Adaptive Management

In the AI Story

The Everglades restoration provides the canonical example. After decades of command-and-control water management that succeeded by engineering metrics while slowly destroying the ecosystem, the Comprehensive Everglades Restoration Plan adopted an adaptive framework. Rather than specifying a target hydrological regime, the plan established restoration goals, implemented a portfolio of interventions, and committed to monitoring outcomes and adjusting based on evidence. Slower than command-and-control. Less certain in specific timelines. More effective because it could respond to surprises — and in a system as complex as the Everglades, surprises were the dominant feature.

AI governance faces the same structural challenge. The system being governed is complex, dynamic, and operating in a phase where conditions change faster than static frameworks can accommodate. An adaptive approach would establish broadly agreed goals — broadly distributed benefit, protection against displacement-driven poverty traps, maintenance of institutional infrastructure for developing human judgment — and implement a portfolio of interventions advancing those goals, while committing to monitoring and adjustment.

Four principles characterize adaptive governance applied to the AI transition: govern for learning rather than compliance; govern across scales rather than within siloed jurisdictions; maintain diversity of approach rather than converging on a single model; and incorporate diverse knowledge, not just expert assessment. The Berkeley workplace researchers' documentation of AI-augmented work's lived reality is exactly the kind of experiential knowledge that expert panels cannot produce but adaptive governance requires.

The difficulty is that adaptive management is harder to implement than conservation-phase governance. It requires institutional arrangements capable of learning. It requires political cultures that tolerate experimentation. It requires fiscal commitments that persist across electoral cycles. These requirements are demanding. They may exceed the capacity of many existing governance systems. But the alternative — static governance applied to a release-phase phenomenon — produces either irrelevance, as the transition outpaces the governance, or harm, as frameworks designed for conditions that no longer exist produce consequences their framers never intended.

Origin

Adaptive management was developed by Holling and colleagues at the University of British Columbia in the 1970s, initially for fisheries and forestry management, and extended through decades of application to ecosystems exhibiting complex nonlinear dynamics. Its extension to social-ecological systems and eventually to technology governance came through the Resilience Alliance and related research networks.

Key Ideas

Treat interventions as experiments. Specify hypotheses; monitor outcomes; adjust based on evidence.

Govern for learning, not compliance. The framework accumulates knowledge; the static rule freezes it.

Polycentric architecture. Multiple overlapping governance bodies at different scales, connected through information flows.

Maintain diversity. A portfolio of approaches run in parallel outperforms premature convergence on a single model.

Incorporate experiential knowledge. Workers, families, and institutional insiders possess knowledge expert panels cannot substitute for.

Debates & Critiques

Critics argue that adaptive management can become an excuse for indecision — continuous experimentation as a substitute for commitment. Defenders respond that adaptive management requires clear goals and rigorous monitoring; what varies is the pathway, not the destination. A further tension concerns democratic accountability: adaptive institutions must balance the flexibility the framework requires with the predictability democratic legitimacy demands. The most defensible implementations embed adaptive mechanisms within stable normative frameworks rather than treating all parameters as contingent.