Regulate against the machine: how the EU mitigates AI harm to democracy

Cupać, Jelena; Sienknecht, Mitja Article — Published Version Regulate against the machine: how the EU mitigates AI harm to democracy Democratization Provided in Cooperation with: WZB Berlin Social Science Center Suggested Citation: Cupać, Jelena; Sienknecht, Mitja (2024) : Regulate against the machine: how the EU mitigates AI harm to democracy, Democratization, ISSN 1743-890X, Taylor & Francis, London, Vol. 31, Iss. 5, pp. 1067-1090, https://doi.org/10.1080/13510347.2024.2353706 This Version is available at: https://hdl.handle.net/10419/308004 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ RESEARCH ARTICLE Regulate against the machine: how the EU mitigates AI harm to democracy Jelena Cupać a and Mitja Sienknecht b a WZB Berlin Social Science Center, Berlin, Germany; b European University Viadrina Frankfurt, Frankfurt, Germany ABSTRACT Democracies are under attack from various sides. In recent years AI-powered techniques such as profiling, targeting, election manipulation, and massive disinformation campaigns via social bots and troll farms challenge the very foundations of democratic systems. Against this background, demands for regulating AI have gotten louder. In this paper, we focus on the European Union (EU) as the actor that has gone the furthest in terms of regulating AI. We therefore ask: What kind of instruments does the EU envision in their binding and nonbinding documents to prevent AI harm to democracy? And what critique can be formulated regarding these instruments? To address these questions, the article makes two contributions. First, by building on a systematic understanding of deliberative democracy, we introduce the distinction between two types of harm that can arise from the widespread use of AI: rights-based harm and systemic harm. Second, by analysing a number of EU documents, including the GDPR, the AI Act, the TTAP, and the DSA, we argue that the EU envisions four primary instruments for safeguarding democracy from the harmful use of AI: prohibition, transparency, risk management, and digital education. While these instruments provide a relatively high level of protection for rights-based AI harm, there is still ample space for these technologies to produce systemic harm to democracy. ARTICLE HISTORY Received 31 October 2022; Accepted 7 May 2024 KEYWORDS Artificial Intelligence (AI); democracy; harm; European Union (EU); AI Act Introduction Concerns about artificial intelligence (AI) negatively impacting democracy are not new. The Cambridge Analytica scandal has already revealed how practices such as psychographic profiling and targeted political advertising could be used for election manipulation. However, recent advancements in AI, notably the emergence of Large Language Models (LLMs) such as ChatGPT, along with models capable of generating images, audio, and video content, have escalated these concerns to a new level. As these technologies continue to spread and capture public attention, we are heading toward © 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons. org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent. CONTACT Jelena Cupać[email protected] DEMOCRATIZATION 2024, VOL. 31, NO. 5, 1067–1090 https://doi.org/10.1080/13510347.2024.2353706 the first wave of elections in which they will be widely deployed. It is expected that quite a few politicians will use synthetic content in their campaigns, whether to launch attacks against their opponents through manipulated videos and imagery or to reduce campaign costs. Against this backdrop, we focus on the European Union (EU) and ask what measures it envisions to mitigate the harm AI poses to democratic process. The EU’s relationship with democracy is not perfect. Since its founding, it has been accused of structural democratic deficit and, more recently, it has had to grapple with a populist wave causing democratic backsliding in several of its member states. Still, there is no denying that the EU is inextricably linked to democracy. Along with human dignity, freedom, equality, the rule of law, and human rights, the Lisbon Treaty lists democracy as one of the EU’s core values. An important implication of this value orientation is that when the EU ventures into a new regulatory territory, it is expected to show concern for democracy and devise mechanisms for its protection, especially if there is ample evidence that a given domain might suffer significant democratic erosion. As the global race towards AI intensifies, we are witnessing a parallel race towards AI regulation, with the EU emerging as a clear frontrunner. 1 Although lagging behind the United States and China in AI development, the EU is actively striving to establish itself as a global standard-setter in digital technologies. It has already achieved this goal through its flagship General Data Protection Regulation (GDPR), with various countries outside Europe adopting similar data protection rules. The EU now seeks to replicate this impact with the Digital Services Act (DSA), aimed at bolstering online safety by tackling harmful content, and more significantly, with the Artificial Intelligence Act (AIA), a comprehensive regulatory framework for AI grounded in a risk-based approach. 2 Despite these documents’obligation to address harm to democracy, a systematic overview of the measures they establish for this purpose is still missing. As a result, we have a limited understanding of the set of actions the EU and its member states have at their disposal to uphold democratic processes in the AI age. The rapid progress of AI technologies underscores the importance of such understanding, not least to identify which areas of democracy are adequately protected and which require further safeguarding. To address this gap, we start by identifying the types of harm a widespread use of AI can cause to democracy. Based on a systematic understanding of deliberative democracy, we distinguish between rights-based harm and systemic harm. The former pertains to using technology to limit people’s participation in the democratic process, while the latter refers to broader societal and political factors that impede democratic deliberation, such as fragmentation, polarization, distrust, and political apathy. In making this distinction, we join a growing number of scholars who call for a broader approach to AI harm, one that goes beyond individual and human rights concerns and pays attention to the wider societal impact. 3 Based on this framework, we have selected four legally binding EU documents for our analysis: the GDPR, DSA, AIA, and the Proposal for a Regulation on the Transparency and Targeting of Political Advertising (TTPA), alongside a range of nonbinding documents (see Appendix, Table 1). Through close examination of these documents, we have identified four key instruments the EU proposes to protect democracy from AI harm: prohibition, transparency, risk management, and digital 1068 J. CUPAĆAND M. SIENKNECHT education. Prohibition entails the outright banning of specific data and AI practices; transparency mandates the disclosure of information concerning the development and deployment of AI systems; risk management involves assessing the risk of an AI system before deployment and monitoring the risk after the system is in use; and digital education aims to raise public awareness of AI risks and empower individuals to use digital technologies responsibly. Although these instruments afford a relatively robust level of protection against rights-based harm, our core observation is that the widespread use of AI still has significant potential to inflict systemic harm to democracy in the EU. Before proceeding, it is important to note this study’s limitations. Our primary goal was to categorize AI harm to democracy and take a bird’s eye view of the EU’s regulatory and policy responses. Ideally, we would have matched each type of harm with specific protective measures and provided a detailed evaluation. However, the intertwined nature of rights-based and systemic