Excerpt

This talk was the keynote for SFI’s 2019 Applied Complexity Network Symposium, held post-cocktails on the evening of the first day. Complexity economics was very much birthed at the Santa Fe Institute in the late 1980s and the 1990s, and indeed for several years it was known as Santa Fe economics. This talk is an informal attempt to tell how complexity economics got its start at SFI, and to make sense of the logic of its approach. I have left out many things about modeling and people and early agent-based work in our program, and our early probability approaches. More details are given in other accounts. But I want to emphasize here —which I didn’t in the talk—that SFI’s ideas arose in a context. They had precursors in the work of Peter Allen and Ilya Prigogine of the Brussels group, nonlinear dynamics, nonlinear stochastic processes, and in theories of induction. The Ann Arbor group directly influenced us through John Holland. 

I’m struck that so much has happened since these early days at SFI, both within SFI and outside. Rob Axtell and Josh Epstein introduced their “generative social science.” Network theory took off. Agent-based computational economics took off. Models that use machine learning and AI to simulate agent behavior have shown up. My own interests turned to theorizing about how technology comes to be, influencing how economies form and re-structure themselves. Some of these new threads were reflected at the symposium, some were not. 

Whatever was included in these symposium talks, complexity economics has now become a large field. And the Santa Fe Institute—particularly Kenneth Arrow, Philip Anderson, and David Pines—deserves much credit for early on backing this work. I am surprised that our small program, “The Economy as an Evolving Complex System,” has come to have so much influence. It was SFI’s first research program, and we were aware—at best vaguely aware—that we stood some chance of doing economics differently. I also think the times deserve much credit. Economics changes with change in technique. And just as algebra and calculus entered economics in the 1870s and brought in neoclassical economics, in the 1980s we all got desktop computers and that helped birth this new approach. We could suddenly allow ourselves to get interested in more complicated models—computation could handle complication. 

In fact, since this talk was given I am struck that computation has gone from being slightly outré to being now highly fashionable. There’s been an enormous burgeoning of agent-based computational economics (ACE) models in the last few years. This is healthy, I believe. In good hands, the sort of detail computation allows makes for more realistic modeling, and we are seeing this currently in coronavirus models. The old equation-based models with buckets of Susceptibles, Infecteds, and Recovereds are too coarse, and computation allows detail on how precisely infection spreads and how it’s related to the niceties of economic arrangements. So we are learning a lot from models that allow more detail.

BACK TO Complexity Economics