Simulation, Not Prediction

mississippi-meander-fisk-bannerAs the CEO of a company that uses computational simulation to help leaders make consequential decisions, I am constantly fielding different versions of this question: can you predict the future?

It’s easy to conflate simulation and prediction. After all, both extrapolate what might happen next week, next month, or next year. But while prediction generally strives for accuracy—for example, by providing a better point estimate of a single metric over a given time interval—simulation aims to compare the relative future impacts of choices you face today.

All decisions are based on models. All of us—every human being—use mental models to synthesize complex information. But these mental models are a jumble of vague limbic associations sprinkled with a few frontal cortex data points. For a choice to be truly informed, a decision maker must develop an understanding of the underlying process dynamics based on their own experiences, expert input, and data. At Epistemix we use computational simulation to turn the implicit and opaque mental modeling process into an explicit and open representation.

Let’s make this concrete. As I write, the world is grappling with COVID-19 and caseloads in the United States are at all-time highs. Someone seeking to make a prediction about this evolving crisis might estimate the number of daily new cases nationwide one month from today. If you wanted to evaluate their model, you could wait a month and see what happens or, as a proxy, you could feed historical data into the model and see how accurately it matches subsequent evidence.

But when Epistemix sets out to simulate COVID-19, we start in a different place entirely. First, we ask: who is this simulation for? For example, for the past few months we’ve been working with school superintendents across the country and since the vaccine rollout has started, we are working with state governments to assess vaccine allocation strategies. Second, we ask: what important decision are they trying to make? Many superintendents are in the difficult position of making crucial judgement calls about when and how to open schools, despite the fact that they do not have any public health experience. Third, we ask: whose lives will be influenced by this decision? Students, teachers, parents, staff, and, through second and third order effects, the larger community.

Superintendents have to balance the needs of a diverse set of stakeholders to find the best path forward for their schools. The paths open to them typically include moving to fully open, fully remote, or a variety of hybrid operations. But the high-level data they have access to from county and state health departments, CDC, and WHO rarely clarifies which path is best for their school. Nor do predictions that simply project that data into the future.

At the end of the day, superintendents don’t need an abstract sketch of what tomorrow might look like, they need to evaluate and compare the options they have today. So Epistemix runs simulations that account for every individual person, household, neighborhood, school, and workplace, as well as local demographic characteristics including age, sex, and race. We incorporate the latest local public health data, calibrate the simulation to local disease conditions, and then use realistic social dynamics based on the actual policies being implemented to estimate the impact of particular interventions on the number of COVID-19 infections among students, teachers, staff, and the larger community.

These simulations aren’t useful solely for their point-estimate accuracy. They are useful because they illuminate the relative merits of different responses to the epidemic.

Leaders across the public and private sector are in the same position as school superintendents: they need to make high-stakes decisions in the face of significant uncertainty in order to guide us through this pandemic. And it turns out that when it comes to computational modeling, simulating the impacts of the specific choices you face can make a much bigger difference than predicting “the future” because there is no single “future,” there are many possible futures that depend on your choices.

Getting a leadership team to arrive at a shared understanding of a complex problem is no easy task. Computational simulation provides a common platform to represent and merge insights from a wide array of disciplines, synthesizing previously disjointed ideas. I like to think of simulation as a small epistemic step toward a deeper understanding of the world. Simulation is a testing ground for making better decisions—and a better future.

You don’t need a crystal ball. You need a compass.


John Cordier is the CEO of Epistemix.

Image credit: Meander Maps of the Mississippi River, Harold Fisk, US Army Corps of Engineers, 1944