Institutional Relays

In the AI Story

The concept addresses a persistent democratic problem: individual citizens possess valuable knowledge about governance failures—the worker knows the workplace is unsafe, the teacher knows the curriculum is failing, the engineer knows the system can be misused—but individual knowledge alone cannot produce democratic accountability. The knowledge must be amplified, verified, and translated into forms that democratic publics can use to exercise the counter-democratic powers of vigilance and evaluation. This amplification is what institutional relays perform. They are the connective tissue between the individual who knows and the public that must act on that knowledge.

The engineer at the AI company described in The Orange Pill performed an act of denunciation: she identified a misuse risk, proposed a redesign, escalated when ignored, persisted for six months. Her denunciation was technically specific, ethically grounded, and directed at people with authority to act. It failed not because it was unpersuasive but because no institutional mechanism existed to convert her persuasion into organizational change or public accountability. The internal hierarchy mediated between her denunciation and any possible action, and the hierarchy's incentive structure (efficiency, speed, competitive advantage) filtered out the signal. She was a sensor in a system with no nervous system—capable of detecting danger, connected to nothing that could process detection into response.

Effective relays possess three characteristics Rosanvallon's framework identifies: independence (the relay is not controlled by the entity it relays information about), authority (the relay has genuine power to investigate, publicize, and demand response), and accessibility (the relay can be reached by ordinary individuals without requiring extraordinary resources, legal sophistication, or institutional position). Whistleblower protections are accessible but often lack authority—they protect the speaker from retaliation but cannot compel the institution to change. Congressional hearings have authority but are not accessible to most individual sources. Independent investigative journalism was historically independent and accessible but has lost authority as the media landscape fragments.

For AI, building effective institutional relays requires creating new forms that satisfy all three characteristics simultaneously. Possibilities include: independent AI safety ombudsmen with authority to investigate complaints, protect sources, and issue public findings; mandatory incident reporting systems for AI-driven harms (analogous to aviation safety reporting); standing congressional committees with genuine technical capacity to evaluate AI governance rather than performing theater of oversight; and public interest technology organizations that advocate for affected communities within regulatory processes, serving as the institutional intermediaries between individual knowledge and collective action. Each possibility faces obstacles—funding, political will, technical capacity requirements—but the obstacles are not insurmountable, and the democratic necessity is clear.

Origin

Rosanvallon developed the concept across multiple works but gave it systematic treatment in Counter-Democracy, where he traced the historical development of institutional mechanisms enabling the many to hold the few accountable. The free press was the first and most important relay—the institution that could receive individual denunciations (leaked documents, whistleblower testimony, investigative leads) and amplify them into public knowledge. The development of press freedom, protections for journalistic sources, and professional norms distinguishing journalism from propaganda were all institutional innovations responding to the democratic need for effective relays.

The twentieth century saw proliferation of specialized relays: regulatory agencies with public complaint procedures, ombudsmen offices, legislative oversight committees, civil society watchdogs. Each was a response to the growing complexity of governance—as governments and corporations grew in scale and technical sophistication, the individual citizen's capacity to monitor them directly declined, requiring institutional intermediaries performing the monitoring function. The AI transition has outpaced this institutional development. The complexity exceeds existing relays' capacity, the speed defeats their response time, and the knowledge gap makes effective translation nearly impossible without new institutional forms designed for the specific challenge of making algorithmic governance democratically accountable.

Key Ideas

Converting individual knowledge into collective accountability. Relays perform the essential democratic function of amplifying, verifying, and translating individual observations of abuse into forms democratic publics can use to exercise counter-democratic powers—without relays, denunciation remains private complaint.

Three characteristics of effective relays. Independence (not controlled by entity being monitored), authority (genuine power to investigate, publicize, demand response), and accessibility (reachable by ordinary individuals without extraordinary resources)—all three required simultaneously.

Engineer as sensor without nervous system. The AI company engineer identifying misuse risks performed counter-democratic denunciation but had no institution to receive her vigilance, investigate her claim, or translate her evaluation into accountability—technical capacity without institutional infrastructure.

Historical relay forms inadequate. Whistleblower protections weak in technology industry, mandatory reporting for AI safety incidents nonexistent in most jurisdictions, congressional hearings revealing questioners' incomprehension, free press fragmented and authority-diminished—existing mechanisms insufficient for AI's opacity and velocity.

New institutional forms required. Independent AI safety ombudsmen, mandatory incident reporting systems, standing committees with genuine technical capacity, public interest technology organizations serving as intermediaries—each facing obstacles, each necessary for democratic accountability of algorithmic governance.