Thursday, 15 December 2016
'Optimising' Failure in Disaster Response
At the end of the 18th and beginning of the 19th centuries, mechanisation of the weaving industry in Manchester, England, began to destroy the cottage industry of loom work. The Luddites were people who sought to destroy the machines that were destroying their livelihoods. Things are different now: to be a latter-day Luddite is akin to being a modern King Cnut, the ruler who sat on his throne amid the waves to show his subjects that he had no power to stop the tide coming in (and was later misinterpreted as the king who thought he could control the tides but failed). Technology is the tide, indeed the unstoppable tsunami. It brings good and bad things with it.
Regarding the modern advance of technology, I have no desire to be a Luddite of any kind. However, I am concerned about how know-how and equipment are being misused in disaster risk reduction.
Thirty-six years ago, when I began to study disasters, we were only a select few people in this new area of scholarship. DRR is now a crowded field. That, of course, is good because it means that the importance to humanity of DRR is recognised and research institutions are taking the field seriously. However, there are several branches of 'disasterology' in which many of the proponents appear not to have an adequate background in the field, nor an appreciation of the reality behind the problems they tackle. The consequence of this is that the ceaseless application of technology becomes a problem, not a solution. It can create vulnerability by inducing reliance on routines or equipment that in a disaster may function badly or not at all.
Humanitarian logistics is a field in which there is a vivacious tendency to produce algorithms. One can imagine cohorts of mathematicians and computer scientists casting around for problems to solve. Suddenly, they see the delivery of humanitarian aid as the answer to their prayers: a field in need of algorithms. A proper critical analysis would examine the key problem of whether the algorithms have any value in the field. I strongly suspect that more often than not the answer is 'no'. The models are based on assumptions. In some cases these are inadequate, and in many they are untested. Few attempts have been made to find out whether the models function during real disasters, whether they would improve the situation and whether they are attractive to emergency managers. In my experience, they are not.
Regarding the very fashionable problem of how to optimise the location of critical facilities, in a recent earthquake that I studied in the field, facilities such as emergency operations centres and warehouses were located in the only places that were available, accessible and functional. There was no element of choice or optimisation. It was a typical event of its kind, and optimisation algorithms would not have helped in any way. Nor would they have been accepted by the emergency managers on site, who overwhelmingly wanted to simplify their decision making, not complicate it with over-sophisticated routines.
Algorithms designed to optimise facilities presuppose that the data will be available to carry out the optimisation during the critical phase of a disaster aftermath. That is unlikely. Alternatively, they tend to presuppose that decisions can be made before disaster strikes. That ignores the range of geographical variation in possible impacts.
I have a strong feeling that humanitarian logistics and common sense have parted company. For instance, consider a recently published model that optimally located shelters and efficiently assigned evacuees to the nearest shelter site. There is no idea in this work of how to preserve social cohesion. My field research, and that of many others, shows that this is a critical factor in the success or failure of shelter strategies.
If it occurs at all, model testing is usually hypothetical, or it is carried out under highly artificial conditions. There is very little research on the effectiveness of algorithms in real crises, and whether they are able to improve situations. This is hardly surprising. Most emergency decision makers are not interested in optimisation routines and are not equipped to understand or use them. They would be very suspicious of any attempt to replace informed judgement with automated routines. The key question is not "how can we optimise the location of a warehouse?", but "are there any suitable warehouses in the area that we can commandeer and use?"
Increasing dependency upon electronic routines may indeed increase vulnerability to disaster. Any failure of sophisticated electronics and software (or batteries, or connections) risks causing serious problems (will there be electricity after the disaster has struck?). This explains the widespread reluctance of emergency managers and responders to use the algorithms. Indeed, there is significant and well-founded opposition to the 'technofix' approach to emergency management. At the very least, users will have to assume the burden of memorising yet another plethora of mnemonics and initials. Complex training will need to be allied with ability to fix software and hardware glitches with extreme rapidity. Rarely if ever do the papers that present algorithms discuss how dependency upon them may be dangerous, or what would happen if the algorithm fails. When the algorithm is described, seldom is any redundancy offered.
'Optimal' and 'optimisation' are misnomers when applied to a problem that has only been thought through partially. Rather than supporting decisions, the algorithms may thwart them by encouraging decision makers not to think problems through.
Many have argued that the 'technofix' approach to disaster is wasteful and damaging. In an age in which technology has become the world's obsession, it cannot easily be dismissed, nor should it be. Indeed, it holds the answer to many complex and intractable problems, but only if it is used with intelligence and insight. What technology has to offer to disaster management depends on its robustness, its ease of use in a crisis, redundancy if it becomes blocked, its cultural and technical acceptability to users, and its ability quickly to provide useful answers to intractable problems. When papers are published that offer technological solutions to disaster problems, they should necessarily address these issues. They should prove that the technology is attractive to users, that it will indeed be adopted and that it will make a positive difference to disaster management.