10 (+1) Lessons from Systems Dynamics & Control

... trust me, important!

TL;DR: Systems dynamics and control provides a framework for understanding why complex systems behave as they do. Its central insight is that behaviour emerges from structure. Feedback, delays, accumulations and constraints are more important than the characteristics of individual components. These ideas originated in engineering, but they illuminate the behaviour of organisations and institutions as readily as they explain physical and technological systems.

I know that @profserious articles cover a wide range, and it may be that if you are interested in higher education you are less interested in national security, or if you subscribed to read about emerging technology then articles about policy are unwelcome. But, I would like you to trust me on this, whichever of these is your primary concern, this article is for you.

There are some things you learn that change the way you see the world. Once introduced to them it becomes difficult to think in quite the same way again. For me, learning about systems dynamics and control is one of those intellectual inflection points. It provides a language for understanding why complex systems behave as they do and, in doing so, changes the questions one asks about almost every problem.

1. Structure drives behaviour. Systems dynamics starts from a simple proposition that has profound consequences. Behaviour is generated primarily by the structure of the system rather than by the properties of the individual components. Structure comprises the network of reinforcing and balancing feedback loops, together with the constraints and relationships between different parts of the system. Similar structures often produce similar patterns of behaviour regardless of the characteristics of the components. This implies a different way of analysing persistent problems. Rather than asking why specific components fail to achieve their goals, this approach asks instead what features of the system generate the undesirable behaviour.

2. Stocks are not flows. A stock is an accumulated quantity. A flow is the rate at which that quantity changes. Stocks therefore change only through the net effect of their inflows and outflows, which means that flows can change immediately while stocks can only change over time. This distinction, although straightforward, is nevertheless frequently misunderstood. Implicitly confusing rates with accumulations gives rise to systematic underestimates of the inertia inherent in complex systems.

3. Feedback entails delay. Feedback is what distinguishes a dynamic system from a passive one. Reinforcing feedback amplifies change, whilst balancing feedback opposes it and tends towards stability. In practice, however, feedback is almost always subject to delay arising from sensing, communication, decision-making or physical processes. Those delays alter the behaviour of the system and can transform stable control into instability if corrective action consistently arrives too late. Much of control theory is concerned with understanding how feedback and delay interact to determine system behaviour.

4. Real systems are nonlinear. Many analysis techniques – and common intuitions – begin by approximating a nonlinear system with a linear one because linear systems are tractable and often provide a good approximation over a limited operating range. The approximation eventually fails because real systems contain thresholds, saturation effects, discontinuities and multiple operating regimes. As a consequence, proportional changes in input do not necessarily yield proportional changes in output, and small changes in conditions can sometimes produce qualitatively different patterns of behaviour. The most interesting and important behaviour of complex systems frequently occurs precisely at those points where linear intuition breaks down.

5. Disturbances are often amplified. Systems composed of multiple interacting stages frequently amplify rather than attenuate disturbances. Delays, local decision-making and imperfect information mean that a small variation introduced at one point can become progressively larger as it propagates through the system. The resulting behaviour is often counter-intuitive because actions that appear locally rational can collectively increase overall instability.

6. Closed-loop systems compensate. Feedback systems often maintain their behaviour in the presence of disturbance. Attempts to change their behaviour are therefore frequently interpreted as disturbances and are partially or wholly compensated by existing feedback mechanisms. Consequently, altering the magnitude of an intervention often produces less change than expected because the system itself responds to preserve its existing mode of operation.

7. Measurement changes behaviour. Feedback control depends upon measurement, but once a measured variable becomes the basis for decision-making the system begins to optimise that variable rather than the underlying objective it was intended to represent. The statistical relationship between the measure and the objective consequently changes, often reducing the usefulness of the measure itself. It is a general property of feedback systems that the act of control alters the behaviour being measured. Targeting a proxy hollows out its value as a measure (popularly, Goodhart’s law).

8. Every controller embodies a model. Every controller operates according to assumptions about the behaviour of the system it regulates (as in the Conant–Ashby theorem). Some of those assumptions are explicit in mathematical models, while others are implicit in rules or policies. Since no model captures reality perfectly, the performance of a controller depends upon the extent to which its assumptions remain valid. Differences between the assumed model and the actual system frequently explain unexpected or degraded behaviour.

9. Performance always involves trade-offs. Control systems cannot simultaneously maximise every desirable characteristic. Increasing responsiveness generally reduces stability margins, while increasing damping suppresses oscillation at the cost of slower response. Improving sensitivity often increases susceptibility to noise. These relationships are fundamental properties of dynamic systems rather than shortcomings of particular designs. They arise because the underlying dynamics constrain which combinations of behaviour are simultaneously achievable.

10. History influences future behaviour. The behaviour of many systems depends not only upon their current state but also upon the path by which that state was reached. This is known as hysteresis. Once a system has crossed certain thresholds, simply removing the original cause does not necessarily restore the prior behaviour because the system has entered a different region of operation. The present observable state therefore does not always contain sufficient information to predict future behaviour without knowledge of the past.

11. Bonus: Complex systems are steered rather than controlled. As systems become more complex, complete knowledge of their state and complete authority over their behaviour become unattainable (Ashby’s law of requisite variety makes this limit explicit). Their evolution emerges from many interacting feedback processes operating at different scales. Consequently, interventions influence rather than determine behaviour, shaping the conditions under which some outcomes become more likely than others.

Systems dynamics and control is not simply an engineering discipline; it is a way of thinking about complex systems wherever they occur. The significance lies less in the mathematics than in the analytical framing they permit. They have shaped the way I view technology, policy and institutions, and how I approach building solutions to complex challenges. I commend these 10(+1) lessons to you.