Systems thinking (see earlier entry) drives us to interdisciplinarity: we need to know *everything* about a system of interest and that means anything and everything that any relevant discipline can contribute. For almost any social science system of interest, there will be available knowledge at least from economics, geography, history, sociology, politics plus enabling disciplines such as mathematics, statistics, computer science and philosophy; and many more. Even this statement points up the paucity of disciplinary approaches. We should recognise that the professional disciplines such as medicine already have a systems focus and so in one obvious sense are interdisciplinary. But in the medicine case, the demand for in-depth knowledge has generated a host of specialisms which again produce silos and a different kind of interdisciplinary challenge.

Some systems’ foci are strong enough to generate new, if minor disciplines. Transport Studies is an example, though perhaps dominated by engineers and economists? There is a combinatorial problem here. In terms of research challenges, there are very many ways of defining systems of interest and mostly, they are not going to turn into new disciplines.

How do we build the speed and flexibility of response to take on new challenges effectively? A starting point might be the recognition of ‘systems science’ as an enabling discipline in its own right that should be taught in schools, colleges and universities along with, say, mathematics! This could help to develop a capability to recognise and work with transferable concepts – super concepts – and generic problems (for which ‘solutions’, or at least beginnings, exist). In my *Knowledge Power* book, I identified 100 super concepts. A sample of these follow.

- systems (scales, hierarchies, ..….)
- accounts (and conservation laws)
- probabilities
- equilibrium (entropy, constraints, ……)
- optimisation
- non-linearity, dynamics (multiple equilibria, phase transitions, path dependence)
- Lotka-Volterra-Richardson dynamics
- evolution (DNA,……)

Each of these has a set of generic problems associated with them – not an argument that was fully developed in the book! The systems’ argument has been dealt with elsewhere. Think of any systems of interest that come to mind and let us explore each of these super concepts and associated generic problems.

Systems entities can nearly always be *accounted for*. In a time period, they will be in a system state at the beginning and a system state at the end; and entities can enter or leave the system during the period. This applies, for example, to populations, goods, money and transport flows. In each case, an account can be set out in the form of a matrix and this is usually a good starting point for model building. This is direct in demographic modelling, and in input-output and transport models.

The ‘behaviour’ of most entities in a social science system is not deterministic, and therefore the idea of *probability* is a starting point. The modelling task, implicitly or explicitly, is to estimate probability distributions. We often need to do this subject to various *constraints* – that is, prior knowledge of the system such as, for example, a total population. It then turns out that the most probable distribution consistent with any known constraints can be estimated by maximising an *entropy function, or through maximum likelihood, Bayesian or random utility procedures* – all of which can be shown to be equivalent in this respect – a superconcept kind of idea in itself. Which approach is chosen is likely to be a matter of background and taste. The generic problem in this case is the task of modelling a *large* population of entities which are only *weakly interacting* with each other. These two conditions must be satisfied. Then the method can generate, usually, good estimates of equilibrium states of the system (or in many cases, subsystem). It can also be used to estimate missing data – and indeed complete data sets from samples following model calibration based on the sample.

Many hypotheses, or model-building tasks, involve *optimisation* and there is a considerable toolkit available for these purposes. The methods of the previous paragraph all fall into this category for example. However, there may be direct hypotheses such as utility or profit maximisation. It is then often the case that simple maximisation does not reproduce reality – for example because of imperfect information on the part of participants. In this case, a method like entropy-maximising can offer ‘optimal blurring’!

The examples to date focus on equilibrium and in that sense on the fast dynamics of systems: the assumption is that, after a change, there will be a rapid return to equilibrium. We can now shift to the slow dynamics – for example in the cities case’ – evolving infrastructure. We are then dealing with (sub)systems that do not satisfy the large number of elements, weak interactions’ conditions. These systems are *nonlinear* and a different approach is needed. Such systems have generic properties: multiple equilibria, path dependence and the possibility of phase changes – the last, abrupt transitions at critical parameter values. Examples of phase changes are the shift to supermarket food retailing in the early 1960s and ongoing gentrification in central areas of cities. Working with Britton Harris in the late 70s, we evolved, on an ad hoc basis, a model to represent retail dynamics which did indeed have these properties. It was only later that I came to realise that the model equations were examples of the Lotka-Volterra equations from ecology. In one version of the latter, species compete for resources; in the retail case, retailers compete for consumers – and this identifies the generic nature of these model-building problems. The extent of the range of application is illustrated by Richardson’s work on the mathematics of war. It is also interesting that Lotka’s, Volterra’s and Richardson’s work were all of the 1920s and 1930s illustrating a different point: that we should be aware of the modelling work of earlier eras for ideas for the present!

The path dependent nature of these dynamic models accords with intuition: the future development of a city depends strongly on what is present at a point in time. Path dependence is a sequence of ‘initial conditions’ – the data at a sequence of points in time – and this offers a potentially useful metaphor – that these initial conditions represent the ‘*DNA*’ of the system.

These illustrations of the nature of interdisciplinarity obviously stem from my own experience – my own intellectual tool kit that has been built over a long period. The general argument is that to be an effective contributor in interdisciplinary work it is worthwhile, probably consciously, to build intellectual tool kits that serve particular systems of interest, and that this involves pretty wide surveys of the possibilities – breadth as well as, what is still very needed, depth.

Alan Wilson, April 2015