Some of the principal characteristics of disaster situations are as follows. (1) Conditions on the ground are uncertain and liable to change in unpredictable ways. (2) Knowledge of conditions is incomplete and likely to remain so for the duration of the emergency, despite the best efforts of all concerned. (3) A common operating picture is created and shared only slowly and with much arduous work. (4) Field commanders and coordinators want simple solutions that require no great effort of thought or computation, not because they are unintelligent, but because they must devote almost all of their attention to directing emergency work. (5) Face-to-face communication is the most vital means of conveying information and the only one that can utilise the full range of people's sensory attributes. (6) Managing the convergence reaction is such a complex logistical exercise that it is inevitable that the outcome is approximate rather than precise.
Recently, there has been a sudden upsurge in the application of operations research (OR) to the management of sudden impact disasters. Mathematical and statistical algorithms are being written in the hope of optimising logistical actions. The circulation of traffic, the evacuation of cities, stockpiling and warehousing relief goods, improving vehicle fleet management, and so on, are some of the goals. Much of the work is inductive and relies on vacuuming up data, whacking them into order and squeezing them through a set of equations or matrices in order to produce the 'perfect' output.
Imagine this in the field. The earth shakes, producing massive destruction and a substantial toll of casualties. We whip out the iPad or laptop computer and turn it on. We turn to WiFi, but it has gone. So has the cable-based network and the local electricity supply. Nevertheless, the device still has some battery power. We crank up the algorithm and work out where best to stockpile relief goods. The solution that the algorithm proposes is unworkable because (a) we have no time and resources to build warehouses; (b) the accessibility of places has changed radically because of route blockages (rubble in the street, buildings in danger of collapse, damaged bridges, etc.); (c) rendezvous points have to be agreed with many organisations.
The process of OR modelling for optimisation of disaster relief is completely dependent on the quality of the assumptions that underlie it. In most cases these are scarce, threadbare, or simply unworkable. They do not mirror the real situation, as most of the modellers have no experience of conception of what that is like.
Science is very susceptible to fads and fashions. Perhaps the stimulus which motivates the current craze for mathematical optimisation of disaster relief is a desire to bring order to chaos. What could be more attractive than turning a messy, inefficient situation into one which is clean, streamlined and super-functional? If this is the motivation, then it ignores--at its peril--the old Aristotlean idea of generatio and corruptio. In disasters, as in so much else, forces are at work that break down order while other forces that create it are also at work and conflict with them. Many disaster plans have gone haywire because people have not done what the assumptions said they would do. One assumption is that we all have the same motives and objectives. We do not.
Perhaps the biggest chasm that would need to be bridged here is the one that occurs between the academics and the field coordinators. How can algorithms work if the cultures are different, have different receptivities, work towards different goals, and have different expectations? Disaster management would be so simple if no people were involved. People are such a nuisance, with their huge variety of attitudes and behaviours (Kirschenbaum 2003).
Will artificial intelligence do the trick? The current craze is to use the analytical hierarchy process (AHP), which uses inductive reasoning to make decisions. Rather than substituting the decision process, AHP merely shifts it to a different part of the process. If the initial decisions are wrong, so is the outcome, a classic "garbage in, garbage out" syndrome (Whitaker 2007).
When we have finally got the algorithm to spit out the perfect solution for step one, and have applied it by directing all traffic down one road to one site, the battery finally fades out and the computer is dead. Has anybody got a spare battery? No.
In disaster relief, we do not optimise: we try to muddle through until the end and keep wastage to less than ten per cent.
References
Kirschenbaum, A. 2003. Chaos, Organisation and Disaster Management. Marcel Dekker, New York, 318 pp.
Whitaker, R. 2007. Criticisms of the analytic hierarchy process: why they often make no sense. Mathematical and Computer Modelling 46(7-8) 948-961.
