AI: Tool for Democratic Renewal or a Hidden Threat?

AI: A Turning Point for Modern Democracy

Throughout history, revolutions in the flow of information have transformed modes of governance. The printing press democratized access to knowledge, facilitating the emergence of the Reformation and representative governments. The telegraph enabled the administration of vast territories like the United States, accelerating modern bureaucratization. Broadcast media created national audiences, strengthening mass democracy. Today, we stand on the brink of a new transformation. Artificial intelligence (AI) is rapidly establishing itself as the primary interface through which we form our beliefs and participate in democratic governance. This upheaval, if not managed, could further undermine already fragile institutions in the United States. However, it also presents an opportunity to address persistent issues such as declining civic engagement and increasing polarization. The future depends on the design choices currently being made, often without our awareness.

The Epistemic Layer: A New Way of Knowing

The way we access knowledge is undergoing a profound change. Increasingly, individuals are turning to AI to determine what is true, understand events, and know whom to trust. The search for information is already heavily influenced by AI. Future AI assistants will be capable of synthesizing, structuring, and presenting information with authority. For many, querying an AI will become the norm for forming an opinion about a candidate, a policy, or a public figure. Therefore, whoever controls the discourse of these models holds increasing influence over popular beliefs.

Personal AI Agents: A Shift in Information Interaction

Technology has always influenced how citizens interact with information. However, a new challenge looms with the emergence of personal AI agents. These systems could transform not only the reception of information but also the actions that follow. They will conduct research, draft communications, advocate for causes, and lobby on behalf of the user. They will guide decisions such as voting on a measure, supporting organizations, or responding to government notices. In this sense, they could redefine the relationship between individuals and the institutions that govern them.

The Risks of Polarization and Radicalization

The experience of social media has shown the dangers of optimizing engagement at the expense of understanding. Platforms do not need an explicit political agenda to foster polarization and radicalization. An AI agent, designed to maintain engagement by understanding the user's preferences and anxieties, presents the same risks. These dangers are even more insidious when an agent presents itself as a personal advocate, speaking and acting on behalf of the user, thereby gaining their trust through this proximity.

Towards a Fragmented Public Sphere

Broadening our perspective, AI agents and humans could soon coexist in the same forums, making their distinction difficult. Even if each AI agent is well-designed and aligned with its user's interests, the interaction of millions of agents could produce undesirable outcomes. Studies show that agents without individual bias can still generate large-scale collective biases. Moreover, a public sphere where everyone has a personalized agent aligned with their existing opinions does not promote the shared deliberation necessary for democracy. It transforms into a multitude of private worlds, each internally coherent but collectively unhelpful for democratic debate.

A Fundamental Transformation of Citizenship

These three transformations — the way of knowing, acting, and participating in collective governance — represent a fundamental change in the texture of citizenship. In the near future, political opinions will be formed through AI filters, civic engagement will be exercised via AI agents, and institutions and public discussions will be shaped by the interactions of millions of these agents.

Democratic Institutions Facing AI

Current democracy is not prepared for these changes. Our institutions were designed for a world where power was exercised visibly, where information circulated slowly, and where reality seemed shared. These foundations were already in decline before the advent of generative AI. However, this need not be a story of decline. To avoid this scenario, it is crucial to design better solutions.

Ensuring Truthfulness and Reducing Polarization

At the information level, AI companies must intensify their efforts to ensure the truthfulness of model outputs. They should also explore promising findings that suggest AI can help reduce polarization. A recent field assessment of AI-generated fact-checks revealed that individuals with diverse political views found AI-written notes more useful than those written by humans. Although this article has not yet been peer-reviewed, it represents a potentially revolutionary finding: AI-assisted fact-checking could achieve bipartisan credibility that has eluded most manual efforts. A better understanding and transparency regarding how models formulate these assertions and prioritize sources could enhance public trust.

Accurate Representation by AI Agents

At the agentic level, it is essential to ensure that AI agents accurately represent their users. An agent should never have its own agenda or distort the opinions of its user, which is technically complex in areas where user preferences are not explicitly expressed. However, accurate representation should not become an accessory to motivated reasoning. An agent that refuses to present uncomfortable information or protects its user from any challenge to their prior beliefs is not acting in the best interest of the individual.

Integrating AI into Governance

Finally, at the institutional level, policymakers must quickly harness the potential of AI to make governance more responsive and legitimate. Several states and municipalities are already using AI-mediated platforms to conduct large-scale democratic deliberations, relying on research showing that AI mediators can help citizens find common ground. As agents become common participants in public contribution processes — and there is already evidence that bots distort these processes — identity verification for humans and their agentic representatives must be integrated from the outset.

Designing a New Generation of Democratic Infrastructures

What is needed is a new generation of democratic, technological, and institutional infrastructures designed for the current world. Failing to design for democratic outcomes means designing for something else. And the history of unaccountable power leaves little room for optimism about what that something else might be.