harm, which is also mirrored in regulatory and policy instruments, makes such an approach challenging. Thus, we have opted for a broader approach, which can serve as a guide for future in-depth analysis. The article proceeds in three steps. First, we define democracy and AI, and discuss rights-based and systemic harm in more detail. Second, we outline our analytical steps: selecting EU documents and identifying instruments to shield against AI harm. Third, we present and critically assess these instruments. In conclusion, apart from summarizing our findings, we delve into the challenges and possibilities of regulating against AI harm, thereby underscoring a significant avenue for future research. Artificial intelligence and democracy: definitions To explore the potential negative impact of AI on democracy, we must first define both concepts. However, defining AI and democracy is notoriously difficult. AI can refer to a broad range of constantly evolving technologies and applications that mimic human cognition, such as learning, problem-solving, and decision-making. Likewise, democracy remains elusive given the diverse ways democratic principles manifest across various societies and cultures. Recognizing these challenges, we have opted for a pragmatic approach, seeking definitions that neither impede thorough analysis nor restrict broader insights into the EU’s AI regulation for democracy protection. Defining artificial intelligence and its application in politics AI can be defined as “Systems that display intelligent behaviour by analysing their environment and taking actions –with some degree of autonomy –to achieve specific goals.” 4 AI systems can make autonomous decisions in various fields of human activity and generate original content such as text, pictures, and videos. Unlike conventional digital technologies based on fixed algorithmic patterns, AI systems rely on neural networks –machine learning algorithms modelled after the human brain’s structure and function –to learn, adapt, and evolve through exposure to large amounts of data. In this way, AI systems can analyse complex patterns, make informed decisions, and create content that exceeds the limits of their initial programming. DEMOCRATIZATION 1069 Currently, numerous AI technologies and their applications are seen as potential threats to democracies. Some are entirely AI-based, while others rely on traditional digital methods. Yet, when augmented by AI, these conventional methods can be significantly enhanced, heightening their potential risk. Broadly speaking, these technologies can be split into two categories along their intended function: those centred on surveillance, such as data collection and profiling, and those geared towards manipulation, encompassing targeting, social bots, and deep fakes. 5 Profiling consists of collecting and analysing data about people in order to classify them based on specific characteristics. 6 In the electoral process, demographic profiling can be used to categorize voters by attributes like age and education, whereas psychometric profiling probes deeper into personality traits, such as being argumentative or compliant. Yet, profiling is not solely informational; it typically supports targeting,a technique intended to influence people’s attitudes and political stances. 7 While micro-targeting relies on this data to identify a specific audience segment, hyper-targeting takes it a step further by relying on even more detailed information such as location tracking and social media activity. Social bots are automated programmes that emulate human actions on social media. 8 Augmented by AI models such as GPT and Midjourney, they could generate and spread vast amounts of text, images, videos, and links. With the ability to amplify particular narratives, sway discussions, and serve as a vehicle for spreading disinformation, social bots pose a significant threat to the integrity of online information. The same applies to deep fakes, multimedia counterfeits that swap one person’s image or video likeness for another using advanced algorithms that analyse their facial cues. 9 The result is a fake yet highly realistic piece of media. It is crucial to highlight that we are also entering an era of rapid development in AIpowered neurotechnology, biometrics, and subliminal messaging, all expected to further exacerbate issues related to profiling and manipulation. 10 Technologies ranging from real-time biometric identification systems to subliminal messaging such as AI-driven “dark patterns”are all poised to amplify the accuracy and subtlety of profiling and manipulation, generating a host of new and often unforeseeable challenges. 11 Lastly, although not yet a reality, there is significant discussion about the development of general-purpose AI –systems capable of learning and executing a broad spectrum of tasks instead of being confined to specific roles or areas. How such AI will be integrated into political systems is difficult to predict, but its impact on disrupting traditional democratic processes and citizen engagement will be profound, so a careful consideration of its implications for governance, democracy, ethics, and society at large will be paramount in shaping its role and impact on political processes. Definition of democracy: a systemic deliberative approach To define democracy, we take a cue from the work of Spencer McKay and Chris Tenove, who settled on the concept of systemic deliberative democracy in their analysis of how disinformation affects the democratic process. 12 The advantage of this approach to democracy is that rather than focusing on the intricacies of multiple democratic arenas and processes, it emphasizes deliberative functions that unite these arenas and processes into a systemic whole. As a result, a key feature of this approach to democracy is identifying normative goods and functions that sustain 1070 J. CUPAĆAND M. SIENKNECHT meaningful deliberation across the system and which, if jeopardized, can threaten democracy in its entirety. Different authors emphasize different normative goods and functions essential for genuine democratic deliberation. 13 In this study, we draw on the work of Jane Mansbridge and her colleagues, whose approach seems particularly pertinent to the information age. They identify three functions of deliberative democracy: democratic, epistemic, and ethical. 14 The democratic function focuses on the equal participation of citizens, stressing the active inclusion of diverse perspectives and the prevention of unjust exclusion. The epistemic function pertains to the quality of the information environment in which deliberation takes place, especially the quality of the information and logics influencing preferences and opinions. Finally, the ethical function emphasizes mutual respect in political discussions, with the emphasis on participants recognizing each other’s viewpoints as valid and deserving of consideration. These three functions suggest that the bedrock of a thriving democracy lies in upholding participation rights, whether for individualsor groups, and in the systemic conditions guaranteeing this participation. The systemic conditions derive from the integrity of the information ecosystem and the quality of interactions among citizens. Against this background, we assume that AI can undermine democracy at its core both by curtailing citizens’participation rights and by introducing disruptions at the systemic level. In other words, we differentiate between two types of AI induced harm to democracy: rightsbased harm and systemic harm. In so doing, we align with the burgeoning literature seeking to map, systematize, and taxonomize AI harm across various domains. 15 At the same time, we align ourselves with a growing number of sociologists, legal scholars, and policy experts who, drawing on foundational insights from Zemiology and Science and Technology Studies, argue that social harm, whether stemming from AI or other sources, should not be viewed in predominantly individualized and human rights terms. 