2. How to make a model: The art of systems modelling
You can’t do much carpentry with your bare hands, and you can’t do much thinking with your bare brain. — Bo Dahlbom
How do you get to grips with a complex system?
Over 2011–14, myself and the members of Boho were in residence at University College London’s Environment Institute grappling with this question. The system we were examining was a music festival. The result of our efforts was a systems model that looked like this:
And like this:
What is a model?
The most straightforward definition, though it requires some unpacking, is something like this: ‘A model is a simplified representation of a real world system.’
My favourite example, courtesy of David Shaw, is a model car.
A model car looks a bit like a real car. It’s smaller, and it doesn’t contain all of a real car’s working and functions, but if you’d never seen a car before and wanted to understand what they were, this would be a good place to start.
A scientific model is the same. It’s a stripped back version of the real world, made simple enough to be useful.
A map is a model. A map is a simplified representation of a system — in this case, a geographic region. Every map, from a high-tech satellite image to directions scribbled on a napkin, is a model.
Everyone models
Modelling is a universal activity.
All living creatures store information from the past and extract regularities from it. These regularities are a model of the environment which that creature uses to anticipate the future. From Bradbury et al:
‘Whether it is a tree responding to shortening day length by dropping its leaves and preparing its metabolism for winter — in advance of winter — or a naked Pleistocene ape storing food in advance of winter for the same reasons, both are using models.’
Mental models vs scientific models
Because we all use models to help us understand the world, the question is not ‘Should we use models?’ but ‘How do different models compare with each other?’
Joshua Epstein makes the distinction between the implicit models we have in our heads, versus explicit models that we write down:
In the models in our heads, Epstein says, ‘the assumptions are hidden, their internal consistency is untested, their logical consequences are unknown, and their relation to data is unknown’.
In contrast, an explicit model can be shared, discussed, tested, and other people’s opinions and expertise can be incorporated.
Why model?
Many of the models we make are predictive. Weather models, climate models and infection models are designed to help us anticipate the future.
Predictive models rarely allow us to foresee an exact event. Instead, they give us an estimation of its likelihood — a rough sense of best and worst case possibilities.
A geophysical model that predicts seismic activity won’t tell us when and where a large earthquake will happen — but they’re still valuable in figuring out what kind of anti-seismic construction materials you might need.
Models don’t have to be predictive to be useful, and some of my favourite models don’t claim to have any predictive power. Their aim is to help us understand systems.
I’m thinking here of Nicky Case & Marcel Salathé’s beautiful illustration of agent-based covid models — this is purely a tool to help get to grips with different ways of imagining infection.
Beyond the realm of prediction, Joshua Epstein’s ‘Why Model?’ lists a variety of other ways in which modelling can improve our understanding of systems: they guide data collection, illuminate core dynamics, suggest dynamical analogies, reveal new questions, illuminate core uncertainties and challenge prevailing theories through perturbations.
In their paper for the Australian Academy of Science entitled ‘What is a model, why people don’t trust them, and why they should,’ Bradbury et al list some of the useful skills that working with models can help to train:
‘The ability to manage limited resources and unexpected feedbacks, to interpret outcomes against expectations, to balance emotional responses (eg. humility, curiosity, frustration and blame-shifting), to tolerate high levels of uncertainty, acknowledge mistakes, to search for counter-evidence and to usefully self-reflect.’
Australia’s Defence Science and Technology Organisation (DSTO) used models to train officers in adaptive thinking. The DSTO’s Anne-Marie Grisogono describes how soldiers were tasked with playing a simple systems model (managing a chocolate factory).
The most effective training tool, Grisogono writes, was when officers were placed in pairs. In turns, one person operated the sim, while their partner commented aloud on their strategies and choices. Simply the fact of having their decision-making processes made conscious was a powerful means to help strengthen them.
Platforms for discussion
For me, one of the most vital roles that models can play is as platforms for discussion. This is where much of my work takes place — creating models that give people a space to consider the systems they’re a part of, to discuss options and plan strategies.
Done correctly, an explicit model can provide a platform for constructive debate between people from varied backgrounds or with opposing viewpoints.
I draw here on the idea of ‘participatory co-models’. This is Pascal Perez’s term for what have also been called companion models, post-normative analytical models or participatory planning models.
This is a model-building practice which focuses on engagement rather than accuracy, which highlights the process rather than the outcome of the model.
The aim is to involve the stakeholders who will be affected by the model’s outcomes in its creation.
Typically, scientists creating a model of a wetlands or a farming basin will gather data from monitoring instruments, survey maps and official records. In a co-modelling process, scientists will instead undertake consultations with local stakeholders to bring in a broad range of specific expertise.
If you interview ten different people about how a complex system works, you’re likely to get ten different answers. The farmers upstream will have a different opinion of how much fertiliser runoff the river can handle than the conservationists working downstream.
Where there is disagreement, the model-builder avoids choosing a correct answer, and instead builds the model to incorporate all the conflicting opinions.
This results in a far less accurate model with, as Perez puts it, ‘limited capacity for prediction’.
The advantage of this approach is that the model has been ‘socially validated’ by a wide spectrum of stakeholders. Consequently, it can be used as a legitimate platform for decision-making.
In essence, a participatory co-model is the least worst depiction of the system that all the relevant groups can agree on.
It becomes a ground for compromise — as in the Democratic Nature project that I worked on with Boho for Swedish NGO Miljöverkstan, where the aim was to bring a group of people around the table to discuss the future of Swedish nature reserves (more on this later).
