Search

Event report: Agent Based Modelling and its Real World Applications

On the 25th November, De Morgan House hosted an evening focused on Agent Based Modelling and its Real World Applications.  The event was organised by GSD in collaboration with the Lighthill Institute for Mathematical Sciences and featured two speakers: Liz Varga from the Complex Systems Research Centre at Cranfield University and Patrick Beautement from the Abaci Partnership.  Both with a background of complex systems in industry, Liz and Patrick gave excellent talks highlighting the uses of agent based modelling in an industrial context as well as pointing out some of its limitations in conjunction with further work that needs to be done in the area.

Liz Varga’s talk started with a brief review of both complex systems and mathematical modelling.  A remark that a complex system is not repeatable and that two starting points will never reproduce the same results sparked some lively debate with the audience and the importance of the context of the complex system became clear. Liz talked about how the more constraints that are added to a system the less likely the system is to represent real life and the need for evolutionary multi-agent behaviour.  Moving onto agent based modelling in practice, programming methods were discussed as well as the importance of choosing an order for the agents to act when developing code – different orders often generate different behaviour.  Two examples were given of agent based models: an economic market model, where the health of the economy fared best with a learning by experiment rule, rather than a Darwinian approach or by imitating the winner, and a contingency model of a paper factory, where multi-agent modelling was able to make factory processes more efficient.

Patrick Beautement followed on from Liz’s talk by considering the question is agent based modelling ‘real world ready’? Decomposing the problem space into ‘puzzles’ and ‘mysteries’, Patrick discussed how agent based modelling can be very effective in the puzzle space but once we move into the mystery space, human decision becomes critical and agent based modelling is of less value. In the mystery space, the problems are focused on shocks to the system rather than day to day processes.  There was some interesting debate regarding how much resources should go into planning for shocks and Patrick felt this was certainly an area where more research was needed.  Some impressive visualisations allowed Patrick to explain the different types of problems and the role of agent based modelling in each. Patrick also recommended the book Reshaping National Intelligence for an Age of Information by Gregory F. Treverton for further reading in this area.