The Field Guide › Paper
Holland's Daedalus essay is where complex adaptive systems acquires its name and its canonical characterisation, and it anchors the systems-science half of the complexity-foundations set alongside Anderson's emergence and Simon's hierarchy. Holland, a computer scientist at Michigan and the Santa Fe Institute (and the inventor of the genetic algorithm), starts from a puzzle about simulation: aircraft wings and bridges yield to equation-based modelling, but economies, ecologies, immune systems, developing embryos and brains do not, because each has an evolving structure that continually reorganises its own parts to adapt to its surroundings, making it a moving target rather than a fixed object. What these otherwise unlike systems share, he argues, is a common kernel, and he pins it to three properties. They evolve: the parts adapt or learn in Darwinian fashion, which he calls the pivotal characteristic. They exhibit aggregate behaviour that is not simply the sum of the parts but emerges from their interactions (the immune system's ability to tell self from other, an economy's supply-and-demand network). And they anticipate: the parts carry internal models that let them act on expected outcomes, so that expectations reshape behaviour even when they never come true (the anticipated oil shortage that raises prices on its own). Holland singles out this last property, the forming and using of internal models, as the fundamental attribute that separates complex adaptive systems from merely complicated ones. A second, load-bearing claim is that such systems have no single governing equation and little or no central control: behaviour is distributed across many parts each following its own condition/action rules, adapted over time by credit assignment (rewarding rules that contribute to good performance) and rule discovery (recombining useful building blocks, the logic of his genetic algorithms), with the system perpetually balancing exploration against exploitation. Critically, complex adaptive systems operate far from equilibrium and never reach a stable optimum; it is the process of becoming, not any end-state, that has to be studied, which is why the equilibrium-based mathematics of linear systems, fixed points and attractors is of little help. For a corpus about organisations this is the source text for the now-routine claim that an organisation is a complex adaptive system: it grounds the probe-and-adapt stance, the primacy of emergence over central design, and the treatment of change as continual adaptation rather than the execution of a plan (the intuition Cynefin, the adaptive cycle and the corpus's own complexity articles all draw on). Its framing should be read critically, though. Holland's account is thoroughly computational, built from rule-following agents, genetic algorithms and massively parallel simulation, and its transfer to human organisations runs through an analogy that can quietly re-mechanise the very thing it means to de-mechanise, recasting people as agents executing and recombining rules; its optimism about policy makers flight-testing societies on a simulator is very much of its 1992 moment; and its vocabulary of adaptation and fitness is easily co-opted into exactly the managerial technique a complexity-informed practice should resist. (Text drawn from the 1992 Daedalus essay, 121(1), pp. 17–30, released under CC BY-NC 4.0.)