Competence versus consent is the oldest unresolved tension in democratic theory, and Rosanvallon's work makes it the central problematic of AI governance. Competence is real—the physician understands the body better than the patient, the engineer understands the reactor better than the community living downstream, the AI researcher understands transformer architecture better than the users whose work it transforms. Competence creates genuine power: the ability to shape outcomes, make consequential decisions, determine the conditions under which others live and work. But in democratic systems, competence alone does not create legitimate authority. Legitimate authority requires consent—the agreement of the governed that those who exercise power do so rightfully. The Abbé Sieyès's 1789 question—'What is the Third Estate?'—was a claim about legitimacy, not competence. He did not argue the people were competent to govern; he argued no one else had the right to govern without their consent. That distinction between competence and legitimacy restructured European civilization and remains the fault line on which every democratic governance crisis turns.
The AI priesthood—builders who understand systems from inside and claim authority as stewards—faces this crisis in its most acute form. The Orange Pill proposes that those who understand the river should tend the dam, and the proposal rests on genuine insight: someone must understand the current, someone must know where the leverage points are, someone must possess the technical capacity to build structures that hold. But understanding confers obligation, not authority. The obligation is to make the knowledge available for democratic decision-making. The authority to decide where the dam goes must come from the people who swim in the pool, through legitimate democratic processes, not from the engineers' superior understanding of hydrology.
Every priesthood in democratic history has justified autonomy on identical grounds: the work is too complex for popular oversight, stakes too high for amateur interference, those who understand the system are better positioned to govern it than those who merely live inside its effects. Central bankers said this about monetary policy; citizens eventually demanded and received legislative oversight that did not require legislators to be economists but did require central banks to justify their decisions in democratic terms. Nuclear engineers said this about reactor safety; communities eventually constructed regulatory frameworks that did not require citizens to understand physics but did require engineers to demonstrate their safety assessments to public satisfaction. Intelligence agencies said this about national security; democratic societies eventually built oversight mechanisms that did not require legislators to be intelligence professionals but did require agencies to operate within legal boundaries subject to review.
In every case, the democratic solution was not to make the public equally competent but to build institutions translating expertise into accountability. Medical licensing boards with public members do not make the public capable of performing surgery—they make physicians accountable to standards the public helped set. Judicial review processes do not make citizens capable of arguing case law—they make judicial decisions subject to appeal through processes citizens can access. Legislative oversight of central banks does not make politicians capable of conducting monetary policy—it makes central bankers justify their decisions in terms politicians, representing the public, can evaluate. The pattern is consistent: expertise serves, through institutions that translate competence into democratic accountability.
The AI context adds specific difficulties. The knowledge gap is deeper—understanding transformer architecture requires years of specialized training, unlike understanding monetary policy, which can be grasped at operational level by any mathematically literate adult. The velocity is faster—AI systems evolve weekly while oversight institutions operate yearly. The opacity is structural—even AI researchers cannot fully explain how large language models produce specific outputs, unlike central bankers who can articulate their reasoning. These difficulties are real. They do not change the democratic principle: competence creates power, consent creates legitimacy, and power without legitimacy is coercion regardless of how well-intentioned or technically sophisticated. The appropriate democratic response is not to abandon oversight as impossible but to invent institutions adequate to the challenge—institutions that can translate AI expertise into democratic accountability despite the depth of the knowledge gap, the speed of technological change, and the structural opacity of the systems themselves.
The distinction has ancient roots—Plato's Republic is an extended argument for rule by the competent (philosopher-kings), Aristotle's Politics a sustained insistence that legitimate governance requires the consent of free and equal citizens. The modern democratic settlement, as Rosanvallon traces it, rejected Platonic competence-based authority in favor of Aristotelian consent-based legitimacy while recognizing that modern governance's complexity requires expertise. The settlement is: expertise informs, democracy decides. Experts serve as advisors, translators, executors of democratically determined policy—but the authority to set policy rests with citizens or their representatives, not with the experts themselves.
Rosanvallon's contribution was to show how this settlement has been continuously renegotiated through institutional innovation as new forms of expertise emerge. Each wave of specialized knowledge—medical, legal, economic, technological—has required new institutional mechanisms subjecting expertise to democratic accountability: professional licensing, judicial review, legislative oversight, regulatory agencies with public participation procedures. The AI wave is the latest and most challenging iteration. The question is whether democratic societies will invent the institutions this challenge requires or whether the knowledge gap will prove so vast that competence-based authority becomes the de facto settlement, with democratic legitimacy reduced to rhetorical decoration on decisions made by technical elites.
Expertise creates power, not authority. Competence gives genuine power to shape outcomes and make consequential decisions, but in democracies legitimate authority requires consent of the governed—agreement that those exercising power do so rightfully, a requirement competence alone cannot satisfy.
Democratic pattern of resolution. Every specialized knowledge domain—medical, legal, economic—eventually subjected to democratic accountability through institutional invention: licensing boards, judicial review, legislative oversight—expertise serves through institutions translating competence into accountability.
AI knowledge gap unprecedented. Deeper (understanding transformers requires years of specialized training), faster (systems evolve weekly), more opaque (even researchers cannot fully explain outputs)—but difficulties do not change democratic principle that power without legitimacy is coercion.
Expertise informs, democracy decides. The modern democratic settlement: experts serve as advisors, translators, executors of democratically determined policy, but authority to set policy rests with citizens or representatives—not with experts themselves, however superior their knowledge.
Translation institutions required. Medical boards do not make public capable of surgery but make physicians accountable to public-set standards; judicial review does not make citizens lawyers but makes judicial decisions appealable; legislative oversight does not make politicians economists but makes central bankers justify decisions in evaluable terms—pattern applicable to AI.