Panel about system thinking, complexity thinking - Group talk (Panel Outline)

A dense 45-minute panel hosted by Nigel Thurlow (Toyota background, “5 Whys as religion”) with Dave Snowden (complexity science / Cynefin), Diana Montalion (systems architect, author of Learning Systems Thinking), and Daniel Terhorst-North (self-described “happy consumer of both” models — around Cynefin long enough to open with “Cynefin is not a quadrant model,” systems thinking ~20+ years).

1. Systems Thinking vs. Complexity Science

1.1 Diana’s framing

• She reframes to “how would I describe systems thinking”: there’s no single agreed description (academics vs. technologists), and debating a concrete definition is itself a linearization of an inherently ambiguous/abstract topic.
• Simply: when you think not about discrete parts and engineering their behavior, but about how relationships produce effect (Ackoff), about the patterns and the asynchronous, in-motion things causing the visible experiences — you’re using a form of thinking more likely to help you engage complexity.

1.2 Dave’s framing — “the clue is in the two languages”

• “Systems thinking” vs. “complexity science” — a real distinction. Scientific metaphor: Newtonian physics isn’t invalidated by quantum mechanics, but it’s bounded; epigenetics changed Darwin without invalidating it. His contention: complexity science has done that to systems thinking — large parts of systems thinking are now invalidated, and “some people can’t escape that.”
• History of systems thinking: two foundations — general systems theory (von Bertalanffy) and cybernetics (communication & control — appeals to software people). Cybernetics splits into Ashby (foundation of modern systems thinking; information = data/signal, hence the “appalling DIKW nonsense”) vs. Bateson (whom Dave works with via Bateson’s daughter Nora — abduction, metaphor, a very different view of information). Then soft systems methodology (Peter Checkland — Dave’s origin) and critical systems thinking (Mike Jackson, Gerald Midgley — Dave holds a visiting chair).
• Complexity splits into: computational complexity (Santa Fe, heavyweight agent-based modeling — “abstractions without context,” seeking universal simulation rules; the modeler who told the US Navy “with enough money I can model the universe” — now an embarrassment) vs. anthro-complexity (different). Systems thinking is largely design without emergence (assumes you can design the system); anthro-complexity stimulates the system, manages only actants and interactions, and as things stabilize changes the energy gradients to favor what’s possible — using emergence as a response mechanism.
• The disagreement with Diana: focus on interactions/actants and scaffolding vs. focus on the individual/actor — the individual focus seen as culturally specific to North America and Northern Europe.

1.3 Daniel’s framing — “attractors, not boxes”

• Systems thinking à la Donella Meadows: “the relationships between things are generally much more interesting than the things.” A system of parts has characteristics — some emergent (unpredictable from the parts), some steady state, some fully reducible (a car’s engine does “engine stuff” predictably; bigger engine → more power).
• Complexity science gives him a vocabulary to reason differently about the emergent parts vs. the stable/predictable ones. His ~10-years-ago epiphany (with Dave): Cynefin domains (complex/complicated/clear/chaotic) are attractors — aspects of a thing, not categories you file things into. Any non-trivial system of work has overlapping characteristics — “enormously liberating.”

1.4 Diana’s rebuttal & the archetypes debate

• She “hates the phrase systems thinking” — her book was nearly titled “Mind Shift.” She does not mean design without emergence; she writes for individuals whose thinking makes engaging complexity hard (a “starter way”; her top online course is paradoxically “Systems Thinking Made Simple”). Defining these streams makes it harder to define what we’re trying to do.
• Meadows’ archetypes describe types of emergence (e.g., “shifting the burden to the intervenor” = essentially addiction as a system property, with no single part causing it).
• Dave’s critique: Meadows/Senge/Forrester (system dynamics) recognize feedback loops but limit their number — can’t account for the variables in a complex adaptive system — and turn archetypes into categories people are forced into. Critical-systems-thinking/cybernetics people reject system dynamics and get angry at the association. Respect for Meadows in her context, “less respect now”; “leverage points” wrongly imply you can engineer the system.
• Dave’s book-title jab: Diana’s book “makes things simple, [but] it doesn’t make systems thinking simple” — Midgley/Jackson would call it simplistic (negative); “your publishers did you no favor with the title” (“except it sold more books”). Both agree: the rungs get you on the journey, then it must evolve (like agile) — build on prior work rather than “Newtonian-ize” it. Dave’s phrase: “practice is theory-informed practice” (a nod to his “revolutionary left” roots).

2. Root Cause & Reductionism

• Dave: complexity science gives you “a scientific understanding of common sense.” Agile’s flaws: the “empirical heresy” (agile people misunderstand empiricism — it’s hypothesis validation, not “what worked on my last three projects” or SAFe’s “half-remembered things codified into a method”).
• Reductionism done right: in a complex adaptive system you must break things down to the lowest level of coherent granularity; the error is assuming the whole can be derived from the parts.
• Nigel’s Toyota challenge: “most people can’t do true root-cause analysis.” How do you root-cause a complex problem?
  • Daniel: “root cause” sounds like a noun/thing, but he looks at what in the structure/characteristics of the system of work made this inevitable — and what unintended consequences a fix would introduce. Google’s massive public outage’s “root cause”: a mudslide destroyed one data center + a fire engulfed another + (probably a shark) severed an undersea cable simultaneously → grown-up answer: “we’ll take those odds.”
  • Diana: don’t assume there is a knowable root cause — what you can understand depends on context and what you already look for; “the call is coming from inside the house” (the problem is inherent in what we value/believe/are certain about); fixing one problem makes new things start occurring. She can “only help form and reform the question.”
  • Dave: in a complex adaptive system there is no linear material causality, so root-cause hunting is a waste of time and produces retrospective coherence (system gets worse). Root causes exist in constrained/ordered manufacturing systems (Goldratt’s Theory of Constraints, “old-style physics” complexity — many parts but mappable). Management methods start in manufacturing (ordered/predictable) then fail in services (complex/unpredictable) — why he devised Cynefin (unpopular with complexity purists who think everything is complex; Cynefin says some things are ordered, and humans can create structure/order — via constructors/scaffolding, distributed micro feedback loops Meadows couldn’t account for). Agile went wrong using “the language of complexity but not the practice.”
• The proposed boundary: parts of systems thinking (system dynamics, cybernetics) only work in constrained environments; soft systems works unconstrained but depends on facilitation → doesn’t scale. Dave’s epistemic-justice work has everyone tell and interpret their own story quantitatively at scale (no facilitator) — revealing how language changed over centuries.

