Written by Mimi Keshani

4 min read

Enabling Better Decisions With Multi-Agent Systems

  • Artificial Intelligence
  • Life Sciences

Simulating Virus Transmission With Multi-Agent Systems

The World Health Organization (WHO) has now officially classified COVID-19 as a pandemic, and governments, in particular those seeing an exponential increase in the number of cases, have stepped up their efforts.

There is an acceptance that we cannot stop widespread population transmission immediately, but by ‘flattening the curve’ we can reduce the burden on health services and ultimately decrease the mortality rate.

However, creating predictive models to make actionable insights from is difficult with today’s technology. As the COVID-19 outbreak unfolds, new data will continually emerge and it’s vital that governments and businesses can rapidly and effectively respond.

Governments must be decisive in order to distribute resources and save lives. Accurate modelling has helped South Korea achieve the lowest death rate of all countries with significant numbers of patients. 

Enabled by early testing, South Korea was then able to model the spread and implement the most effective models quickly – “deploying a range of legal, medical, technological and public communication efforts.” As such, they were able to achieve inspiring results without implementing draconian measures such as cordoning off entire cities or implementing nationwide shutdowns.

Enabling Better Decisions With Multi-Agent Systems | Hadean

Taken from Flattening the Coronavirus Curve by The New York Times

How Can Governments Better Advise The Public In The Event Of A Health Crisis?

Modelling different scenarios via a multi-agent system simulation can highlight behaviour patterns that emerge over time.

Multi-agent systems consist of autonomous entities known as agents working collaboratively. Agents solve tasks by learning and acting on interactions with one another and the environment they’re in. These entities could represent anything, including different computers in a network, different pieces of software, or even a person

When multi-agent systems are used to process lots of real-world data from an emerging situation such as an outbreak, it can allow civil servants, doctors, and organisations to create malleable models of different potential scenarios. These simulations reveal non-obvious outcomes, which otherwise may have gone unnoticed allowing leaders to make informed decisions. An illustrative example of this is Using Agent-Based Models in Contagion Modelling.

UK Multi-Agent Systems Symposium At The Turing Institute

Recently, I was fortunate enough to be invited to attend a Turing Institute event focusing on multi-agent systems and their application within the world today.

From Autonomous Vehicles to Healthcare, Industry 4.0 to Gaming, multi-agent systems are commonplace. Researchers from both industry and academia are currently investigating the topic including Microsoft, Deepmind, the University of Oxford, and the University of Edinburgh.

Enabling Better Decisions With Multi-Agent Systems | Hadean

Slide created by Stefano V. Albrecht, University of Edinburgh

However, as access to accurate data and the availability of computing power continues to rise we expect a proliferation of multi-agent systems entering society. For example, as artificial intelligence becomes more integrated into our healthcare system, it's integral that these systems work together to help improve the efficiency of our services.

Hadean and Multi-Agent System Modelling

Last week we published our work modelling the spread of COVID-19, illustrating the effect of combining multi-agent systems with spatial simulation technology. In the space of 48 hours, an agent-based model of 100,000 entities was created using Aether Engine, mapping the potential spread of the pathogen throughout the country.

But this is only a prototype. The complexity of a model can grow infinitely, layering in additional data – including point-based (e.g. individual people), graph-based (e.g. transport networks), or grid-based (e.g. transmission/spread of droplets) – to refine the accuracy of the simulation.

These models can then be manipulated to simulate “what ifs” quickly. Traditionally these models require significant computing power and time to run the simulations – but, by contrast, the Aether Engine simulation can be up in a matter of minutes, and scale into the cloud with no additional effort from developers. Data is processed more efficiently and the probability of potential outcomes is more accurately predicted, allowing for informed decisions to be made quickly. 

Update: We are delighted to announce that we are renewing our partnership with the Francis Crick Institute. The project will combine analysis of person-to-person interaction with insight into how COVID-19 transmits within an individual, providing a complete picture of the pathogen’s spread. Find out more here.