Lost in Abstraction: The complexity of Real Environments vs. the Assumptions of Models

Many if not most fire evacuation models are relatively insensitive to the complexity of human behaviour in the built environment. The complexity of human behaviour and the very long feedback loop on building disasters can lead to the approval of structures that have massive inherent vulnerabilities. Buildings last a very long time. How does the designer propose to validate the model for the lifetime of the building? What keeps the occupant characteristics within the range proposed by the engineer? What keeps the reactions of the occupants predictable in the future based on the reactions of people with the current characteristics? Fire evacuation models routinely include what is claimed to be “engineering judgement”. But while engineering judgement is well designed to interpolate data between known data points in the presence of an adequate covering law, it cannot be extended to extrapolating from mere observational data in the absence of a suitable “covering law”.

This is especially true with regards to complex structures, i.e. buildings that involve complex variables that lack supporting data or are difficult to predict. We need better tools to understand the link between people and the built environment. In the process of abstraction and simplification, significant details and complex variables tend to be lost leading the models to overoptimistic outputs.

Some modelers may think that the inputs to behavioral models represent universal reality and they do not take into account Unknown Unknowns, which include:

• Wayfinding in unconventional evacuation paths (old and new buildings)
• Unconventional walking surface (old buildings)
• Cultural confusion (old and new buildings, e.g. airports)
• Unpredictable human behaviours

This paper will illustrate a small portion of this core issue in the use of behavioural models. We examine the assumptions routinely used in such models when compared to the reality of complex structures.