Provocations

Democracy's deepest survival trait is its capacity to correct itself without violence, which depends on citizens being able to encounter what they do not already believe — what Arendt called “representative thinking.” Political representation has historically carried that function, living in the gap between the represented and their representative. Sorota's provocation is that “mirror representation” is incoherent as a design goal for civic agents: AI built as a flawless extension of a single user, admitting no other standpoint. His essays ask whether agentic representation can instead be recovered by widening, not narrowing, the standpoints a citizen reckons with.

Democratic institutions are failing not because the public sphere has collapsed, but because the operating system beneath it — optimization, maximization, humans inserted in the loop of AI — extends a design stance into a domain that asks for an intentional one. Across three essays, Tang argues for a shift from MaxOS to an ethic of care, from humans in the loop of AI to AI in the loop of humanity, and from one rational public sphere to many bounded, federated communities: what Tang calls techno-communitarianism, with the practical goal of making polarization-reduction measurable the way CO₂ became measurable for climate, so the plural, not the consensus, becomes democracy's basic unit.

Ovadya's contribution maps what the intelligence infrastructure underlying democracy's checks and balances could look like as AI advances reshape it, intentionally or not. He sketches good versions of this infrastructure, focusing largely on the legislative branch but extending to executive, judicial, and other checks and balances, and flags the likely challenges ahead. The aim is to clarify what people and systems democratic institutions will need to hire, build, and procure to function in the age of AI: in effect, a “meta-org chart” for democracy that maps out the right mix of human and AI roles.

Noveck's provocation argues that democracy lacks a transformative vision because it has misdiagnosed the challenges of the information age. She contends that democratic institutions were built for information scarcity and have failed to adapt to an era of abundance, where public voice has expanded but institutional capacity to listen and learn has not. She calls for a democratic “moonshot”: a sustained effort to redesign institutions as informational systems capable of gathering collective intelligence at scale and measuring success through meaningful outcomes rather than participation alone.

Hadfield's provocation argues that the central democratic deficit is not a lack of public participation but the absence of institutions that translate collective deliberation into meaningful governance. He contends that AI makes it possible to build this missing infrastructure by enabling structured public dialogue at scale, identifying shared priorities and points of disagreement, and translating them into transparent governance principles. In this vision, democratic legitimacy depends not on AI itself but on making the entire process — from public input to synthesis, implementation, and accountability — open, inspectable, and repeatable.

The end of social media has arrived. The question is how what comes after gets shaped. That era promised participation but built an economy of capture, rewarding conflict, outrage, and affective escalation because those extract data best. As it is displaced by algorithmic broadcasting, semi-private enclaves, and AI-mediated infrastructures, a rare moment of institutional plasticity opens. Across three essays, Törnberg develops a political economy of attention, on the claim that attention is a democratic resource that cannot be left to commodification.

If democratic theory is shifting from communication to infrastructure, then the ways AI systems represent and surface uncertainty become constitutive, rather than peripheral, to democratic life. Drawing on participatory research with GPs and educational assessors, as well as the NAVIGATE research programme, Delacroix argues that current LLM design often treats non-quantifiable uncertainty as a calibration problem to be eliminated, undermining the professional norms and collective judgment that sustain expertise. Her provocation contends that informational democracy requires conversational infrastructures that preserve, rather than suppress, productive uncertainty.

Since the “Transformer” turn, LLMs have become central to the machine-learning revolution — but that revolution was more contingent than it now appears. Pénigaud's provocation is that today's LLMs constitute weapons of mass destruction for human morality and responsibility, replacing humanly grounded motivation with “orphan reasons” that blur who is accountable for a decision. But LLMs and chatbots should not be conflated: unlike human-like chatbots, LLMs' capacities might instead be leveraged to improve democratic processes, if what AI is for, how it is financed, and who controls it are rethought.