It’s difficult to create predictive models quickly with existing technology – without accurately understanding the true or likely impact of an emerging situation it is impossible for governments and businesses alike to effectively distribute resources, guide public health efforts, or plan for wider socioeconomic fall out.
Coronavirus (COVID-19) is directly affecting thousands of people. Understanding its spread is an extremely complex scientific and engineering problem that must be addressed by governments looking to take mitigating steps as quickly as possible. Ideally we would like to support informed policy decisions and minimise draconian measures such as quarantining an entire city.
At Hadean we have explored the role a multi-agent systems simulation of COVID-19 might be able to play in modelling the risk of infection and transmission across the UK. Over the course of three days we’ve designed an initial simulation involving 100,000 entities, although it could be scaled to the entire population if required.
Currently we don’t know the R0 number for COVID-19, and in the context of a complex, large-scale simulation, small errors become dramatically magnified. If one part of the framework is misjudged or data is input incorrectly, the repercussions can be significant – especially if decisions are made based upon the inaccurate output. Indeed, as it stands data sources for such a model are diverse and create a complex contagion model, including virus genomes, crowd sourced data and social data.
One way around this is to create models which don’t rely on these inputs. The Aether Engine simulation instead focussed on movement patterns to try and isolate specific scenarios. Focussing on two different hubs – Liverpool and London – the simulation shows how the pathogen might spread from different locations along the UK road network. It simulates in super-real-time and can be quickly re-run as new data becomes available.
Multi-Agent Spatial Simulation [MAS] can be used to increase the flexibility of contagion models and is a better way of modelling geographic data than network theory approaches. The initial demo can be scaled upwards by continuing to add layer-upon-layer of additional input data, from transport infrastructure to further sociodemographic information. Without using such a model it becomes difficult for governments to plan and effectively distribute the resources that guide public health efforts and curb the infection.
Contagion modelling has been a theoretically interesting but practically limited area which makes it difficult for organisations to implement appropriate measures in time. However, it’s becoming easier to harness the increased data and computational power we can build out accurate, scalable models.
The multi-agent spatial method provides an intuitive geographic representation and provides insight into what might be achieved. There are also a number of other projects currently being worked on for contagion models using artificial intelligence and machine learning techniques, including those being run by BlueDot, Flowminder, and Metabiota.