In 1984, people who no longer wanted problems sorted by department founded an institute in New Mexico with no departments
Universities are organised into buildings by discipline, and there is a class of problems that refuses to sit in any of them: how an economy grows cycles out of countless individual trades, how an immune system recognises a virus it has never met with nobody in charge, how a language replaces its own grammar with no one legislating. What these share is not subject matter but structure — many interacting parts producing collective behaviour that is not present in any single part.
In 1984 a group of senior researchers around Los Alamos National Laboratory (the convener was the chemist George Cowan; the early core included the physicists Murray Gell-Mann and Philip Anderson and the economist Kenneth Arrow) decided to build a house for exactly those problems: the Santa Fe Institute, universally abbreviated SFI.
Its intellectual colour traces back to Anderson's 1972 essay More Is Different: each new level brings new regularities, and those regularities cannot be derived from the equations one level down — so the programme of "finish physics and you have explained everything" breaks at every level boundary. If that is right, someone has to work on the boundaries themselves.
Rule 3 is the real device. The usual way an interdisciplinary conversation dies is that everyone trades metaphors and nobody can falsify anyone. A model written as code forces each participant to hand over something checkable — what the rules are, what comes out, and which step fails to match reality. The model's job here is less to explain the world than to give people from different fields a shared object to argue about.
In 1987 SFI locked ten physicists and ten economists into the same ten-day workshop; the proceedings became The Economy as an Evolving Complex System (1988). One consequence was a modelling line that replaced the omniscient rational agent with agents that learn — the artificial stock market of Arthur, Holland, LeBaron, Palmer and Tayler came out of it: give every trader their own set of forecasting rules and let profit and loss retire the bad ones, and prices spontaneously show the clustered volatility and self-fulfilling technical trading of real markets.
There is also Holland's complex-adaptive-systems framework and genetic algorithms → ref · genetic algorithms, Kauffman's autocatalytic sets and NK landscapes, West's scaling laws for metabolism and cities, and Mitchell's book, now a standard entry point. Plus one by-product: the self-identification "complexity science" itself — Waldrop's 1992 bestseller Complexity wrote this group up as a school, and the name stuck.
SFI is not a model and explains no natural phenomenon. What it explains is an episode in the history of science: why the scattered tools of power laws, networks, criticality and adaptation coalesced into a self-declared field in the late 1980s instead of staying in their separate departments.
Part of the answer is unromantic: who was in the room, and what craft they brought. The founders were largely physicists out of the weapons laboratories, holding computing resources that were rare at the time and a physicist's habit of seizing the dominant mechanism and discarding detail; economics and biology supplied problems that would not behave. That combination fixed which questions the field asked first — it favours mechanisms a minimal model can catch, and has correspondingly been weak on institutions, culture and meaning, which decline to be caught that way. Knowing this explains where complexity science is sharp and where it is blunt.