Digital citizen participation tools have proliferated over the past ten years. They do not always deliver on their promises. In Reboot: AI and the Race to Save Democracy, Beth Simone Noveck argues that AI can help connect public knowledge, deliberation, and institutional action, under certain conditions of design and governance. The tool alone is not enough; what also matters is the capacity of institutions to listen, act, and account for their decisions.
The Essentials
- Noveck proposes a broader framework of “democratic AI,” centered on the capacity of institutions to listen, deliberate, act, and integrate transparency, accountability, and participation.
- Several participation experiments show the importance of clearly defining the role of participants, the rules of the process, and the follow-up given to contributions.
- AI expands the potential of these mechanisms, but it introduces risks of technical capture and power concentration that require explicit governance.
- More powerful participatory tools can either foster broader democracy or further concentrate power among the intermediaries who design them.
Who is Beth Simone Noveck
Beth Simone Noveck directs the Governance Lab at New York University, which she founded after coordinating the open government policy of the Obama administration. She has overseen concrete digital participation projects in several countries, notably in Italy where she advised the Liguria region, and in Uruguay. Reboot is her third work on the subject following Wiki Government (2009) and Smart Citizens, Smarter State (2015). Her trajectory—academic, practitioner, government advisor—gives her a rare position: she has seen these mechanisms work or fail from the inside.
The Thesis: Participation Without Decision-Making is Mere Theater
Noveck starts from an empirical observation. Many digital citizen participation platforms suffer from the same structural flaw: they collect opinions, produce visualizations, and generate reports, but the link between citizen contributions and the final decision remains opaque or nonexistent. The citizen neither knows if their opinion was read nor why it was retained or dismissed. Under these conditions, participation can reinforce distrust rather than reduce it: the citizen may understand that their opinion was sought without having any visible effect.
Reboot proposes a broader framework of “democratic AI,” centered on the capacity of institutions to listen, deliberate, act, and integrate transparency, accountability, and participation. For a decision already in the pipeline, Noveck recommends targeted engagement, limited in time and linked to a real outcome; she also acknowledges an open listening phase to surface issues to be placed on the agenda. Institutions should then clearly communicate how the results of the participatory process were taken into account and what was implemented.
The link to a decision and feedback are important criteria, but they are not alone sufficient to evaluate the democratic quality of a tool. Inclusion, accessibility, available resources, process integrity, and administrative capacity matter equally.
Lessons from Experiments
Noveck draws on a comparative body of work covering multiple geographic areas. In Europe, participatory budgets in cities like Porto and Gdańsk illustrate the importance of making participation rules, available amounts, and implementation timelines clear. Clearly defined objectives, rules, and follow-ups can promote participation and its credibility, but they are not sufficient by themselves. In Latin America, some legislatures have experimented with digital consultations on laws under debate. In Brazil, some citizen ideas online can become formal parliamentary proposals.
In Uruguay, PROPACI collects comments and proposals, but available official sources indicate the absence of clear rules on how they are taken into account.
In the United States, some states have tested randomly selected citizen panels, in the manner of the Irish model, to address polarizing subjects like electoral redistricting or the regulation of psychoactive substances.
The common thread is not technological. The tools vary considerably from case to case: online forums, mobile applications, collaborative mapping platforms, consultations via messaging. What can distinguish the most robust mechanisms is the prior institutional decision to define the real role of participants and the expected effects of the process. In several cases, elected officials and administrations accept not deciding alone on the scope open to participation.
Noveck formulates this without condescension toward institutions. She recognizes that delegating a decision carries real political risks for those who undertake it. An elected official who commits to following the result of a consultation may find themselves constrained by a citizen majority that contradicts their beliefs. This structural tension can explain why some mechanisms are designed to consult without constraining.
AI as Amplifier and as Risk
The book’s second thesis concerns artificial intelligence. Noveck sees AI as a potential amplifier of these mechanisms: automatic synthesis of massive contributions, instant translation, detection of thematic convergence across thousands of comments, customization of interfaces to reach populations usually absent from consultations. These capacities are real. AI can help process far larger volumes of contributions, but no general guarantee allows us to assert an absence of loss of analytical quality when scaling from 1,000 to 1,000,000. This quality depends on data, models, evaluation methods, traceability, and human control.
But AI introduces a second level of delegation that the book examines carefully. When algorithms sort and synthesize contributions, the choices of system designers and governors can strongly influence the agenda; guarantees of transparency, supervision, and contestation are therefore necessary. Noveck knows this tension. She does not resolve it with default optimism. She poses the question of governance of the tool itself as a condition for the legitimacy of the process.
The work of Lê Nguyên Hoang on participatory governance of algorithms provides complementary perspective. Hoang, through the Tournesol project, has attempted since 2020 to build a recommendation platform whose encoded values would be defined collectively by its users rather than imposed by its developers. The experience documents a problem that Noveck touches on without fully developing: participating in an algorithm’s governance requires a level of technical abstraction that most participants do not master.
Tournesol remains a small-scale experiment precisely because pluralistic governance of models hits a competence bottleneck. If digital participation can be captured by those who control its architecture, the institutional guarantees that Noveck calls for must extend to the design of the tools themselves, not just their use.