3. Ashby’s Law & the Problem of Language

• Ashby’s law of requisite variety: if a system’s internal variety < its environment’s, the environment dominates and destabilizes it.
• Diana: start with defining variety — e.g., 92% of developers are male (2022), a diversity problem for how we think/experience the work. Language is the challenge: trying to describe what she sees as invisible doesn’t fit existing vocabularies/frameworks → we end up debating language. Reductionism, predictability, command-and-control fail us even at agreeing what “system”/“complexity” mean or what’s okay to bring into the conversation — so finding space for sufficient diversity of thinking is a real challenge.
• Dave: works in an 80% female company (different dynamics, “more vicious politics,” alpha-male/alpha-female behavior); feminist philosophy uses new language (new materialism, epistemic justice) for new concepts. Ashby is not a “law of gravity” — it builds on Shannon’s restricted view of information (brilliant but limited); human info flow (pheromones for trust; constructor/assembly theory; the theoretical-physics claim that information has momentum → new materialism’s “attractor wells,” e.g. Trump/Brexit) exceeds it. Post-9/11 counter-terror work (with Boisot) faced symmetric governments out-pointed by asymmetric threat — you reverse Ashby’s law by using citizens as sensor networks. Accepting only Shannon/Ashby’s information theory holds back software’s value.
• Language solidifies into dogma — Houston Smith: “into every religion, when the prophet dies, comes everything they abhorred.” The “Constantine moment” (Catholicism became a state religion; and the near-miss where Jesuits almost merged Confucianism + Catholicism in China before the Dominicans stopped it) — insights born in imperfect language/metaphor calcify into fixed meaning. Dave has already moved on from “complexity science” as a term because it now connotes agent-based modeling.
• Daniel: requisite variety “feels very one-dimensional” — his counter is “people are messy,” so you need a more nuanced, informed model, not a blanket overlay.
• The abductive/inductive key point (echoing Simon Wardley’s keynote): humans reason abductively, AI reasons inductively, yet most software is written for inductive reasoning and not to enable abductive reasoning.

4. Closing — AI Is Complication, Not Complexity

• Nigel: will AI make the world more complex? Advice for leaders?
  • Diana: she’d call it complication, not complexity. The “source of truth” is in motion, interaction, asynchronicity over time in context; AI is about fast perspective, not inference/relation — is it doing anything effective or just efficient?
  • Dave: “you stole my language — I wrote down ‘AI isn’t making things complex, it’s making them complicated’” — and dependent on algorithms. New program (US through November, UK from April) to measure abductive capability in an organization via its interactions → an executive early-warning: are you letting AI do things it does better while humans fail to develop higher capacity, and is the balance right? His 20-year-old phrase (often stolen unattributed): “human systems are messily coherent” — the coherence matters and the mess matters.
  • Daniel: chess study — real-time MRI shows a grandmaster isn’t scanning future moves; they “remember a game that never existed” (recall, a much smaller solution space, less effort) — the abductive thing GenAI can’t do (GenAI is brilliant at scanning but doesn’t know the right move). Rant — follow the money: AI is a rich field (ML, image recognition, stochastics, neural nets, DNA computing) now underfunded because money chases LLMs — “a really useful translation tool at scale” that is “always hallucinating (bullshitting)” (“hallucinating” wrongly implies cognition), sometimes usefully. He’d love machine learning / computer-aided computing to get attention back from GenAI hype and “terrified VC investors.”

People & References Cited

• Panel: Dave Snowden (Cynefin/complexity science), Diana Montalion (Learning Systems Thinking), Daniel Terhorst-North, host Nigel Thurlow (Toyota / The Flow System).
• Von Bertalanffy (general systems theory), Ashby, Bateson (& daughter Nora), Peter Checkland (soft systems), Mike Jackson / Gerald Midgley (critical systems thinking), Donella Meadows / Peter Senge / Jay Forrester (system dynamics), Russell Ackoff, Goldratt (Theory of Constraints), Shannon (information theory), Boisot (counter-terror work), Houston Smith, Simon Wardley (keynote), Dreyfus brothers (chess/expertise).
• Concepts: systems thinking vs. complexity science, actants/interactions/scaffolding, emergence as response, Cynefin (aparetic domain), archetypes-as-categories critique, retrospective coherence, requisite variety, new materialism / epistemic justice, abductive vs. inductive reasoning, “complication not complexity.”

Video: https://www.youtube.com/watch?v=45U1E7SXT7E — Transcript via yt-transcript.sh; outline generated from the transcript.