16 Instead, special attention should also be paid to societal harm, including harm to systems and structures upholding societies. 17 We contribute to this strand of literature by distinguishing between rights-based and systemic harm within the specific context of democracy, instead of addressing it through the lens of a society as a whole. 18 This is an analytical move that we hope provides a framework from which AI harm to democracy can be studied in a broad yet more focused manner. AI harm to democracy: two types Rights-based harm to democracy In the context of this article, rights-based harm to democracy refers to using AI technologies to violate individual, collective, and group rights essential to ensuring the meaningful participation of diverse voices and perspectives in a democratic society. 19 Individual rights and freedoms, such as privacy, freedom from discrimination, and freedom of thought and expression, are widely recognized as essential for democratic participation. Collective rights belong to groups based on shared identity, such as language or culture, and are vital for protecting minority rights and ensuring their representation. Finally, group rights apply to any group of individuals and encompass such rights as forming associations and organizations and engaging in collective bargaining and advocacy. DEMOCRATIZATION 1071 Based on the extant literature and real-world cases, we can discern several scenarios of digital technologies powered by AI causing significant harm to these rights. One such instance is the potential harm to privacy resulting from AI systems collecting large volumes of personal data through monitoring online behaviour and social media activity without consent. 20 This data can then be used to profile and target individuals, modifying theirinformation environment without their awareness and influencing how and whether they engage in the democratic process. Furthermore, combining generative AI with AIpowered social bots could enable malicious actors to flood social media with an endless stream of messages that disparage and reinforce stereotypes about particular minority groups, corroding the moral respect they need to be recognized as equal democratic interlocutors. 21 Such information operations can spill over into the real world, contributing to these groups being further marginalized and even excluded from politics. Similar things can happen to groups that organize around political interests rather than strictly their identities, such as political parties and social movements. In addition to large quantities of disparaging information, their political participation can also be jeopardized by the compromising fabrications created by deep fakes. Political adversaries could use deep fakes to depict each other engaging in disreputable behaviour, making controversial statements, or participating in criminal activities while portraying themselves in a favourable light. Finally, groups and collectives may find their voices drowned out not because of the nature of AI-generated content but its massive quantity on social media and other online platforms. 22 Systemic harm to democracy The systemic harm that the widespread use of AI technologies can inflict on democracies refers to the impairment of general societal, informational, and political conditions in which democratic participation and deliberation unfold. Such harm can come in the form of societal and political polarization, fragmentation, pervasive distrust, and widespread political apathy and indifference. These distortions need not be produced by AI. They can arise due to a variety of political, cultural, societal, and economic reasons and, as we know, from digital technologies that are not augmented by AI. However, AI-based technologies, and especially generative AI, threaten to exacerbate them to historically unprecedented levels. As already indicated, rights-based harm and systemic harm to democracy are not strictly separable. The reason for this lies in the fact that the infringement of individual, collective, and group rights constitutes one means by which AI can inflict systemic harm on democracy. To illustrate, considerthe case of pervasive AI propaganda campaigns targeting a minority group on social media platforms. Such campaigns can simultaneously erodethegroup’s collectiverightsand fostera climate ofintensifiedsocietalradicalization and polarization, thus developing from group rights-harm to systemic harm. However, the distinction is nonetheless useful, considering that AI can undermine democracy at a systemic level even in the absence of explicit violations of individual, group, and collective rights. Strategies for achieving this can include spreading a plethora of conspiracy theories or amplifying extremist views. Such campaigns can lead to the formation of isolated, like-minded groups that are increasingly disconnected from the broader society, resulting in societal fragmentation. At the same time, the proliferation of conspiracy theories can deepen the divide between those who subscribe to them and those who reject them, thereby intensifying polarization. This widening rift 1072 J. CUPAĆAND M. SIENKNECHT can make it more challenging for individuals and groups to engage in constructive dialogue and find common ground, ultimately undermining the democratic process. Multiple scenarios can also be imagined in which the widespread use of AI can lead to distrust and political apathy. For instance, they may inadvertently prioritize content that triggers strong emotional reactions, such as outrage, fear, or disgust, leading to a more negative and cynical perception of politics and politicians. Using AI in mass surveillance may result in perceived or actual privacy loss, making people more cautious about expressing their opinions and engaging in political activities. Lastly, AI-produced disinformation and misinformation can erode trust in traditional knowledge sources, such as universities, leading citizens to regard all truth claims with scepticism, perceiving them as politically motivated. Before we proceed, it should be emphasized that the literature recognizes the challenges of addressing and regulating systemic harm. A primary concern is the issue of temporality. On the one hand, there are systemic harms that remain unanticipated and thus elude regulation. On the other hand, the systemic harm we are familiar with does not arise from a singular use of AI but evolves over time from multiple deployments by various actors and in combination with other social factors. Consequently, this type of harm is accumulative, making it challenging to identify a direct cause or determine the responsible party. 23 The EU’s response to AI harm to democracy: a critical survey Methodology How does the only supranational organization of its kind comprising 27 democracies, the EU, respond to the outlined dangers posed by AI technologies to democracy? What measures is the EU taking to meet the challenge of regulating a constantly evolving technology? In this section, we survey the EU’s policies and regulations aimed at mitigating rights-based and systemic harm the widespread use of AI may cause to democracy. Since the mid-1990s, the EU has regularly introduced policies, action plans, codes of conduct, and regulations to support member states’digital transition. To select documents relevant to our analysis, we first compiled a detailed list of the EU’s digital regulations and policies by using the EU’s website and EUR-Lex search function (see Appendix, Table 1). The resulting list is not exhaustive. When it comes to nonlegal instruments, it only features documents with a direct or indirect connection to AI, rather than covering digitalization more broadly. To select relevant documents for our analysis, in line with our definition of democracy, we identified documents concerned with rights to political participation and those addressing the integrity of the information environment. From the shortlisted documents we then selected for our analysis those addressing AI technologies and applications, specifically