The art of modelling
The key criteria for any model, no matter what it’s of or who it’s for, is that it is useful.
There are no points for creating the most detailed or most beautiful representation of a system, unless that detail serves a purpose. The often stated ideal for models is that they be ‘as simple as possible but no simpler’.
If you’re drawing a map on a napkin to guide you from the restaurant to the train station, you don’t need a detailed rendering of the building facades. You probably don’t even need street names. ‘Third right, first left, cross at the lights’ is probably sufficient — and much more detail is likely to be distracting as much as it is helpful.
A more recent example: the iconic ‘flatten the curve’ graph showing the potential impact of a Covid-19 outbreak is incredibly simple. There are no numbers, no scale, no variability. But it does the one thing that’s required of it; to explain why social distancing is needed to prevent an exponential multiplying of cases.
Even the massive global climate models that give us our longrange scenarios for global warming are as simple as they can get away with. When you’re using petaflops of processing power, it behooves you to be parsimonious.
This illustration of global tipping points around climate from PNAS journal in 2018 does everything it needs to in terms of communicating the escalating risk of climate disaster with the minimum amount of detail.
Modelling =/= understanding
There is a crucial caveat in all of this — that modelling is not the same as understanding.
A model is a tool to help us understand a system — but the map is not the territory, and the way to understand a territory is not simply to make more and more detailed maps.
In New Dark Age, James Bridle notes how our focus to data, modelling and computer simulations cuts us off from other ways of understanding the world:
‘As life accelerates, the machine steps in to handle more and more cognitive tasks, reinforcing its authority — regardless of the consequences. We refashion our understanding of the world to better accommodate the constant alerts and cognitive shortcuts provided by automated systems. Computation replaces conscious thought. … That which is possible becomes that which is computable. That which is hard to quantify and difficult to model … is excluded from the field of possible futures. Computation projects a future that is like the past — which makes it, in turn, incapable of dealing with the reality of the present, which is never stable.’
I think of this tendency every time I hear a government advisor spouting half-digested scientific buzzwords, or tech utopianists promising an app for whatever practical or social problems we face. When we’re talking about real world complex problems, a model never gives us the answer — and we should mistrust anyone who claims they do.
But the other extreme is to discard models altogether, to dismiss data as an elitist obsession, and to instinctively go with your gut. This is the kind of mendacious leadership currently being endured by the people of the Philippines, the USA and Brazil, among anothers.
Watching men (always men) who take pride in their refusal to learn anything fumble their way through a major humanitarian crisis is exhausting and heartbreaking.
Meteorologist David Lallemand at the Earth Observatory Singapore talks about the challenge of making models for policy-makers. In the face of an oncoming typhoon, the temptation is for meteorologists to provide government with detailed models, including all their assumptions and various scenario.
David points out:
‘When a local mayor receives a 50mb pdf containing 10 different maps showing 10 different things, they’ll likely end up throwing it all out completely and phoning up the local police station and asking them to walk around and tell them what they see. So by giving them so much information we make our work redundant.’
Too much detail can be worse than not enough.
Experiential models
In my practice, I create ‘experiential models’ — models designed for people to interact and engage with.
Some are models of complex systems, others model possible future scenarios.. Though informed by data, they are qualitative rather than quantitative, and rarely rely on software simulations. They are intended to illuminate aspects of a system, to help people think through complex problems, and to provide a space for conversation and reflection.
Boho’s Best Festival Ever is an example of this kind of model. Using a fictional music festival as its subject, the work is designed as a primer to the practice of modelling.
Through an interactive workshop, we introduce people to various types of systems models by presenting various models of the same system.
This is a model of a music festival.
This is a model of the same festival.
What’s on the table is a physical model, designed to show the physical layout of the space and the transport network of the festival. What’s on the board is a conceptual model that shows how some parts of the festival influence other parts.
Participants navigate through the experience in a third kind of model — a narrative. The story that takes them through the three day festival (and its disastrous consequences) is a different kind of model, with its own rules. More on that soon.
1. We live in systems: Becoming aware of what surrounds us
The disasters were designed by us — cuckoo clocks vs ants nests — the humility of systems thinking — seeing deep patterns — my practice as a writer, theatre artist and game designer, tools & techniques for thinking about the world
2. How to make a model: The art of systems modelling
What is a model? — maps on napkins vs satellite images — as simple as possible but no simpler — Best Festival Ever: modelling a disaster
3. A snapshot of everything: Tools for systems mapping
Mapping a Swedish forest — thousand year old oak trees — resilience assessments — a walk through the woods — Democratic Nature —
4. The future doesn’t exist: Scenarios and prediction
Why bother trying to predict the future? — the practice of creating scenarios — there are four possible futures — CrimeForce: LoveTeam
5. Narrative in systems: How to tell stories about complexity
Are theatre shows systems models? — underdog narratives & police procedurals — perspectives on an Indonesian rainforest in 95 Years or Less
6. Creating an experience: What design and dramaturgy teach us about worldbuilding
Theatre as rehearsal for revolution — dramaturgy & design thinking — tactility in Get The Kids and Run — collective experiences in Gobyerno
7. Don’t play games, make games: Interactivity in complex systems
Games are systems — game theory in Temperature Check — calculating risk in Busy Mayors — skilltesters vs decision-makers in Run A Bank
8. Lessons learned
Final thoughts — steering 9 billion people through a century of climate and global change — working with time instead of against it -