At the end of March, the Royal Society issued a clarion call to modellers. As the coronavirus crisis tightened its grip on the country, the UK’s science academy was asking researchers and data scientists to join an effort to boost the country’s capacity to create mathematical representations of the pandemic.
In a statement published on its website on March 29, the Society said that the UK’s “small but highly effective community” of pandemic modellers was overstretched, busy creating models to inform the government’s response. According to the Royal Society there was a need for modelling experts from other industries – such as urban planning, finance, marketing, meteorology – to ease that overload by volunteering their help. That help could range from providing advice and insights, to vetting existing models, to assistance with the software design and the data analysis. The ultimate goal was beefing up the UK’s modelling firepower so that the country could find its way out of the lockdown.
“We need to assist our pandemic modellers by building capacity to explore and optimise possible exit strategies,” the Royal Society said in its statement. Less than a week later, it announced that the response from volunteers had been “enormous.”
The importance of good models to read and interpret the unfolding crisis has become apparent in the last few weeks, especially after the accuracy of some of the models guiding the government’s action – notably those devised by Imperial College London – has been called into question by academics. But it was very clear within the government from the very beginning. On March 16, senior Number 10 advisor Dominic Cummings – who has often written about the use of models and simulations for policymaking – chaired a meeting with over 40 founders and executives from the technology industry, asking them to contribute to the country’s fight against the pandemic. Among those invited was one unicorn that has become a byword for simulations and modelling – London-based Improbable.
Founded in 2012, Improbable made a name for itself by creating SpatialOS, a platform for developing and hosting large-scale digital simulations of entire worlds. Such methods are typically used to create video games, especially multiplayers set in open-world environments. But Improbable’s technology can also help understand city mobility by simulating vehicle traffic, or to assess the incidence of crime in a certain area. The company has also assisted Britain’s ministry of defence with the creation of wargame-like simulations for training purposes.
“We combine scientific modelling, artificial intelligence and data analytics to create, essentially, a platform orientated towards the simulation of real-world environments,” says Joe Robinson, Improbable CEO for defence and security. “The aim of all of this really is to improve decision-making and increase the effectiveness of the whole of a government’s preparedness activities whilst also drastically reducing the cost.”
Robinson will not get specific about how the company is working with the government to confront the coronavirus emergency, except that they have been “incredibly busy” of late. A person familiar with the Number 10 meeting says that Improbable offered to make its simulation technology available to aid the government’s epidemic modelling.
So what would that look like? One of Improbable’s specialties is agent-based modelling. That is a technique that builds simulations for every individual element in a certain environment – such as people, cars, or businesses.
“These [modelled entities] can be civilians, or institutions, or even inanimate objects, like a power plant,” says Christoforos Anagnostopolous, a principal scientist at Improbable. “That allows for very rapid extensions, because you can stack up behaviours on the agent’s level, and then ask different experts to tell you how the agent would behave in different circumstances or different aspects of their behaviour.”
Imagine, for instance, creating a simulation – think of a sort of low-fi version of The Sims, only massive. You can populate it with sixty-six million “agents” – digital stand-ins for people – each with their movement patterns, working routines, social networks, and health conditions and then fast-forward the simulation to observe the progression of the coronavirus crisis. You can add a gauge, showing how the infection propagates over time. What is more, you can run the same simulation several times, changing the rules governing the environment – for instance, you could close the schools, or ban large gatherings – and watch how the same environment behaves under different conditions.
Read more: If we're living in a simulation, this UK startup probably built it
“Agent-based models are particularly good in situations where you need to explicitly model the interactions and the behaviour of the individual components of a system,” says Nick Malleson, a professor of spatial science at the University of Leeds, who has worked with Improbable to study crime patterns. “I think the reason that they've become popular for [studying] disease spread is that very often in a disease spread, you might need to look at how people are interacting – when they come into contact in shops, how the social networks affect how people move, how they behave, how they interact, all these kinds of things.”
Nigel Gilbert, a professor in computational social science at the University of Surrey, and a major proponent of agent-based modelling, offers a word of caution. “These kinds of models will not make accurate predictions. There’s a lot of randomness,” he says. “That means a few things: you need to run the model many times and then get the average. And that’s still an approximation – you’ll get an idea of what might happen but not of what will happen.”
The Imperial College research makes use of individual based models, a technique similar to agent-based simulations. What Improbable’s SpatialOS brings to the table, Anagnostopolous says, is first and foremost its potential to minimise fragmentation.
“In a pandemic, you really want to be able to couple together a model of the outbreak itself – the infection, the epidemiological model – with a model of the economic impact on small businesses, as well as possibly a behavioural model of compliance,” he says.
But often, each of those models would be run by different researchers who are mainly interested in one particular facet of the outbreak. Anagnostopolous says that Improbable is currently working “with a very large network of academic institutions” to address that issue and enable the merging of different models into a single simulated environment. The company is also working to help researchers to update their models as new information comes in.
“Models are often built with a certain type of situation in mind. And when the crisis actually erupts some of these baseline assumptions will be different. And repurposing a model is quite challenging,” says Anagnostopoulous. “Now, we're trying to offer them tools that they can use to help assimilate evidence as this accumulates.”
Of course, in order to update simulations in any significant way the new data has to be plentiful, and reliable – and that is not always the case in the current circumstances, as reports on slipshod data collection practices in China or Italy show. But, Anagnostopolous explains, agent-based modelling is not the only way simulations can help experts and researchers make sense of the coronavirus crisis. For instance, probabilistic models – which enable more randomness to be factored in – might need less data and still return useful insights.
“By relying on techniques like probabilistic modelling – which allow for a lot of uncertainty to be captured in the model – you're able to produce forecasts or possible outcomes on counterfactuals that capture both what you know and what you don't know,” Anagnostopolous says.
“Our platform is able to support both kinds [of simulations] and indeed, interactions between these two types of models. We're working really hard to support as much variety as possible.”
Gian Volpicelli is WIRED's politics editor. He tweets from @Gmvolpi
😓 How did coronavirus start and what happens next?
❓ The UK's job retention furlough scheme, explained
💲 Can Universal Basic Income help fight coronavirus?
🎲 Best video and board games for self-isolating couples
👉 Follow WIRED on Twitter, Instagram, Facebook and LinkedIn
This article was originally published by WIRED UK