data collection, processing, profiling, targeting, social bots, deep-fakes, and disinformation campaigns, as well as other technologies that could broadly be defined as AI. This selection process yielded four legally binding documents: The General Data Protection Regulation (GDPR), the Digital Services Act (DSA), the Artificial Intelligence Act (AIA), and the Proposal for a Regulation on the Transparency and Targeting of Political Advertising (TTPA). The GDPR is a privacy law that protects EU residents’ data by restricting its unconsented collection and processing, including for automated DEMOCRATIZATION 1073 decision-making. The DSA is a legislative framework for digital services, aiming to enhance online safety by combating disinformation and harmful content. The EU AI Act is the first legislation aimed at directly regulating the development and deployment of AI systems within the common market. Finally, the TTPA is a proposed law designed to harmonize political campaigning across the continent, including the use of AI technologies such as profiling and targeting. The European Media Freedom Act (EMFA) and the Digital Markets Act (DMA) are examples of the documents we excluded from our analysis, although they may appear relevant at first glance. While the EMFA emphasizes democratic participation, it does not directly address AI. Conversely, the DMA mentions AI but does not focus on democracy as we define it here in this article. As for non-binding legal and nonlegal documents, we considered documents in our analysis that fulfilled the same criteria. Thus, we included declarations, communications, white papers, policies, and codes of practices that deal with threats to democracies and the influence of AI. For a detailed list, see the bolded text in the table located in the Appendix. To identify the specific EU regulatory and policy measures intended to protect the continent’s democracy from AI risks, we started by inductively analysing the four legally binding documents. By grouping the articles according to their similarity, we can show that the measures the EU is currently taking to protect democracy from AI-related harm fall into four broad categories: (1) prohibition, (2) transparency (subcategories: transparency to individuals, operational transparency, reporting transparency); (3) risk management; and (4) education (see Appendix, Table 2). In the course of the analysis, two more categories were considered: socio-technical design of AI systems and the governance of the digital sphere. However, we decided to omit them, considering them as already contained within our four primary categories. While specific articles in the AIA, like Article 15, explicitly require developers to adhere to certain design principles, we contend that the bulk of the articles within the prohibition, transparency, and risk management categories also guide the design of AI systems. These articles, by dictating what AI systems should or should not do to avoid causing harm, or asking developers to disclose features of design, essentially serve as directives for how these systems should be made. For this reason, we have opted not to treat socio-technical design as a separate category, although we acknowledge that some articles might be more explicit about it. Similarly, given that governance plays a crucial role in enforcing any EU regulation, we see it as unnecessary to treat it as a distinct category within the context of digital sphere governance. In the next section, we discuss each measure, along with their strengths and weaknesses in reducing rights-based and systemic AI harms to democracy. In doing so, we leverage insights from recent research in prohibition, transparency, risk, and education within the digitalization field. EU instruments against AI harm to democracy Prohibition The EU’s most far-reaching instrument to mitigate the harmful effects of AI, including harm to democracy, is the prohibition of specific AI systems and practices. Each of the four documents examined –the GDPR, DSA, TTPA, and the AIA –includes such prohibitions. The GDPR bans processing biometric data and data relating to ethnic 1074 J. CUPAĆAND M. SIENKNECHT While the EU’s initiative to educate its citizens on safeguarding their various rights in the face of an increasingly AI-driven democracy and averting systemic harm is commendable, expecting individuals to manage their private information and discern false information consistently and collectively may be overly optimistic. To begin with, access to digital education is not equally available. Most EU campaigns target young people in educational settings, leaving adults and non-digital natives, who are no longer part of the educational system and might need AI curricula tailored to their cognitive level, somewhat overlooked. 59 Moreover, the EU has not sufficiently addressed a crucial point highlighted in the literature: training AI developers on AI harm to ensure the creation of systems that adhere to the standards of fairness, accountability, transparency and ethics, and democracy in general. 60 Still, even with citizens and developers being highly educated about AI risks and the measures available to mitigate them, protecting democracy at both rights-based and systemic levels is not guaranteed. While digital literacy is necessary, it is not sufficient. Even experienced social media moderators find identifying disinformation campaigns and information propagated by social bots challenging. This challenge grows exponentially with large language models capable of flooding the system with misleading content. Moreover, even digitally literate individuals might become distrustful and apathetic if they feel overwhelmed by the constant need to differentiate between genuine information and disinformation and handling their private data. Ironically, this could contribute to systemic harm to democracy rather than protecting against it. Conclusion In this study, we mapped the diverse measures the EU has implemented or plans to implement to protect democracy from the growing harm associated with AI technologies. The mapping was guided by the distinction we made between rights-based and systemic AI harm to democracy, resulting in four categories of measures: prohibition, transparency, risk management, and education. As a mapping exercise, our analysis trades an in-depth examination of the EU regulation for a bird’s eye perspective on the regulatory landscape. Yet, this approach enables us to infer several insights that could prompt a deeper analysis of these measures in the future. We note that when we expand our focus beyond the AIA to other binding and nonbinding EU documents that address to some degree the harm of AI to democracy, we see that the EU has developed a quite robust toolbox of protective measures, catering to both rights-based and systemic harm. Especially in the realm of systemic harm, which existing literature identifies as inadequately addressed, we map a variety of measures that can be seen as efforts to institute continuous oversight of AI systems’development and deployment. 