Does AI weaken democratic institutions?

Most of this wiki maps the hopeful case for civic AI. This page holds the strongest argument against it, because a serious field has to take its critics seriously.

The argument: AI infeebles institutions

Legal scholars Woodrow Hartzog and Jessica Silbey (Boston University) make a blunt claim: today’s AI systems are built to function in ways that degrade and are likely to destroy the civic institutions a democracy depends on — journalism, higher education, the rule of law, civic life. The harm isn’t a bug to be patched; it’s how the tools work as designed, dropped into an economy with extractive incentives. They name three mechanisms:

It erodes expertise — by substituting plausible answers for the slow work that builds and certifies knowledge.
It short-circuits decision-making — collapsing the deliberation, friction, and contestation that good institutional decisions require.
It isolates people — weakening the human bonds and shared purpose that democratic participation is built on.

Their counterintuitive conclusion: efficiency is not the primary goal of institutions like courts or universities. They need friction, points of human contestation, and deference to human expertise in order to thrive. So AI should be let in judiciously — bespoke over general, only where it genuinely serves the organisation and the institution it belongs to.

A reframe: worthy, not trustworthy

A related move (from Mozilla’s Nabiha Syed) is to stop asking whether an AI system is trustworthy and ask whether it’s worthy — worthy of our time, our resources, of taking something from the commons. That turns a technical question into a democratic one: who gets to decide what’s worthy, and how are resources allocated to match?

The counterweight: a mirror, not a verdict

Not everyone on the critical side goes as far. Angela Oduor Longe (Ushahidi) frames technology as a mirror of society — not inherently good or bad. Strong democratic foundations let AI amplify them; weak ones get their dysfunction amplified too. On that view the problem isn’t AI as such but its current design and political economy: the handful of actors at the helm, and the incentives they answer to. The response is to build alternatives — locally-grounded, publicly-governed systems that demonstrate a different utility — rather than to abandon the technology. The global-south optimism and global-north anxiety about AI track exactly this difference in who is building it and to what end.

Yoshua Bengio adds the structural worry that ties it together: AI concentrates power in private hands, and democracy is about sharing power — so the two are in tension well before any science-fiction scenario arrives.

A subtler erosion: government that acts before you ask

The arguments above are about AI in society at large; a quieter version shows up inside the state. Gustavo Maia’s forthcoming book Zero-Click Government makes the optimistic case for anticipatory governance — public institutions using administrative data and AI to act before citizens ask, delivering a benefit or catching a need without the forms. In an afterword, Beth Simone Noveck (who has spent a career on open government) grants the appeal but names what it risks losing. A benefit application or a complaint was never only friction: it was a feedback signal, a moment where a citizen told the state what they needed and the state had to answer. Act on inferred demand instead, and that signal — along with the chance to contest a decision you never knew was being made — can quietly disappear. Her response isn’t to refuse the tools but to insist that anticipatory systems be co-designed with the communities they serve, kept transparent and contestable, and treated as a place where participation has to be built in rather than engineered out. It’s the bespoke-over-general rule again: efficiency is not the only thing a public institution is for.

The honest position sits between the poles: be precautionary about where AI enters public life, insist on democratic governance of the stack, and judge each deployment by whether it strengthens or hollows out human agency. See also why civic tech projects fail and synthetic participation.

Sources

Woodrow Hartzog & Jessica Silbey, “How AI Destroys Institutions,” 2025: papers.ssrn.com · scholarship.law.bu.edu
Mozilla Foundation, “AI for democracy or democracy for AI?” panel (Nabiha Syed, Angela Oduor Longe, Arnau Monterde, Woody Hartzog, Claudia Chwalisz), 2025: youtube.com/watch?v=bIGWZY-fCk0
Yoshua Bengio, remarks — Collective Intelligence Project “Democratic AI” panel, 2024: youtube.com/watch?v=OQxMGMB5kL8
Beth Simone Noveck, “Zero-Click Government: Omakase or Loss of Agency?” — Reboot Democracy, 2026 (afterword to Gustavo Maia’s Zero-Click Government): rebootdemocracy.ai