Beyond Panaceas

In the [YOU] on AI Field Guide

The cycle’s core methodological commitment—to hold the complexity of the AI moment without collapsing it into premature resolution—is the practical expression of Ostrom’s anti-panacea principle. The triumphalist who celebrates AI as pure liberation and the alarmist who condemns it as pure catastrophe are both reaching for panaceas: clean resolutions that substitute ideological comfort for the honest difficulty of diagnosis. The silent middle that the cycle identifies—those who hold both truths simultaneously—is the position that Ostrom’s diagnostic approach demands.

In the specific domain of AI governance, the principle challenges every faction in the debate. Just make it open—open-source everything and transparency will solve the problem—is a panacea: a single solution proposed for a problem whose defining feature is its diversity. Just regulate it—pass the right law and the harms will be contained—is a panacea: a single authority presuming to govern what no single authority can know. Just align it—solve the technical alignment problem and governance becomes unnecessary—is a panacea: a technical fix proposed for what is fundamentally an institutional and political challenge. Ostrom’s diagnostic alternative asks: what kind of resource is this specific AI application, at what scale does this specific harm live, in what community, with what available institutional repertoire? The answers differ by case, and so should the governance responses.

Origin

The explicit formulation of the anti-panacea principle came in a 2007 paper titled “A Diagnostic Approach for Going Beyond Panaceas,” co-authored with Marco Janssen and John Anderies, and was consolidated in Ostrom’s 2009 Nobel lecture. The paper identified two dominant panaceas in environmental governance: privatization (give individuals property rights and let markets allocate) and centralized state management (impose rules from above and enforce them through bureaucracy). Both had been applied wholesale across radically different contexts, often producing the opposite of their intended effects. The solution Ostrom proposed was not a third panacea but a method: a systematic diagnostic approach that maps the features of the resource, the community, and the institutional environment before prescribing anything.

The principle had a self-reflexive moment that Ostrom acknowledged with characteristic honesty. She came to regret that the term “design principles” had confused many readers into treating the eight principles as a checklist rather than a diagnostic lens. In a footnote to her Nobel lecture she confessed she might have used “best practices” instead, precisely because “design principle” sounded too much like a blueprint. A framework that applies its own central warning to itself—insisting that even the design principles must not become a panacea—is the rarest kind of intellectual achievement: a theory that means it.

Key Ideas

The appeal of elegance over truth. The deeper reason panaceas are seductive is not merely practical but aesthetic: a single clean solution is beautiful, and the messy plurality of diagnostic, context-specific governance is ugly by comparison. Ostrom understood this temptation and refused it on principle. The rejection of complexity in favor of the elegant model is not rigor. It is a failure of nerve dressed up as parsimony. The AI commons is genuinely complex, and a governance theory that flatters us with simplicity is lying.

Diagnosis as the alternative. The diagnostic method Ostrom developed asks a structured set of questions about any governance problem: what are the attributes of the resource system; what are the attributes of the resource units; what is the governance system; what are the attributes of the users; and how do these interact to produce outcomes? Applied to AI, this means refusing to treat “AI governance” as a single problem with a single answer, and instead asking: which AI application, which harms, which communities, which institutional capacities, at which scale? The different answers to these questions point toward genuinely different institutional responses—the data commons governance problem is not the compute-concentration problem, which is not the catastrophic-misuse problem, which is not the model-fairness problem.

The panacea test. A useful heuristic for evaluating AI governance proposals: if the proposal would govern all AI applications in all contexts through the same mechanism, it is probably a panacea and should be held with suspicion proportional to its universality. The more an approach insists on its universal applicability, the more likely it is that the insistence substitutes confidence for diagnosis. This is not an argument against coordination or common standards—which are themselves context-appropriate responses to specific problems of interoperability and catastrophic risk—but an argument against the reflex to reach for the single solution before completing the diagnostic work that would reveal which solutions are actually matched to which problems.