Lovely, clear explanation. When considering organisations, I think in terms of agility v stability (instead of damping and sensitivity) as the key trade-off. I like SD, but I’m not sure it’s as all embracing as its exponents believe - other methods have their place, in my view, too. I’m trying to get my head around this and Jonathan Klein (Soton) and I are grappling with why and where the top end of systems engineering gives way to soft systems and social science.
This article crystallised part of this for me, something I had been puzzling over, which is that the long term structure also changes - sometimes in response to the feedback loops themselves - and this, SD models struggle to address.
Another thing is the failure of tekkie types to turn their insights into policy. Famously, just as Stafford Beer felt he was about to get the Chilean economy under control by throwing out yet about control loop, his champion, Allende, was overthrown. In the many recent obits and comments on Alan Greenspan’s tenure at the Fed, there is his repeated concern that his model of reality (and thus the need for more or fewer feedback mechanisms of greater or lesser strength) was faulty. The Times obit states: ‘In an interview with the Financial Times in 2013 when Greenspan published The Map and the Territory, a reflection on his changing approach in the light of the crisis, he spoke of the devastating impact of 2008 on everything he had stood for. “The whole period upset my view of how the world worked,” he said. “The models failed at a time when we needed them most … and the failure was uniform.”’
What I’d like to see next is you applying your 10+ rules to two or three crises this century (global financial, covid, energy, the PO scandal, universities…?) and see what sort of pattern emerges as your template is applied.
Yes. This lands, yet left me wondering whether the 10+1 each has its own domain of validity. I re-read the post, using the Cynefin framework as a lens. Most lessons look ‘complicated-domain valid’ (assume something stable enough to model). The +1 (steered, not controlled) feels most valid in a complex system. My recency bias is from working with leaders about agents and agency (self driving things). So while Goodhart barely bites when you're metering inventory (7), it's brutal the moment the measured thing can read the meter (hello AI agent). It would, be interesting to see a map of the 10+1 by domains in which each is strongly or weakly load-bearing
Very timely. I think the obvious lack of understanding of dynamics, and especially of possible spontaneous acute loss of stability (as latent parameters move slowly, chronically, over hidden thresholds) is something that is not really considerd by various natiopnal authorities (who should know better).
Here is a recent sumary, and it includes the rather sad history of catastrophe theory, which was later reinvented as "path dependence" in Sante Fe.
I think, as someone who's lived and practiced this stuff for many years, the thing that matters is not the theory but the practice.
Somehow we need to get under the skin of our institutions, their actors and (most importantly) the boardroom. Somehow we need to use technology intelligently (and not be diverted by "intelligent technology"!), and somehow we should pass this all - knowledge and practice - on to our students.
Cleanest short statement of the tradition I have read in a while, and the bonus lesson is the one most people skip.
The thing I keep circling is that this machinery tells you how a system moves but not whether it can afford to keep moving. Structure drives behavior, agreed. But two systems with identical structure can sit on opposite sides of viable: one replenishing the stocks it draws on, the other quietly liquidating them. So I find myself wanting a verdict on top of the description. At what cost is this behavior being held, and for how long, before the stocks it depends on run out.
Your lesson 9 is where it bites. The tradeoffs are not only design constraints. When a system optimizes its visible target by spending the buffers that look like waste under normal conditions, you get rising metrics and accumulating fragility at the same time, and the description reads as success right up to the threshold in lesson 10. The hardest part is that nothing in the measured behavior tells you which side of viable you are on.
Most systems are comprised of human beings, at some level. And they behave in ways that are sometimes not intelligible in terms of the system in which they are imbedded. This is the agent structure debate, and whilst structures are clearly (IMO) more important, human agents create and shape those structures: their logics are constructed and are constantly being remade.
This is very elegant. I’m taken back to my early teens when I spent a couple of years fiddling with electronic components trying to do hobby projects. Trying to understand why they didn’t work was in a way more interesting than playing with them when they did. (Then I decided to stick to maths - equations behave themselves, transistors don’t).
Excellent exposition! Do I detect some of Ludwik's thinking? (Hardly surprising, given he influenced the ideas of most people he fell into conversation with!) Best wishes.
Very helpful analytical framework. I’m an economist who has worked mostly on emerging economies. Your analysis is a very helpful reminder that recommendations, even if fully taken, don’t have straightforward and predictable consequences.
