Surviving the complexity winter

Complexity science in 2024: claiming to be what it has never been, but gladly giving away its Nobel prize. This is a reflection about where the science of complex systems stands today, and where it could and should go next. Cover painting by Isaac Levitan.

I’m as excited as everyone else that Hopfield and Hinton got the Nobel Prize in physics. The biological foundation of computation has been a part of many of the “holistic tribes”—the disconnected scientific movements of the 20th century centered around explaining systems by zooming out rather than in. It would be entirely in order to uncork another bottle of Krug and party like it’s 2021 all over again. But, oh no. My complexity colleagues unanimously call this “a prize to AI” and join in the discussion on whether AI is physics or not.

The Nobel part of the story doesn’t end there, as the Nobel Foundation was kind enough to give AI another round of awards—in chemistry. The striking thing is that the chemistry prize, at least Hassabis and Jumper’s, is a brilliant example of AI that is not complexity science: all prediction, all black box—very cool and well deserved, but not complexity science. AI and complexity science overlap, but one does not contain the other.

I also caught the AI bug, and many of my ongoing projects try to find nifty uses of LLMs to ultimately learn about people. I also have some (admittedly less interesting) let’s-explore-the-capabilities-of-AI-by-letting-them-do-something-that-we-know-how-humans-do research. I would never dream of calling that complexity science, but leafing through the abstracts of recent complex systems conferences, I see many such studies—cool science, but hardly complexity science.

Another sign of the times is that many of the core ideas of complexity science of the 90s have been absorbed back into the traditional academic disciplines. I recently chatted with a management scholar who pointed out how ideas from complexity science—systems performing optimally at the edge of chaos, etc.—have been integrated, textbook-ized, and (hmm) gentrified to the point nobody sees a need to look to complexity science for new ideas.

Further evidence of how systems-related topics have partly fallen out of fashion and partly merged with other fields of knowledge is their fading presence in public political discourse. In the wake of Limits to Growth, governments appointed systems-science advisors, founded systemsy institutes (like The Institute for Futures Studies that once paid part of my salary), etc. Even the Vatican held complexity science conferences. Not much such stuff these days.

This is not the end, however. Just like the famous AI winters of the past, this complexity winter will give way to another spring. The reason I’m fairly sure of this is that the desire to understand is a fundamental human trait—”Nothing seems to be more prominent about human life than its wanting to understand all and put everything together” as Bucky Fuller articulated it in his Operating Manual for Spaceship Earth. Ultimately, we can’t do that without the most fundamental and radical of complexity science ideas—that causality is not the atom of scientific explanation; we need explanation in the form of models, systems diagrams, systems thinking, etc., feedback loops included. For a moment, we might be infatuated by causal inference and black-box prediction, but that crush will not last forever. (And for the record, I’m not a prediction-is-not-understanding memer, almost the opposite (which I should also write (at least) a blog post about).)

To end on a high note. I just finished teaching complex systems for the first time in ten years, and I can tell many students really liked the material. Instead of following the prototypical syllabus—essentially a storyline and an accompanying skill set that were ready in place in Hermann Haken’s 1981 Synergetics—I focused only on the zoo of ideas and frameworks and their different takes on the pillars of complexity science, holism, emergence, etc. Those ideas will never die . . or at least never not be reborn. The complexity science we have today could maybe die, though: by presenting itself as something popular that it isn’t and then being swallowed by that (two-lettered) something.