61 However, despite the inclusion of these provisions, AI still has significant potential to produce systemic harm to democracy. Based on this, we propose that the challenge with AI systemic harm in the EU’s AI regulation is not just its insufficient coverage, as highlighted by existing research, but also that such harm is inherently difficult, perhaps even impossible, to effectively regulate. First, as underscored throughout the article, highly protected rights and an AI-literate public can, paradoxically, engender systemic harm to democracy. Many people consenting to being profiled and targeted, even with non-deceptive content, can DEMOCRATIZATION 1081 foster the creation of echo chambers, which, in turn, may lead to political polarization and fragmentation. Moreover, a public possessing a high level of knowledge about AI practices may grow disillusioned with its integration into the democratic process, consequently becoming more distrustful and politically apathetic. Second, the aggregated nature of systemic harm, coupled with the fact that it arises not solely from AI but from its intertwining with pre-existing social conditions, makes pinpointing its exact causes challenging. However, without such precision, it is difficult to establish parties accountable for systemic harm or a threshold at which measures targeting only its AI component would be effective. For example, while social media companies contribute to increasing political polarization, it is unclear whether eliminating misand disinformation from these platforms would alleviate this issue or, importantly, what amount would need to be removed before we see a positive shift. Against this backdrop, we propose two avenues for further research: one analytical and the other normative. Analytically, there is a need for a thorough examination of the EU’s regulatory instruments to mitigate the harms AI can cause to democracy at a fundamental level. This examination should be mindful of both rights-based and systemic AI harm as well as of the broader regulatorylandscapeweoutlined.Scrutiny of this type could act as a catalyst for regulatory improvements and policy recommendations, which are urgently needed given AI’s rapid advancement and widespread adoption. Normatively, it may be time to acknowledge that regulation and policy can go only so far and that we need to start thinking about a novel paradigm of democracy in which AI would not be only a problem to be managed but its integral component. 62 Notes 1. Smuha, “From a ‘Race to AI’to a ‘Race to AI Regulation’.” 2. After adopting the Artificial Intelligence Act (AIA) on 13th March 2024, the European Parliament issued corrections on 16th April 2024. This corrected version is cited in this paper and represents the final draft of the AI Act, which is awaiting final approval by the Council of the European Union at the time of writing this paper. 3. See: Hildebrandt, “Algorithmic Regulation and the Rule of Law”; Kolt, “Algorithmic Black Swans”; Smuha, “Beyond a Human Rights-Based Approach to AI Governance”; Smuha, “Beyond the Individual”; Uuk, “Manipulation and the AI Act”; van der Sloot and van Schendel, “Procedural Law for the Data-Driven Society”; Yeung, “AI Governance by Human Rights-Centered Design.” 4. European Commission’s High-Level Expert Group on Artificial Intelligence, “ADefinition of AI”,1. 5. Cf. Kaplan, “Artificial Intelligence, Social Media, and Fake News,”153. 6. See: Brkan, “Artificial Intelligence and Democracy”; Djeffal, “AI, Democracy and the Law”; Eubanks, Automating Inequality; Hinsch, “Differences That Make a Difference”; Kertysova, “Artificial Intelligence and Disinformation”; McSweeny, “Psychographics, Predictive Analytics, Artificial Intelligence & Bots”;O’Neil, Weapons of Math Destruction; von Ungern-Sternberg, “Discriminatory AI and the Law.” 7. See: Brkan, “Artificial Intelligence and Democracy”; Djeffal, “AI, Democracy, and the Law”; Ienca, “On Artificial Intelligence and Manipulation”; Kaplan, “Artificial Intelligence, Social Media, and Fake News”; Kertysova, “Artificial Intelligence and Disinformation”; König and Wenzelburger, “Opportunity for Renewal or Disruptive Force?”; Milan and Agosti, “Personalisation Algorithms and Elections”; Mogaji et al.“Using AI to Personalise Emotionally Appealing Advertisement”; Narayanan, Understanding Social Media Recommendation Algorithms; Zachary, “Digital Manipulation.” 1082 J. CUPAĆAND M. SIENKNECHT 8. See: Brkan, “Artificial Intelligence and Democracy”;Diakopoulos,“Automating the News”;Ferrara et al., “TheRiseofSocialBots”; García-Orosa et al., “Algorithms and Communication”; Keller et al. “Social Bots in Election Campaigns”;Shaoetal.,“The Spread of Low-Credibility Content.” 9. See: Habgood-Coote, “Deepfakes and the Epistemic Apocalypse”; Hameleers et al., “You Won’t Believe What They Just Said!”; Jacobsen and Simpson, “The Tensions of Deepfakes”; Westerlund, “The Emergence of Deepfake Technology”; Whyte, “Deepfake News.” 10. For an overview see: Farahany, The Battle for Your Brain; Neuwirth, “Prohibited Artificial Intelligence Practices”; Neuwirth, The EU Artificial Intelligence Act. 11. See: Neuwirth, The EU Artificial Intelligence Act, 25–26 and 94. 12. McKay and Tenove, “Disinformation as a Threat to Deliberative Democracy”; Tenove, “Protecting Democracy from Disinformation.” 13. For an overview, see: Warren, “A Problem-Based Approach to Democratic Theory.” 14. Mansbridge et al., “A Systemic Approach to Deliberative Democracy”, 11-12. 15. For an overview, see: Shelby, “Sociotechnical Harms of Algorithmic Systems.” 16. Smuha, “Beyond a Human Rights-Based Approach to AI Governance”; Smuha, “Beyond the Individual”;Yeung, “AI Governance by Human Rights–Centered Design.” 17. Hildebrandt, “Algorithmic Regulation and the Rule of Law”; Kolt, “Algorithmic Black Swans”; Smuha, “Beyond the Individual”; Uuk, “Manipulation and the AI Act”; van der Sloot and van Schendel, “Procedural Law for the Data-Driven Society.” 18. Cf. Jungherr “Artificial Intelligence and Democracy”; König and Wenzelburger, “Opportunity for Renewal or Disruptive Force.” 19. In “Beyond the Individual,”Smuha differentiates between individual, collective, and societal harm AI can cause. Conversely, we adopt a perspective grounded in the systemic deliberative definition of democracy, which emphasizes participation rights for both individuals and groups, along with the systemic conditions necessary for exercising these rights. Consequently, we offer a binary categorization of AI harm: rights-based and systemic harm. While there are significant overlaps, our classification diverges from Smuha’s by being anchored in a specific understanding of democracy rather than a broad societal viewpoint. 20. See for a discussion on the group right to privacy in the context of biomedical data, Floridi 2014. 21. Cf., DiResta et al., “The Tactics & Tropes.” 22. Cf., Woolley and Guilbeault, “United States.” 23. See: Smuha, “Beyond the Individual,”10; Kernohan, Andrew. “Accumulative Harms.”; Kolt, “Algorithmic Black Swans.” 24. GDPR, Regulation (EU) 2016/679, Article 9. 25. DSA, Regulation (EU) 2022/2065, Article 26 and 28. DSA also prohibits designing online interfaces in a manipulative way, Article 25. 26. European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Article 18. 27. Some of these prohibitions come with exceptions, for details see: European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Article 5. For detailed analysis and criticism of these practices see: Neuwirth, “Prohibited Artificial Intelligence Practices,”; Neuwirth, The EU Artificial Intelligence Act. 28. Cf. Smuha, “Beyond the Individual.” 29. Neuwirth, “Prohibited Artificial Intelligence Practices,”4. 30. Jobin et al., “The Global Landscape of AI Ethics Guidelines.”See also: Mike and Crawford. “Seeing Without Knowing”; Gorwa and Ash, “Democratic Transparency in the Platform Society”; Larsson, Stefan, and Fredrik Heintz, “Transparency in Artificial Intelligence.” 31. Albert Meijer as quoted in Diakopoulos, “Transparency,”198. 32. Heald, “Varieties of Transparency.” 33. GDPR, Regulation (EU) 2016/679, Article 12 and 13. Transparency measures relating to the handling of private data can also be found in GDPR’s Articles 5, 6, 7, 8, 11, 14, 15, 16, 17, 18, 19, 20, 21, 22, and 34. 34. DSA, Regulation (EU) 2022/2065, Articles 14, 17, 20, 26, 27, 28, 38, and 39; European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Articles 11, 12, 18, and 19; European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 13 and 50. It is important DEMOCRATIZATION 1083 to highlight that Article 50 extends to AI under a free and open-source licence. Some critics view this as an insufficient measure for this type of AI, which is often regarded as particularly risky for its potential to harm democracy, including via the widespread dissemination of disinformation. See, e.g.: Giannaccini and Kleineidam, How the EU’s Soft Touch on Open-Source AI Opens the Door to Disinformation. 35. See: Andrada et al., “Varieties of Transparency”; Diakopoulos, “Transparency”; Walmsley, “Artificial Intelligence and the Value of Transparency.” 