In 1973, C.S. Holling, a Canadian ecologist, published Resilience and Stability of Ecological Systems, introducing a distinction foundational to systems thinking. Stability is the system's tendency to return to equilibrium after a small disturbance; resilience is its capacity to absorb disturbance and reorganize while maintaining function. A system can be stable but fragile — returning to equilibrium under small disturbance, then flipping into a different regime past a threshold. The distinction matters because engineered systems are often optimized for stability in ways that reduce resilience; industrial agriculture is the canonical example, high-yielding under normal conditions and dramatically vulnerable outside its design envelope. Forty years later Nassim Nicholas Taleb's Antifragile: Things That Gain from Disorder (2012) extended the dichotomy — fragile systems break under stress, robust systems resist it, and antifragile systems gain from it.
Holling distinguished three dynamical regimes. Engineering resilience — a single stable equilibrium the system returns quickly to after disturbance — is the classical engineer's definition, useful but limited. Ecological resilience, Holling's contribution, recognizes that a system can sit in any of multiple stable states, and the relevant question is the size of the basin of attraction: how large a disturbance can be before the system flips to a different regime. Lake eutrophication is the canonical case — a clear lake can absorb some nutrient loading and stay clear, but past a threshold it flips to a turbid, algae-dominated state that is also stable and difficult to reverse. Adaptive resilience is the most powerful: the system's ability to reorganize its structure under disturbance, evolving into something qualitatively new. The key insight is that high engineering resilience often comes at the cost of ecological resilience: tightly-coupled financial systems engineered for efficiency are more fragile to liquidity shock than slack-rich ones, and just-in-time supply chains are more fragile to disruption than warehouse-buffered ones, a lesson re-learned during COVID-19. Taleb's antifragility extends the framework with a triad: fragile systems break under stress (glass, planned economies), robust systems resist it (steel, large-cap balance sheets), and antifragile systems gain from stress, volatility, and disorder. The immune system is strengthened by challenges; muscles grow under stress; evolution uses mortality to improve the population. The argument is that much of the world is non-linear in stress, and the right strategic question is often not how to become robust but how to become antifragile. The framework is most useful for naming the systematic underestimation of tail risk; the standard caveats survive — antifragility is hard to measure in advance and often only diagnosed post-shock, many apparently antifragile systems are antifragile up to a point and fragile beyond it.
Climate policy is increasingly framed in resilience terms — adaptation is the complement to mitigation, and the engineering-resilience approach of building for current climate and defending it is itself fragile to climate change beyond the design envelope. COVID-19 was a stress test of resilience across multiple systems at once: healthcare with insufficient surge capacity, supply chains with just-in-time fragility, financial systems largely robust thanks to post-2008 reforms. AI safety runs the same diagnostic — whether systems are fragile (break under adversarial input), robust (handle it), or antifragile (improve from it) is a live engineering question. When analyzing any system under stress, the useful question is not is this stable but what kind of resilience does it have: engineering resilience (returns to normal) is one thing, ecological resilience (how large a shock crosses the threshold) is more useful, and antifragility is the most useful diagnostic.