and regulated entities through public scrutiny (Lee, Reference Lee2014, p. 179; Lee & Abbot, Reference Lee and Abbot2003, p. 83).
Another advantage of involving people in decision-making processes is improving the legitimacy of environmental decisions by adding a direct public endorsement and counter-balancing experts’ intervention in the process (Lee, Reference Lee2014, pp. 177–178; Lee & Abbot, Reference Lee and Abbot2003, pp. 83–84). Especially considering that those are complex and technical decisions, mostly made by experts within administrative bodies that lack direct democratic legitimacy, allowing the public to partake in those decision-making procedures adds the demos’ seal of approval to environmental decisions.
Yet, transposing an identical participation scheme to Public-Sector AI development would likely carry over the same limitations found in environmental decision-making (Lee, Reference Lee2014, pp. 173, 176). To begin with, engaging the lay public in the decision-making process of AI development would be challenging, as the matter is abstract and its potential impacts on people are not concrete.Footnote 101 Thus, if, as in environmental decision-making, EU and national authorities were required to identify the relevant public to participate, including the people affected or likely to be affected in the decision to implement AI in the public sector, this would add an even heavier burden on public entities, which is unlikely to be borne – especially since AI harms know no temporal and/or territorial barriers.Footnote 102
Additionally, even if engaged, ordinary individuals might have nothing to say about AI implementation. At the EU level, there is the extra challenge of involving people from diverse languages, cultures and backgrounds (Lee, Reference Lee2014, p. 193). This implies that, as in environmental decision-making, civil society organizations and interest groups would likely dominate participation, especially at the European level, where EU authorities would find it complicated to identify all individuals concerned in implementing AI systems.
In the latter context, even if civil society and interest groups’ involvement remains a possible solution, experience in the environmental field has something to teach us. It is questionable whether civil society organizations can gather public input and channel it to administrative institutions (Lee, Reference Lee2014, p. 193). Rather, there is a risk that these organizations will be captured by elites and economic powers (Lee, Reference Lee2014, pp. 195–196; Lee & Abbot, Reference Lee and Abbot2003, p. 87). Even if we address this issue, the big players still end up excluding ordinary individuals, notably those who are poorer, less educated and from minority groups (Lee, Reference Lee2014, p. 202; Lee & Abbot, Reference Lee and Abbot2003, p. 81). Therefore, despite the good intentions, public participation in AI development by public administrations might eventually face the same obstacle as in the environmental realm: it tends to exclude some more than it includes all.Footnote 103
Given the already acknowledged shortcomings of public participation in environmental decision-making, applying these avenues to Public-Sector AI development may cause more harm than good. Although transposing the participatory model for environmental plans and programs to AI uptake in the public sector would have the merit of expanding participation to the general public, experiences in environmental law show that it fails to attract the general public to highly technical procedures. Even worse, it may have the perverse effect of further alienating marginalized groups.
Notwithstanding these shortcomings, this transposition could have value in itself by alerting the population to public authorities’ plans to deploy AI across public-sector activities. This information is particularly important given the numerous AI deployment scandals in the public sector, in which the public was unaware of the algorithm’s use. However, there are less burdensome ways to introduce transparency in this deployment – such as implementing AI registers – than by putting in place a complex process to identify the relevant public to participate in the development of AI tools. Therefore, even though environmental law may be a useful source of inspiration for drawing future participatory mechanisms in Public-Sector AI development, I believe that, for now, the shortcomings of this proposed solution outweigh its potential benefits.
7. Conclusion
Public participation can play a significant role in shaping the public values guiding the design and development of AI systems for the public sector. Yet, if the latter is true, the existing legal avenues are manifestly insufficient to ensure individuals and organizations’ participation in AI development by the EU and national administrations. On the one hand, individuals and organizations will lack the legal means to take part in the overall authorization process of implementation of AI systems by public entities and to set the systems’ specifications, since participation in administrative decisions with a general scope is mainly excluded. Although natural and legal persons may have a right of access to the documents related to the development process of Public-Sector AI, they will be incapable of influencing that development and could only utilize that information after an individual administrative decision adversely affecting them is made. Furthermore, even though public procurement frameworks offer some indirect channels for public participation, they are not only restricted to the cases in which public authorities contract out AI development but are also conjectural. Therefore, in general, participation in administrative activity remains tilted to the level of individual decision-making, which places significant burdens on individuals exercising their right to be heard and makes their attempt to protect the public values in Public-Sector AI development too little, too late.
On the other hand, the avenues for stakeholder participation in the AIA mainly concentrate on the drafting of standards and codes of conduct or the governance system of the regulation. In addition, these avenues are not mandatory in the context of AI development by public entities, where broad participation by individuals is essential.
Furthermore, even if taking the example of environmental law, transposing its procedural solutions for public participation could result in harvesting more detrimental effects by implementing a formalistic and exclusive participation instead of improving the deployment of AI systems by public administrations and enhancing their accountability and legitimacy. Therefore, under the current legal framework, we find more dead ends than pathways supporting and implementing meaningful public participation in Public-Sector AI development. The key question that remains is what national and local public administrations in Europe – and worldwide – can still do to fix these shortcomings and open up more participatory pathways.
Acknowledgements
First drafts of this article were presented at the Young Digital Law Conference organized by SciencesPo – École de Droit in June 2024, and at the REALaw Young Researchers’ Forum organized by the Pázmány Péter Catholic University in September 2024. I am grateful to the participants and discussants for their feedback and helpful comments. I would also like to thank Deirdre Curtin, Filipe Brito Bastos and Marco Almada for their feedback and valuable suggestions on previous versions of this article. Any remaining errors are my own.
Funding statement
No funding was received for the conduct and writing of this article.
Competing interests
The author declares none.
Estela Lopes is a Ph.D. Researcher at the European University Institute, where her research explores how public administrations outsource the development of artificial intelligence (AI) to private actors from legal and empirical perspectives. She also serves as a Scientific Co-coordinator of the NOVA Platform for European Administrative and Regulatory Law, Lisbon, Portugal. She was recently a visiting researcher at the New York University School of Law.
Open access