36. GDPR, Regulation (EU) 2016/679, Article 25, 40, 41, and 42.; DSA, Regulation (EU) 2022/2065, Articles 44 to 77; European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA (2024)0138, Articles 10, 11, 12, 13, 15, 17, 18, 44, 47, 48, 49 and 77. 37. European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 13 and 53. 38. GDPR, Regulation (EU) 2016/679, Article 30 and 32; DSA, Regulation (EU) 2022/2065, Articles 37 and 40; European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Articles 9 and 19; European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 12, 19, 53, and 55. 39. DSA, Regulation (EU) 2022/2065, Articles 10, 15, 24, and 42. 40. European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Articles 14, 16, 17, and 20. 41. European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 20, 27, 52, 77, and 91. 42. Ananny and Crawford, “Seeing Without Knowing.” 43. GDPR, Regulation (EU) 2016/679, Articles 35 and 36; DSA, Regulation (EU) 2022/2065, Article 34; European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Article 19. 44. European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 27 and 43 and Annex III(8). 45. European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 57 and 60. 46. GDPR, Regulation (EU) 2016/679, 33; DSA, Regulation (EU) 2022/2065, Articles 16, 18, 22, and 23; European Parliament and the Council of the European Union, Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, Article 15; European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Article 9, 14, 55, 72, 73, 74, 75, 76, 89, and 90. 47. Remedial and mitigation articles can be either independent or a part of some other article. For example, see: DSA, Regulation (EU) 2022/2065, Articles 9, 16, 18, 22, 23, 35, 36, and 48; European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Articles 15, 20, 55, and 57. 48. The GDPR establishes the Data Protection Officer, Lead Supervisory Authority, European Data Protection Board, and National Data Protection Authorities; the DSA establishes the Digital Services Coordinator, European Board for Digital Services; and the AIA establishes the AI Office, European Artificial Intelligence Board, Advisory Forum, Scientific Panel of Independent Experts, Notifying Authority, Market Surveillance Authority, and Conformity Assessment Body. 49. European Commission, Commission is Gathering Views on Draft DSA Guidelines for Election Integrity. 50. Austin, “The DSA now Applies in Full.” 51. Veale and Borgesius, “Demystifying the Draft Eu Artificial Intelligence Act,”105. 52. Edwards, Regulating AI in Europe. 53. European Commission, COM (2020) 790 final, 3. 54. See: https://ec.europa.eu/eurostat/cache/metadata/en/isoc_sk_dskl_i21_esmsip2.htm 55. European Commission, Standard Eurobarometer 98, QF8.1, T150. 56. On trustworthiness of sources, see: European Commission, COM (2020) 790, 23. On sharing the information, see: European Declaration on Digital Rights and Principles for the Digital Decade (2023/C 23/01); The Strengthened Code of Practice on Disinformation (2022 COM (2020) 624 final); Tackling Online Disinformation: A European Approach (COM (2018) 236 1084 J. CUPAĆAND M. SIENKNECHT final); White Paper on Artificial Intelligence: A European approach to excellence and trust (COM (2020) 65 final). 57. European Parliament, Corrigendum, Artificial Intelligence Act, P9_TA(2024)0138, Article 3 (56). 58. European Commission, COM (2020) 790 final, 3. 59. Chan, “A Comprehensive AI Policy Education Framework”; Kaplan, “Artificial Intelligence, Social Media, and Fake News”; Laupichler et al., “Artificial Intelligence Literacy in Higher and Adult Education”; Lee, “Fake News, Phishing, and Fraud”; Yang, “Artificial Intelligence Education for Young Children”. 60. Bogina et al. “Educating Software and AI Stakeholders”; Borenstein and Howard, “Emerging Challenges in AI”; Schiff,“Education for AI, not AI for Education.” 61. Cf. Smuha, “Beyond the Individual,”16-23. 62. See: Susskind, On Freedom and Democracy; Coeckelberghm, Why AI Undermines Democracy. Acknowledgment The authors wish to express their gratitude to Yushu Soon for her assistance and patience during the literature review for this article, and to the reviewers for their insightful and constructive comments. Disclosure statement No potential conflict of interest was reported by the author(s). Notes on contributors Jelena Cupaćis a Post-Doctoral Research Fellow at the WZB Berlin Social Science Center. She holds a PhD from the European University Institute (EUI) in Florence. Her research explores the transformation of international organizations, the global anti-gender movement, and efforts to govern AI’s impact on democracy. Mitja Sienknecht is a postdoctoral researcher at the Europen New School of Digital Studies/ European University Viadrina. Her current research focuses on the transformation of war through AI technologies, the general impact of AI systems on democratic societies, and related ethical questions. ORCID Jelena Cupaćhttp://orcid.org/0000-0002-7471-7624 Mitja Sienknehct http://orcid.org/0000-0001-5217-1432 Bibliography Ananny, Mike, and Kate Crawford. “Seeing Without Knowing: Limitations of the Transparency Ideal and its Application to Algorithmic Accountability.”New Media & Society 20, no. 3 (2018): 973–989. Andrada, Gloria, Robert W. Clowes, and Paul R. Smart. “Varieties of Transparency: Exploring Agency within AI Systems.”AI & Society 38, no. 4 (2023): 1321–1331. Bogina, Veronika, Alan Hartman, Tsvi Kuflik, and Avital Shulner-Tal. “Educating Software and AI Stakeholders about Algorithmic Fairness, Accountability, Transparency and Ethics.” International Journal of Artificial Intelligence in Education 32, no. 3 (2021): 808–833. Borenstein, Jason, and Ayanna Howard. “Emerging Challenges in AI and the Need for AI Ethics Education.”AI and Ethics 1, no. 1 (2021): 61–65. Brkan, Maja. “Artificial Intelligence and Democracy: The Impact of Disinformation,: Social Bots and Political Targeting.”Delphi 1, no. 2 (2019): 66–71. Chadha, Anupama, Vaibhav Kumar, Sonu Kashyap, and Mayank Gupta. “Deepfake: An Overview.”In Proceedings of Second International Conference on Computing, Communications, and CyberDEMOCRATIZATION 1085 Security, edited by Pradeep Kumar Singh, Sławomir T. Wierzchoń, Sudeep Tanwar, Maria Ganzha, and Joel J. P. C. Rodrigues, 557–566. Singapore: Springer Singapore, 2021. Chan, Cecilia Ka Yuk. “A Comprehensive AI Policy Education Framework for University Teaching and Learning.”International Journal of Educational Technology in Higher Education 20, no. 1 (2023): 1–25. Coeckelbergh, Mark. Why AI Undermines Democracy and What to Do about It. Cambridge: Polity Press, 2024. Davis, Austin. “The DSA Now Applies in Full: What Can we Expect?”, Democracy Reporting International, 21 February 2024.: https://democracy-reporting.org/en/office/EU/publications/dsaday-taking-stock-and-looking-ahead-to-a-super-election-year. Diakopoulos, Nicholas. “Transparency.”In The Oxford Handbook of Ethics of AI, edited by Markus D. Dubber, Frank Pasquale, and Sunit Das, 197–213. New York: Oxford University Press, 2020. Diakopoulos, Nicholas. Automating the news: How algorithms are rewriting the media. Cambridge, Massachusetts: Harvard University Press, 2019. DiResta, Renee, Kris Shaffer, Becky Ruppel, David Sullivan, Robert Matney, Ryan Fox, Jonathan Albright, and Ben Johnson. “The Tactics & Tropes of the Internet Research Agency.”Congress of the United States: U.S. Senate Document, 2019. https://digitalcommons.unl.edu/cgi/ viewcontent.cgi?article=1003&context=senatedocs. Djeffal, Christian. “AI, Democracy, and the Law.”In The Democratization of Artificial Intelligence, edited by Andreas Sudmann, 255–284. Bielefeld: Transcript, 2019. DSA, Digital Services Act,: Regulation (EU) 2022/2065, October 19, 2022. Edwards, Lilian. Regulating AI in Europe: Four Problems and Four Solutions. London: Ada Lovelace Institute, 2022. Eubanks, Virginia. Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor. New York: St. Martin’s Press, 2018. European Commission’s High-Level Expert Group on Artificial Intelligence: A Definition of AI: Main Capabilities and Scientific Disciplines, Brussels, 18 December 2018. European Commission. “Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the Committee of the Regions on the European Democracy Action Plan,”COM (2020) 790 final, December 12, 2020. European Commission. Standard Eurobarometer 98 - Winter 2022-2023. Public opinion in the European Union, 2023. https://europa.eu/eurobarometer/surveys/detail/2872. European Commission. “Commission is Gathering Views on Draft DSA Guidelines for Election Integrity,”Press Release, 8 February 2024. https://digital-strategy.ec.europa.eu/en/news/commiss ion-gathering-views-draft-dsa-guidelines-election-integrity. European Parliament. “Corrigendum to the Legislative resolution of 13 March 2024 on the position of the European Parliament adopted at first reading with a view to the adoption of Regulation (EU) 2024/ ……of the European Parliament and of the Council laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139, and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797, and (EU) 2020/1828 (Artificial Intelligence Act), P9_TA(2024)0138, (COM (2021)0206 –C9-0146/2021–2021/0106(COD))”. European Parliament and the Council of the European Union: Regulation on the Transparency and Targeting of Political Advertising, PE 90 2023 INIT, February 29, 2024. Farahany, Nita A. The Battle for Your Brain: Defending the Right to Think Freely in the Age of Neurotechnology. New York: St. Martin’s Press, 2023. Ferrara, Emilio, Onur Varol, Clayton Davis, Filippo Menczer, and Alessandro Flammini. “The Rise of Social Bots.”Communications of the ACM 59, no. 7 (2016): 96–104. Floridi, Luciano. “Open Data, Data Protection, and Group Privacy.”Philosophy & Technology 27 (2014): 1–3. García-Orosa, Berta, João Canavilhas, and Jorge Vázquez-Herrero. “Algorithms and Communication: A Systematized Literature Review.”Comunicar 31, no. 74 (2023): 9–21. GDPR. General Data Protection Regulation, Regulation (EU) 2016/679, April 4, 2016. Giannaccini, Francesca, and Tobias Kleineidam. “How the EU’s Soft Touch on Open-Source AI Opens the Door to Disinformation,”The Parliament, 16 April 2024. https://www.theparliament magazine.eu/news/article/how-the-eus-soft-touch-on-opensource-ai-opens-the-door-to-disinfor mation. 1086 J. CUPAĆAND M. SIENKNECHT Gorwa, Robert, and Timothy Garton Ash. “Democratic Transparency in the Platform Society.” In Social Media and Democracy: The State of the Field and Prospects for Reform, edited by Nathaniel Persily, and Joshua A. Tucker, 286–312. Cambridge: Cambridge University Press, 2020. Habgood-Coote, Joshua. “Deepfakes and the Epistemic Apocalypse.”Synthese 201, no. 103 (2023): 1–23. Hameleers, Michael, Toni G.L.A. van der Meer, and Tom Dobber. “You Won’t Believe What They Just Said! The Effects of Political Deepfakes Embedded as Vox Populi on Social Media.”Social Media + Society 8, no. 3 (2022): 1–12. Hamon, Ronan, Henrik Junklewitz, and Ignacio Sanchez. “Robustness and Explainability of Artificial Intelligence.”Publications Office of the European Union 207 (2020. Heald, David Albert. “Varieties of Transparency.”In Transparency: The Key to Better Governance? Proceedings of the British Academy, edited by Christopher Hood, and David Heald, 25–43. Oxford: Oxford University Press, 2006. Hildebrandt, Mireille. “Algorithmic Regulation and the Rule of Law.”Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 376, no. 2128 (2018): 20170355. https://doi.org/10.1098/rsta.2017.0355. Hinsch, Wilfred. “Differences That Make a Difference: Computational Profiling and Fairness to Individuals.”In The Cambridge Handbook of Responsible Artificial Intelligence, edited by Silja Voeneky, Philipp Kellmeyer, Oliver Mueller, and Wolfram Burgard, 229–251. Cambridge: Cambridge University Press, 2022. Ienca, Marcello. “On Artificial Intelligence and Manipulation.”Topoi 42 (2023): 1–10. Jacobsen, Benjamin N., and Jill Simpson. “The Tensions of Deepfakes.”Information, Communication & Society (2023): 1–15. https://doi.org/10.1080/1369118X.2023.2234980. Jobin, Anna, Marcello Ienca, and Effy Vayena. “The Global Landscape of AI Ethics Guidelines.” Nature Machine Intelligence 1, no. 9 (2019): 389–399. Jungherr, Andreas. “Artificial Intelligence and Democracy: A Conceptual Framework.”Social Media + Society 9, no. 3 (2023): 1–14. Kaplan, Andreas. “Artificial Intelligence, Social Media, and Fake News: Is This the End of Democracy?”In Digital Transformation in Media & Society, edited by Ayşen Akkor Gül, Yıldız Dilek Ertürk, and Paul Elmer, 149–161. Istanbul: Istanbul University Press, 2020. Keller, Tobias R., and Ulrike Klinger. “Social Bots in Election Campaigns: Theoretical,: Empirical, and Methodological Implications.”Political Communication 36, no. 1 (2019): 171–189. Kernohan, Andrew. “Accumulative Harms and the Interpretation of the Harm Principle.”Social Theory and Practice 19, no. 1 (1993): 51–72. Kertysova, Katarina. “Artificial Intelligence and Disinformation: How AI Changes the Way Disinformation Is Produced: Disseminated, and Can Be Countered.”Security and Human Rights 29, no. 1–4 (2018): 55–81. Kolt, Noam. “Algorithmic Black Swans.”Washington University Law Review 101, (forthcoming). König, Pascal D., and Georg Wenzelburger. “Opportunity for Renewal or Disruptive Force? How Artificial Intelligence Alters Democratic Politics.”Government Information Quarterly 37, no. 3 (2020): 101489. Larsson, Stefan, and Fredrik Heintz. “Transparency in Artificial Intelligence.”Internet Policy Review 9, no. 2 (2020): 1–16. Laupichler, Matthias Carl, Alexandra Aster, Jana Schirch, and Tobias Raupach. “Artificial Intelligence Literacy in Higher and Adult Education: A Scoping Literature Review.”Computers and Education: Artificial Intelligence 3, no. 100101 (2022): 1–15. Lee, Nicole M. “Fake News,: Phishing, and Fraud: A Call for Research on Digital Media Literacy Education Beyond the Classroom.”Communication Education 67, no. 4 (2018): 460–466. Mansbridge, Jane, James Bohman, Simone Chambers, Thomas Christiano, Archon Fung, John Parkinson, Dennis F. Thompson, and Mark E. Warren. “A Systemic Approach to Deliberative Democracy.”In Deliberative Systems: Deliberative Democracy at the Large Scale, edited by Jane Mansbridge, and John Parkinson, 1–26. Cambridge: Cambridge University Press, 2012. McKay, Spencer, and Chris Tenove. “Disinformation as a Threat to Deliberative Democracy.”Political Research Quarterly 74, no. 3 (2021): 703–717. McSweeny, Terrell. “Psychographics,: Predictive Analytics, Artificial Intelligence, & Bots: Is the FTC Keeping Pace?”Georgetown Law Technology Review 2 (2018): 514. DEMOCRATIZATION 1087 Milan, Stefania, and Claudio Agosti. “Personalisation Algorithms and Elections: Breaking Free of the Filter Bubble.”Internet Policy Review, 7 February (2019). https://policyreview.info/articles/news/ personalisation-algorithms-and-elections-breaking-free-filter-bubble/1385. Mogaji,Emmanuel,SundayOlaleye,andDandisonUkpabi.“Using AI to Personalise Emotionally Appealing Advertisement.”In Digital and Social Media Marketing: Emerging Applications and Theoretical Developments, edited by Nripendra P. Rana, Emma L. Slade, Ganesh P. Sahu, Hatice Kizgin, Nitish Singh,BiditDey,AnabelGutierrez,andYogeshK.Dwivedi,137–150. Cham: Springer, 2020. Narayanan, Arvind. Understanding Social Media Recommendation Algorithms. New York: Kings First Amendment Institute at Columbia University, 2023. Neuwirth, Rostam J. “Prohibited Artificial Intelligence Practices in the Proposed EU Artificial Intelligence Act (AIA).”Computer Law & Security Review 48 (2023): 105798. Neuwirth, Rostam J. The EU Artificial Intelligence Act: Regulating Subliminal AI Systems. New York: Routledge, 2023. O’Neil, Cathy. Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. New York: Crown, 2017. Susskind, Jamie. The Digital Republic: On Freedom and Democracy in the 21st Century. London: Bloomsbury Publishing, 2022. Schiff,Daniel.