Experimentalism is Unger's systematic alternative to the premature settlement pattern that has marked every previous technological revolution—an institutional program that embeds the capacity for continuous reconstruction into governance architecture itself. Rather than allowing first arrangements to harden into naturalized permanence, experimentalist governance treats every institutional response as provisional, requires sunset provisions forcing periodic re-evaluation, supports parallel testing of multiple models, and constructs democratic channels through which experiential knowledge informs revision. This is not vague incrementalism but specific institutional design: standing governance bodies operating at technological transformation's pace, community-level assessment processes, structured comparison of organizational experiments, international networks sharing results. The AI transition demands experimentalism with unprecedented urgency because settlement is proceeding at unprecedented speed—arrangements crystallizing in months that in previous transitions took years, with democratic deliberation operating at the old tempo while the new tempo sets the formative context by default.
Experimentalist governance differs categorically from both comprehensive planning (which assumes knowledge of optimal arrangements in advance) and laissez-faire (which assumes market selection produces optimal outcomes). It assumes that optimal arrangements are unknown, that they can only be discovered through structured experimentation comparing alternatives, and that the experimentation must be democratically governed rather than left to market forces or expert prescription. The experimental method is borrowed from science—hypothesis, test, evaluation, revision—but applied to institutional design rather than natural phenomena. What science does to theories, experimentalism does to governance arrangements.
The institutional mechanisms experimentalism requires are specific and constructible. Sunset provisions: every AI governance framework includes built-in expiration, forcing democratic communities to review effects, evaluate against stated purposes, and make deliberate decisions about continuation, modification, or replacement. Parallel experimentation: rather than selecting one organizational model for universal adoption, multiple models (vector pods, integrated teams, cooperative structures, public utilities) are supported simultaneously, with systematic comparison of their effects on productivity, worker development, democratic participation, community well-being. Democratic evaluation channels: structured processes bringing experiential knowledge of parents, teachers, workers, students—those living inside arrangements—into assessment and revision processes.
The experimentalist commitment is uncomfortable because it refuses the psychological comfort of settlement. Every arrangement remains provisional. Every framework stays open to reconstruction. The governance process never reaches a finished state where institutional design is complete and democratic communities can return to low-energy maintenance. This permanent provisionality is demanding—it requires sustained institutional imagination, continuous democratic engagement, tolerance for the uncertainty that experimentation entails. But the demand is matched by necessity: in conditions of rapid transformation, the governance system that cannot experiment cannot learn, and the system that cannot learn cannot govern.
Current AI governance approaches are, with rare exceptions, anti-experimentalist. The EU AI Act settles on supply-side regulation through risk classification—a single comprehensive framework installed rather than multiple approaches tested. American executive orders establish principles assumed to be correct rather than hypotheses to be evaluated. Corporate AI Practice frameworks are adopted as settled best practices rather than provisional experiments. Educational institutions add AI literacy modules to existing structures rather than testing fundamentally different models. Each represents the premature settlement pattern: first response treated as permanent answer, foreclosing the experimentation that could reveal whether better alternatives are available.
Experimentalism has roots in American pragmatism (Dewey's experimental intelligence, James's pluralism) and was developed into a systematic governance framework by Unger across his political writings and by his collaborators Charles Sabel and Michael Dorf in their 1998 Columbia Law Review article "A Constitution of Democratic Experimentalism." The concept addresses a specific democratic deficit: existing institutions can deliberate about policies (actions within frameworks) but lack mechanisms for deliberating about frameworks themselves. Experimentalism constructs those mechanisms, making institutional design a continuous democratic practice rather than an occasional crisis response.
Provisional rather than permanent. Every institutional arrangement treated as hypothesis to be tested rather than conclusion to be enforced—sunset provisions, evaluation requirements, built-in revision capacity.
Parallel rather than winner-take-all. Multiple organizational models, governance approaches, educational frameworks tested simultaneously with systematic comparison rather than premature selection of single approach.
Democratic rather than expert-driven. Experimentation governed by affected communities with technical expertise informing rather than determining—distributed governance knowledge as essential complement to concentrated technical knowledge.
Pace matching transformation. Experimentalist governance operating at tempo adequate to technological change rather than legislative calendars—standing bodies with real-time adjustment authority under democratic mandate.