<i>Why AI Undermines Democracy and What to Do About It</i> by MarkCoeckelbergh, Cambridge, UK: Polity. 2024. pp. 144. $22.95 (pbk). ISBN: 9781509560936.<i>Algorithmic Institutionalism: The Changing Rules of Social and Political Life</i> by RicardoMen

Nearly 4000 years ago, Babylonians inscribed algorithms on clay tablets, sometimes for the love of knowledge and often for practical tasks like constructing efficient cisterns. The term algorithm originates from the Latinized name of the ninth-century mathematician and astronomer, Muhammad al-Khwarizmi, but today frequently refers to processes in digital technology like artificial intelligence (AI). An algorithm is a set of rules or instructions designed to perform specific tasks. While simpler algorithms can handle basic tasks on their own, more advanced algorithms are needed in AI systems for processing data, recognizing patterns, and generating models. Paired with massive amounts and varieties of data, they have become powerful predictive instruments. These technologies are commonly likened to recipes with apolitical connotations. However, interdisciplinary scholarship has increasingly challenged the notion that their programming, output, and reception are neutral. Extensive work on algorithmic bias, for example, has illuminated how these technologies often misrepresent minorities and perpetuate existing inequalities related to race, gender, and class. Equally important is the focus on the inaccessibility and lack of transparency in algorithmic decision-making processes. Moreover, the political implications of algorithms and AI, including their role in surveillance, are overt and widespread among policymakers, legal institutions, and scholars. It is in this critical tradition that the two texts reviewed here are situated, as both underscore how these advancements provoke new and complex challenges, continuing to question how algorithms impact our personal and political lives. Ultimately, the authors ask whether these technologies are benign instruments for humanity, capable of bolstering democratic ideals, or whether they threaten the integrity of our political institutions and the values they are meant to uphold. They ask important questions, critique current practices, and suggest a demanding research agenda for those who might try to provide answers. Mark Coeckelbergh, in Why AI Undermines Democracy and What to Do About It, contends that AI, as it is currently conceived and deployed, poses a significant threat to the foundational principles and norms of democracy. He argues that AI is not a neutral tool but a potent force that can erode essential democratic values like freedom, equality, and trust. Coeckelbergh elucidates how AI serves elite interests at the expense of the broader populace, thereby undermining the knowledge and trust vital for democratic governance and potentially paving the way for totalitarianism through the spread of disinformation. Coeckelbergh's exploration of the political implications of AI is both timely and astute, making it a valuable resource for those interested in the intersection of technology and democracy. Coeckelbergh states that it is “high time” that Hannah Arendt's analysis of the origins of antisemitism, imperialism, and totalitarianism gets deployed to help us think about our current conditions in the context of digital technologies (p. 5). But Coekelbergh overlooks much of the considerable work that has already been done on this through the lens of Arendt, bypassing theoretical discussions that could have enhanced his analysis, particularly in diagnosing the totalitarian trends in AI and proposing democratic solutions. For example, Lisa Herzog in “Old Facts, New Beginnings: Thinking with Arendt about Algorithmic Decision-Making,” Jennifer Forestal in “Constructing Digital Democracies: Facebook, Arendt, and the Politics of Design,” and Ewa Płonowska Ziarek in “Against Digital Worldlessness: Arendt, Narrative, and the Onto-Politics of Big Data/AI Technologies” have drawn incisively from Arendt to critique AI and digital politics, highlighting the erosion of democratic values. Coeckelbergh's failure to engage this literature is unfortunate. Further, Shoshana Zuboff's The Age of Surveillance Capitalism explicitly draws from the Origins of Totalitarianism to describe and judge new forms of power. However, Coeckelbergh only references Zuboff in passing when discussing capitalism and AI, without acknowledging her use of Arendt. Zuboff relies on Arendt to illuminate how our failure to address social instability, the erosion of trust, and extreme wealth inequality has allowed corporations to exploit these vulnerabilities. Had Coeckelbergh engaged more with other interpretations of Arendt in this context, he might have further advanced his analysis of problems related to authority, trust, and knowledge. In a similar vein, the book does itself a disservice by ignoring authors who are critical of the application of Arendt's ideas to the epistemic ramifications of digital technology, such as Simone Chambers. While Coeckelbergh's text claims to be “following in the footsteps of Hannah Arendt” to show how the use of AI “risks undermining the knowledge and trust basis of democracy through the creation of power asymmetries, manipulation, the erosion of the distinction between what is real and fake, and the creation of epistemic bubbles” (p. 7), it fails to engage with Chambers’ critical perspective on this very issue. In her article “Truth, Deliberative Democracy, and the Virtues of Accuracy: Is Fake News Destroying the Public Sphere?,” Chambers challenges the applicability of Arendt's ideas to contemporary epistemic crises associated with post-truth politics and digital technology. According to Chambers, Arendt's focus on power dynamics rather than democratic processes, and her pessimism about the potential for citizen resistance to the manipulation of truth, may contribute more to the panic about post-truth politics rather than provide practical solutions. Furthermore, Chambers critiques Arendt's notion that truth is inherently despotic and antithetical to politics, advocating instead for a view of truth as corrigible and fallible—one that aligns more with deliberative democracy and is better suited to the challenges posed by fake news and digital misinformation. While Coeckelbergh's ambition to chart a “way out” of the authoritarian and antidemocratic tendencies of AI is commendable, the solutions he proposes—though conceptually rich—lack political traction. Coeckelbergh outlines several pathways toward more democratic AI, such as regulating and readjusting the power balance between private corporations and democratic institutions, embedding political values in technology development, and strengthening democratic institutions through AI-assisted deliberation. However, the mechanisms for achieving these goals are not well fleshed out. For example, while Coeckelbergh suggests that parliaments and citizens, rather than private companies, should decide ethical boundaries for AI, he provides limited guidance on how these complex transformations could be achieved in today's sociopolitical environment, or whether and how AI could be employed to advance democratic politics. Coeckelbergh advocates for a cultural and educational Renaissance inspired by humanism, republicanism, and enlightenment values. While the call for interdisciplinary education and a renewed focus on the common good is inspiring, how to achieve these lofty goals remains unclear. Much soil has been turned and some seeds planted, but there's little yield, as the book's ambitious vision for democratic AI lacks the mechanisms for it to take root. Ricardo Mendonça, Virgilio Almeida, and Fernando Filgueiras, in Algorithmic Institutionalism: The Changing Rules of Social and Political Life, conceptualize algorithms as inherently political institutions that shape social norms and behaviors and are capable of either reinforcing or undermining democratic processes. Like traditional institutions, algorithms establish rules and norms that influence collective action, suggesting that insights from institutional theories could inform their development. Conversely, understanding the dynamics of emerging technologies may necessitate a reorientation of existing theories to better address the challenges these technologies pose. The authors call for a critical reassessment of political values and a democratization of algorithmic systems to ensure they enhance democratic ideals rather than entrench authoritarianism. Mendonça, Almeida, and Filgueiras’ text is as ambitious as Coeckelbergh's and demonstrates similar shortcomings. The text introduces a novel framework for understanding algorithms as institutions, yet it engages only superficially with the democratic values it seeks to uphold and lacks the practical depth necessary to propose meaningful solutions. Though the authors acknowledge the influence of algorithms on democratic values, they leave critical questions unanswered, making their calls for democratizing algorithms more aspirational than actionable. The authors of Algorithmic Institutionalism position algorithms as institutions that shape social behavior and influence politics. While they convincingly identify algorithms’ impact on democratic values such as participation, equality, pluralism, public debate, accountability, and liberty, their analysis falls short in specifying which conception of each value should be prioritized or how pursuing one value often requires sacrificing aspects of others. Participation is an admirable goal, but what kind of participation is optimal, and what level is adequate within the digital public space? What is the appropriate balance between online freedom and security? These are important questions, yet Algorithmic Institutionalism dedicates only a few pages to each. This level of engagement is troubling because other scholars have examined many of these concepts in depth. For instance, Ruha Benjamin's Race After Technology examines various conceptions of equality and solidarity, revealing how digital technologies can either exacerbate or challenge existing social hierarchies. Frank Pasquale's The Black Box Society similarly complicates the understanding of safety, security, and accountability, emphasizing the need for transparency and public scrutiny in algorithmic governance. In a different vein, Alnemr's work on democratic self-government and algorithmic governance explores the complexities of participation, highlighting the tensions between ensuring meaningful civic engagement and the risks of participative tokenism. By not fully grappling with these perspectives, Algorithmic Institutionalism fails to explore the various conceptions of these democratic values or provide clear guidance on how they should be implemented. Their discussion