“Education for AI,: not AI for Education: The Role of Education and Ethics in National AI Policy Strategies.”International Journal of Artificial Intelligence in Education 32, no. 3 (2022): 527–563. Shao, Chengcheng, Giovanni Luca Ciampaglia, Onur Varol, Kai-Cheng Yang, Alessandro Flammini, and Filippo Menczer. “The Spread of Low-Credibility Content by Social Bots.”Nature Communications 9, no. 1 (2018): 1–9. Shelby, Renee, Shalaleh Rismani, Kathryn Henne, Ajung Moon, Negar Rostamzadeh, Paul Nicholas, N’Mah Yilla-Akbari, et al. “Sociotechnical Harms of Algorithmic Systems: Scoping a Taxonomy for Harm Reduction.”In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (2023): 723–741. Smuha, Nathalie A. “Beyond a Human Rights-Based Approach to AI Governance: Promise,: Pitfalls, Plea.”Philosophy & Technology 34, no. Suppl 1 (2021): 91–104. https://doi.org/10.1007/s13347020-00403-w. Smuha, Nathalie A. “Beyond the Individual: Governing AI’s Societal Harm.”Internet Policy Review 10, no. 3 (2021): 1–32. Smuha, Nathalie A. “From a ‘Race to AI’to a ‘Race to AI Regulation’: Regulatory Competition for Artificial Intelligence.”Law, Innovation and Technology 13, no. 1 (2021): 57–84. Tenove, Chris. “Protecting Democracy from Disinformation: Normative Threats and Policy Responses.”The International Journal of Press/Politics 25, no. 3 (2020): 517–537. Uuk, Risto. “Manipulation and the AI Act.”The Future of Life Institute (2022). van der Sloot, Bart, and Sascha van Schendel. “Procedural Law for the Data-Driven Society.” Information & Communications Technology Law 30, no. 3 (2021): 304–332. Veale, M., and F. Z. Borgesius. “Demystifying the Draft EU Artificial Intelligence Act—Analysing the Good, the Bad, and the Unclear Elements of the Proposed Approach.”Computer Law Review International 22, no. 4 (2021): 97–112. von Ungern-Sternberg, Antje. “Discriminatory AI and the Law: Legal Standards for Algorithmic Profiling.”In The Cambridge Handbook of Responsible Artificial Intelligence, edited by Silja Voeneky, Philipp Kellmeyer, Oliver Mueller, and Wolfram Burgard, 252–278. Cambridge: Cambridge University Press, 2022. Walmsley, Joel. “Artificial Intelligence and the Value of Transparency.”AI & Society 36, no. 2 (2021): 585–595. Warren, Mark E. “A Problem-Based Approach to Democratic Theory.”American Political Science Review 111, no. 1 (2017): 39–53. Westerlund, Mika. “The Emergence of Deepfake Technology: A Review.”Technology Innovation Management Review 9, no. 11 (2019): 39–52. Whyte, Christopher. “Deepfake News: AI-enabled Disinformation as a Multi-Level Public Policy Challenge.”Journal of Cyber Policy 5, no. 2 (2020): 199–217. Woolley, Samuel C., and Douglas Guilbeault. “United States: Manufacturing Consensus Online.”In Computational Propaganda: Political Parties, Politicians, and Political Manipulation on Social Media, edited by Samuel C. Woolley, and Philip N. Howard, 185–211. Oxford: Oxford University Press, 2018. 1088 J. CUPAĆAND M. SIENKNECHT Yang, Weipeng. “Artificial Intelligence Education for Young Children: Why,: What, and How in Curriculum Design and Implementation.”Computers and Education: Artificial Intelligence 3, no. 100061 (2022): 1–7. Yeung, Karen, Andrew Howes, and Ganna Pogrebna. “AI Governance by Human Rights–Centered Design, Deliberation, and Oversight.”In The Oxford Handbook of Ethics of AI,editedbyMarkus D. Dubber, Frank Pasquale, and Sunit Das, 77–106. New York: Oxford University Press, 2020. Zachary, G. Pascal. “Digital Manipulation and the Future of Electoral Democracy in the US.”IEEE Transactions on Technology and Society 1, no. 2 (2020): 104–112. Appendix Table 1. Binding and Non-Binding Documents of the European Union Relating to Digitalization. Binding legal instruments Regulation and Proposals for Regulation General Data Protection Regulation (GDPR) Digital Services Act (DSA) Artificial Intelligence Act (AIA) Transparency and Targeting of Political Advertising (TTPA) Digital Markets Act (DMA) Media Freedom Act Data Governance Act Cybersecurity Act Regulation on the Protection of Natural Persons with Regard to the Processing of Personal Data by the Union Institutions, Bodies, Offices, and Agencies and on the Free Movement of Such Data Gigabit Infrastructure Act European Digital Identity The European Health Data Space Data Act Chips Act Proposal for Regulation on the Digitalization of Judicial Cooperation and Access to Justice in Cross-Border Civil, Commercial and Criminal Matters, and Amending Certain Acts in the Field of Judicial Cooperation Proposal for Amending Regulation (EU) No 904/2010 as Regards the VAT Administrative Cooperation Arrangements Needed for the Digital Age Directives Data Protection Law Enforcement Directive e-Privacy Directive Directive on combating the sexual abuse and sexual exploitation of children and child pornography Audiovisual Media Services Directive (AVMSD) Directive concerning measures for a high common level of security of network and information systems across the Union Non-Binding Legal instruments Declarations Declaration on European Digital Rights and Principles for the Digital Decade Declaration: Cooperation on Artificial Intelligence Communications Communication on protecting election integrity and promoting democratic participation Tackling Online Disinformation: a European Approach Action Plan against Disinformation Artificial Intelligence for Europe Coordinated Plan on Artificial Intelligence Building Trust in Human Centric Artificial Intelligence Fostering a European Approach to Artificial Intelligence (+Annex) Protecting Election Integrity and Promoting Democratic Participation Digitalization of Justice in the European Union EU Policy on Cyber Defence Media and Audiovisual Action Plan European Strategy for Data Digital Education Action Plan 2021–2027 A Chips Act for Europe DEMOCRATIZATION 1089 Digitalization of Justice in the European Union Union of Equality: Strategy for the Rights of Persons with Disabilities (2021-2030) A Digital Decade for Children and Youth: The New European Strategy for a Better Internet for Kids (BIK+) A European Health Data Space: Harnessing the Power of Health Data for People, Patients and Innovation A European strategy on Cooperative Intelligent Transport Systems, a Milestone Towards Cooperative, Connected and Automated Mobility FinTech Action plan: For a more competitive and innovative European financial sector White papers White Paper on Artificial Intelligence: A European Approach to Excellence and Trust Non-legal instruments Policies, Programmes, Initiatives Shaping Europe’s Digital Future European Democracy Action Plan Digital Education Action Plan 2021–2027 European Green Digital Coalition Next Generation Internet Initiative (NGI) Policy Guidance on AI for Children Global Gateway Get Digital Initiative Codes of Practice The Strengthened Code of Practice on Disinformation Table 2. Classification of the EU’s measures to mitigate AI harm to democracy*. Transparency EU documents Prohibition Transparency to individual Operational transparency Reporting Transparency Risk management Education** GDPR Articles 9 Articles 5, 6, 7, 8, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 34 Articles 25, 30, 32, 40, 41, 42 Article 31, 47 Articles 33, 35, 36 Article 57 DSA Articles 25, 26, 28 Articles 14, 17, 20, 26, 27, 28, 38, 39 Articles 37, 40, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77 Articles 10, 15, 24, 42 Articles 9, 16, 18, 22, 23, 34, 35, 36, 48 AIA Article 5 Articles 13, 50 Articles 10, 11, 12, 13, 15, 17, 18, 19, 44, 47, 48, 49, 53, 55, 77 Article 20, 27, 52, 77, 91 Articles 9, 14, 15, 20, 27, 43, 55, 57, 60, 72, 73,74, 75, 76, 89, 90 Article 3 TTPA Article 18 Articles 11, 12, 18, 19 Article 9, 19 Articles 14, 16, 17, 20 Article 15, 19 *The table selectively highlights articles most pertinent to rights-based and systemic AI harm to democracy, excluding those with lesser relevance. Articles may cross several categories, reflecting their interconnected roles in the documents. ** The EU deals with education-related measures primarily in its non-binding legal instruments and non-legal instruments while regulations only sporadically cover this aspect. 1090 J. CUPAĆAND M. SIENKNECHT