Lovely, clear explanation. When considering organisations, I think in terms of agility v stability (instead of damping and sensitivity) as the key trade-off. I like SD, but I’m not sure it’s as all embracing as its exponents believe - other methods have their place, in my view, too. I’m trying to get my head around this and Jonathan Klein (Soton) and I are grappling with why and where the top end of systems engineering gives way to soft systems and social science.
This article crystallised part of this for me, something I had been puzzling over, which is that the long term structure also changes - sometimes in response to the feedback loops themselves - and this, SD models struggle to address.
Another thing is the failure of tekkie types to turn their insights into policy. Famously, just as Stafford Beer felt he was about to get the Chilean economy under control by throwing out yet about control loop, his champion, Allende, was overthrown. In the many recent obits and comments on Alan Greenspan’s tenure at the Fed, there is his repeated concern that his model of reality (and thus the need for more or fewer feedback mechanisms of greater or lesser strength) was faulty. The Times obit states: ‘In an interview with the Financial Times in 2013 when Greenspan published The Map and the Territory, a reflection on his changing approach in the light of the crisis, he spoke of the devastating impact of 2008 on everything he had stood for. “The whole period upset my view of how the world worked,” he said. “The models failed at a time when we needed them most … and the failure was uniform.”’
What I’d like to see next is you applying your 10+ rules to two or three crises this century (global financial, covid, energy, the PO scandal, universities…?) and see what sort of pattern emerges as your template is applied.
Thoughtful as well as serious. Thank you!
Yes. This lands, yet left me wondering whether the 10+1 each has its own domain of validity. I re-read the post, using the Cynefin framework as a lens. Most lessons look ‘complicated-domain valid’ (assume something stable enough to model). The +1 (steered, not controlled) feels most valid in a complex system. My recency bias is from working with leaders about agents and agency (self driving things). So while Goodhart barely bites when you're metering inventory (7), it's brutal the moment the measured thing can read the meter (hello AI agent). It would, be interesting to see a map of the 10+1 by domains in which each is strongly or weakly load-bearing
Very timely. I think the obvious lack of understanding of dynamics, and especially of possible spontaneous acute loss of stability (as latent parameters move slowly, chronically, over hidden thresholds) is something that is not really considerd by various natiopnal authorities (who should know better).
Here is a recent sumary, and it includes the rather sad history of catastrophe theory, which was later reinvented as "path dependence" in Sante Fe.
P. Grindrod , Resilience, Tipping Points, and Hysteresis, Complexities, 2026 https://www.mdpi.com/3042-6448/2/2/10/pdf
This is really good to see - thank you!
I think, as someone who's lived and practiced this stuff for many years, the thing that matters is not the theory but the practice.
Somehow we need to get under the skin of our institutions, their actors and (most importantly) the boardroom. Somehow we need to use technology intelligently (and not be diverted by "intelligent technology"!), and somehow we should pass this all - knowledge and practice - on to our students.
So there's a lot to do...
Cleanest short statement of the tradition I have read in a while, and the bonus lesson is the one most people skip.
The thing I keep circling is that this machinery tells you how a system moves but not whether it can afford to keep moving. Structure drives behavior, agreed. But two systems with identical structure can sit on opposite sides of viable: one replenishing the stocks it draws on, the other quietly liquidating them. So I find myself wanting a verdict on top of the description. At what cost is this behavior being held, and for how long, before the stocks it depends on run out.
Your lesson 9 is where it bites. The tradeoffs are not only design constraints. When a system optimizes its visible target by spending the buffers that look like waste under normal conditions, you get rising metrics and accumulating fragility at the same time, and the description reads as success right up to the threshold in lesson 10. The hardest part is that nothing in the measured behavior tells you which side of viable you are on.
Almost thinking like a social scientist ;)
Most systems are comprised of human beings, at some level. And they behave in ways that are sometimes not intelligible in terms of the system in which they are imbedded. This is the agent structure debate, and whilst structures are clearly (IMO) more important, human agents create and shape those structures: their logics are constructed and are constantly being remade.
This is very elegant. I’m taken back to my early teens when I spent a couple of years fiddling with electronic components trying to do hobby projects. Trying to understand why they didn’t work was in a way more interesting than playing with them when they did. (Then I decided to stick to maths - equations behave themselves, transistors don’t).
Excellent exposition! Do I detect some of Ludwik's thinking? (Hardly surprising, given he influenced the ideas of most people he fell into conversation with!) Best wishes.
Very helpful analytical framework. I’m an economist who has worked mostly on emerging economies. Your analysis is a very helpful reminder that recommendations, even if fully taken, don’t have straightforward and predictable consequences.