This dimension is central to understanding why, as an analysis published here on AI in Asian administrations showed, digital tools deployed by public institutions can, in certain contexts, reinforce existing power logics rather than redistribute them. The parameters of a tool can influence its results and distributive effects; they must therefore be transparent, controllable, and subject to pluralistic governance.
The Book’s Limitations
Reboot has honest limitations. The first is empirical. The experiments cited are real and documented, but their selection tends toward cases where participation worked. Noveck identifies conditions for success, but she treats conditions for large-scale failure less systematically. In certain French cities, notably Paris, participatory budgets also raised questions about the participation of underrepresented groups and inequalities in access to mechanisms.
The digital tool did not resolve this initial inequality; it can sometimes accentuate it by favoring those with the time, digital skills, and habit of public debate.
The second limitation concerns political scalability. The most convincing experiments operate on delimited scopes: a district budget, a law article, a localized development project. Noveck discusses possibilities opened by AI for extending these mechanisms to national decisions, but remains cautious about the political conditions for such scaling. A national government that accepts delegating a budget decision to a citizen panel faces institutional resistance of a different order of magnitude than a mayor who submits their roadway plan. The delegation mechanics that the book theorizes supposes a political courage that local experiments do not presuppose to the same degree.
The third limitation is conceptual. Noveck considers institutional design as an important lever of trust, but not as a sufficient solution in itself. Work on budgetary trajectory as a political act reminds us that trust or distrust of institutions obeys dynamics that exceed the procedural quality of participation. A citizen who no longer believes in institutions does not necessarily participate better because the platform is better designed.
Noveck knows this, but she chooses to work on what is actionable: improving institutional design. This is a defensible pragmatic choice, not naïveté.
The Question Structuring the Next Fifteen Years
The horizon that Reboot opens exceeds the frame of each individual experiment. If digital tools and AI make it possible to involve more people in complex public decisions, two plausible trajectories open up, and they do not lead to the same place.
In the first, institutions progressively accept delimiting real decision scopes, accounting for their choices, and allowing consultation results to inform policies. This scenario supposes a cultural transformation of administrations and political acceptance of the risk that the citizen majority go against elected officials’ preferences. Signals exist: Ireland has shown that a citizen assembly can address polarizing constitutional subjects and substantively inform Parliament’s work and referendums; some American states have followed comparable paths on electoral redistricting. But these examples remain demanding to replicate.
In the second trajectory, tools spread without clear institutional commitments being made. Platforms proliferate, consultations multiply, interfaces become more sophisticated, and the result can be surface-level participation that absorbs civic energy without converting it into real influence. In this scenario, distrust does not necessarily decline; it can take a new form, more technical and therefore harder to name. A citizen who devoted time to a participatory platform without perceiving an effect on the final decision may derive increased skepticism. Poorly designed or poorly institutionally constrained tools can worsen the problem they purport to solve.
Noveck proposes design frameworks for linking listening, engagement, decision, implementation, and feedback, rather than a single two-question test. She recommends adapting the form of participation to the stage of the public policy cycle, tying decision commitment to a real outcome, and closing the loop after the decision. The link to a decision and feedback are important criteria, but they must be supplemented by inclusion, accessibility, transparency, integrity, and institutional capacity.
These criteria also apply, with slight extension, to AI systems that claim to integrate user values into their recommendations: on what criteria, decided by whom, verifiable how.
The question remains open. It deserves to be asked of each new mechanism that presents itself as a democratic advance.
Reasons to Read This Book
Reboot is first addressed to practitioners: elected officials, civil servants, civil society leaders, public policy designers. It offers a framework for reflection on listening, deliberation, action, and feedback. But it also interests anyone thinking about the relationship between technology and democracy without wanting to choose between digital enthusiasm and rejection. Noveck refuses both postures. She stands in an uncomfortable place: the tool can work, but only if institutions accept a transformation they are reluctant to undertake.
It is there, in this discomfort, that the book’s interest lies.
For a French reader, the relevance is immediate. The debate on the Citizens’ Convention for Climate, on the participatory mass gatherings that followed the yellow vests movement, on the limits of the Grand National Debate, finds in Reboot a reading grid that helps interrogate the conditions under which these experiments produce influence or frustration. Just as AI applied to education can amplify inequalities rather than reduce them, participation tools can reinforce the illusion of inclusion without modifying power relations. Noveck provides elements to make the difference.
Bibliographic Information
Title: Reboot: AI and the Race to Save Democracy Author: Beth Simone Noveck Publisher: Yale University Press Publication Date: June 23, 2026 Pages: Not specified
Sources
- Beth Simone Noveck, Reboot: AI and the Race to Save Democracy, Yale University Press, 2026, https://yalebooks.yale.edu/book/9780300279993/reboot/
- Title, content and general framework of Reboot
- Official interview with Beth Simone Noveck
- Noveck’s participation framework
- OECD, challenges of emerging technologies for participation
- OECD, digital risks, inequalities and capture
- Participation portals and feedback to contributors
- Federal Senate of Brazil, e-Cidadania
- Uruguay, official diagnosis of PROPACI
- Apur, participation in Paris’s participatory budget
- Citizens’ Assembly of Ireland