of liberty is a case in point. They argue that “our comprehension of democracy entails both negative and positive liberties. Freedom from and freedom to are necessary to the autonomous participation of citizens in the construction of the rules that govern their lives” (p. 145). This may be true. However, these two visions of freedom are often contentious. And differing perspectives on privacy, nudging, and transparency, for example, exacerbate longstanding tensions between individual freedom and the common good that the authors do not address. Institutions like the European Union have been quite explicit about democratic values in the context of digital transformation. For example, The European Declaration on Digital Rights and Principles offers a framework for action centered on six key democratic values: placing people at the center of digital transformation, promoting solidarity and inclusion, ensuring freedom of choice, encouraging participation in the digital public space, enhancing security and empowerment, and advocating for sustainability. Along with a yearly review of practices in the European Union, the declaration serves as an educator so that citizens know their digital rights and that lawmakers and businesses have a clear set of standards. This kind of clarity goes a long way in creating a trustworthy environment that fosters many of the values alluded to in Algorithmic Institutionalism. The book does not address this declaration or any others that are attempting to secure democratic values in the face of rapid technological change. The book examines several important case studies, including algorithmic systems used in security (Chapter 3), policymaking (Chapter 4), and recommendation systems (Chapter 5). However, by opting to cover numerous examples in each chapter, the authors miss the opportunity to thoroughly examine the intricacies of any single institution, which could provide insights into the practical application of their theoretical framework. This zoomed-out analysis neglects the lived experiences of those who develop and interact with algorithms and AI, which would be helpful for understanding the real-world implications of algorithmic governance. For instance, in the realm of security, algorithms play a critical role in surveillance and risk assessment, directly impacting individuals’ sense of freedom and privacy. A more ethnographic approach could explore how people subject to these surveillance systems perceive and respond to the constraints imposed on their freedoms. Similarly, in the context of policymaking, algorithms are increasingly used to allocate resources and shape public policy decisions. Yet, there is little exploration of how these algorithmic decisions are experienced by those affected, especially in terms of their ability to influence or challenge these decisions. In the case of recommendation systems, which are pervasive in digital platforms for music, movies, and dating, the book touches on how these algorithms influence user choices and preferences. However, an inductive investigation—akin to the fine-grained analysis found in Nick Seaver's Computing Taste or the critical examination of technological mediation in Paul Rekret's Take This Hammer—could provide a richer understanding of how individuals experience and navigate these systems. Seaver's work, for instance, reveals how those working in music recommendation often combine technical logic with their own ideas, exposing the subjective underpinnings of what appears to be a purely algorithmic process. And Rekret's analysis demonstrates how technological infrastructures and broader contours of power shape, but do not wholly determine cultural production. Mendonça, Almeida, and Filgueiras present a useful critique of the influence of algorithms on democratic institutions. Not least, the text is valuable for their many examples of algorithmic power and resistance in the Global South. Their framework for understanding algorithms as institutions is innovative, even if it does not adequately engage with the complexities of the democratic values it aims to defend. Together, these books remind us of the complex relationship between emerging technologies and democratic ideals. They present thought-provoking assessments of the challenge of integrating AI and algorithms into democratic frameworks without compromising fundamental values. Their perspectives are panoramic and bear the benefits and shortcomings inherent to such overviews. They help us see the forest. But we must also tend to the trees. For that task, greater attention to available theoretical scholarship and on-the-ground implementation is required. The rush to address AI's tremendous impact on society has led to a proliferation of critical scholarship. This is undoubtedly a positive development. However, much of this work is produced with such urgency that contributions from neighboring disciplines and political movements are frequently overlooked. This situation reflects more of our reactive relationship to technology and usurious demands to publish than a set of thoughtful solutions to these challenges. To understand the implications of AI on democracy and exercise our judgments carefully, we should slow down, learn from each other, and think more about what we are doing. We must also cultivate kinship and act together to challenge the authoritarian and exploitative deployments of these technologies.