Actor Networks, Rare Events and Antifragility

In a recent blog I discuss some aspects of antifragility as suggested by Nassim Taleb’s recent book on Antifragility. Thinking a bit more about the nature of fragile and antifragile networks of relations could be of use in planning for rare events and their impacts. A well-aligned and well co-ordinated network of actors with a dense set of relations defining and binding their netowrk tightly may mean that the network is deeply embedded but this may be a disaster when a rare event hits. As I mentioned before, an event can illuminate the structure and relations in a network. A rare event, a major disruption, puts the spotlight on the fragility (or otherwise) of the web of relations. A well-aligned and co-ordinated network may function excellently for specific actants under ‘normal’ conditions, but in a rare, extreme event these relations may not be able to function. A dense network of relations may be too dense under these extreme conditions. The failure of one relation or the disappearance of one actant may produce a domino effect and trigger the unravelling of the whole web. A dense and highly focused actor network may be fragile to such disruption. A less dense and less well-aligned actor network may be at a disadvantage under ‘normal’ conditions but may have the flexibility to form new relations in disruptive events due to this weaker alignment and co-ordination of relations. Similarly, an actant with the flexibility to activate a different set of relations from the actor network it is usually associated with may be more able to survive and thrive in an disruptive event than a more specialist and network dependent actant or even a whole network.
If correct, then the above suggests that the density (and strength) of relations that define an actor network as well as the specialisation of actants will affect the fragility and antifragility of this network to rare events. Where an actor network has dormant relations, ones that are either unnoticed or unused during ‘normal’ periods, then there is a chance that the actor network could survive by activating these relations in times of crisis. The actor network that emerges, however, would be different from the one that entered the crisis. The dormant relations would now be known to the actants and be active rather than passive. The current banking crisis could be viewed in this light. When the crisis hit the usual sources of safety in the network failed. It was only when the dormant relationship between finance and the state was explicitly activated to prevent those ‘too big to fail’ from failing that some degree of stability was felt by the financial sector (OK oversimplifying like mad but you get the idea). But now that dormant relation is clear and present, everyone knows about it and the new financial network is being constructed with that relation in clear focus and all the issues of moral hazard and tax-payer bail-out that it brings.

There is an assumption in the above, however, that all rare events are the same. This is not necessarily the case as a recent paper by Lampel, Shamsie and Shapira (2009) in Organization Science (you need an account to access the journal). The paper ‘Experiencing the improbable: Rare event and organizational learning’ is a brief summary of the ideas in the special issue of papers on rare events and organizational learning. Importantly, they provide a four-fold classification of the types of learning that rare events produce in organizations based on the potential relevance of the event and the potential impact as in the table below.

                                                                       Potential Impact


Potential Relevance                                 High                          Low

High                                               Transformative              Reinterpretative

Low                                               Focusing                       Transitory


Table: Types of learning associated with rare events
Leaving aside the detail of the table (the subject of future blog!), the idea that a rare event has different affects depending upon the nature of the organization it impacts upon can be translated to actor networks as well. A rare event that is high on both criteria will have the potential to transform the nature of the network. In this case the points about relation density, dormant relations and actant characteristics are highly relevant. These are the rare events that can expose antifragility. A rare event with high potential relevance for a network but low potential impact (such as near-misses) can act as a means of forces reinterpretation of the current web of relations. The impetus to act on reinterpretation will, however, be determined by the interests of the actants and the ease with which the relations that define the network can be altered. Effort is required to overcome resistant to change in the absent of an event that causes transformation. If handled appropriately though this type of rare event could enable the actor network to alter and so improves its robustness or even atnifragiltiy to rare events without having to go through the pain of a transformative event. Drawing the lessons from such events and finding the will amongst key actants is however a major barrier as it is likely that no-one organziatino can affect such leanrign on its own - a sector-wide or even government-led inititative maybe required. A rare event that has high relevance but low potential impact for a network can, similarly, focuses attention on specific issues and problems within the network. Once again, however, change will depend upon who defines these problems and the willingness or ability of actants to alter the relations that define the network.

ACTANTS AND ASH CLOUDS

The risks raised by the ash cloud that swamped Europe in April/May 2010 could be thought of in terms of a set of actants (things, people, institutions anything entity really that has the ability to act upon and be acted upon by other entities). Relations between these actants are not fixed but change as the interactions between the actants change. Some relations and actants are harder to change, more entrenched, than others but all are capable of change even if this change is more painful to some than to others.

Figure 1 illustrates the main actants involved in the ash cloud story. The actants are presented as simple boxes but this hides a great deal of differentiation within each box. All airlines, for example, are not the same and or, initially anyway, were they response to the ash clod. Some airlines complained bitterly after a few days of grounding, others took the air in uninstrumented flights to ‘prove’ the safety of the airspace. Likewise, the government is likely to have had different factions pushing for grounding and for letting flights take place. All the actants relations end up focusing on airspace, the theatre in which the drama is played out.


Figure 1 Main actants in ash cloud drama

Importantly, none of the boxes is isolated; many of the boxes are intricately interlinked. Some of the links are relatively straight-forward. The Met Office and CAA, for example, are linked in a very formal manner. The CAA have set criteria for dust concentrations deemed safe. The Met Office provided that information based on computer modelling and data from instrumented flights. The Met Office may also provide the CAA with information on hazardous weather conditions but again the link is formal and highly structured. The link between the met Office and government is more of an economic link, the government paying for an impartial service, whilst the CAA has a regulatory link to the government in setting the legal parameters of responsibility for the airlines. Links need not be singular in nature. The airlines pay tax to the government (economic link), but also lobby on environmental issues and apply pressure when they interests are threatened.

The whole network trundles along, changing and developing as the actants interact, each trying to make the whole network function for their benefit. Each actant has a role in the network. The Met Office has a ‘scientific’ role of monitoring, the CAA a regulatory role, the airlines an economic role. This does not mean to say that each actant will not press into service different aspects of their character in pursuit of their goals, in their attempts to align the network and how it operate to their benefit. The Met Office tries to monitor the ash concentrations, measure and characterize the ash partilces and transmit this information effectively to all actants. The resultant grounding of flights, based on the CAA interpretation this information, meant the network wasn’t functioning in a manner that matched the desires of the airlines. The airlines tried to usurp the role of the Met Office by undertaking their own ‘tests’, flying unistrumented planes into the ash cloud and then transmitted this information through the network and beyond. The airlines tried to take on a role where they collected and transmitted information about the ash to parts of the network where that information could be understood in a way that benefited them. The general public could understand a plane going through an ash cloud and coming out the other side – could they understand complicated mathematically models that predicted ash concentrations? The airlines played to the general public, part of a wider network, to influence the government and CAA, part of the immediate network focused on the UK airspace.

Expanding the network out, it is relatively easy to include other actants (Figure 2). The CAA insisted that they were setting limits based on advice from VAAC and engine manufacturers. It didn’t take long before the economic relation between engine manufacturers and airlines resulted in the release of new information from the engine manufacturers as to the limits of operation in ash. Similarly, the wheel network could be expanded out to include the general public. There is however a danger with this type of analysis. You must always be aware that drawing a box around group doesn’t mean that that group is real or that that group is static. Entities evolve and are differentiated. Airlines are not all the same nor they necessarily behave in the same way to each hazard that they encounter. Likewise, the general public will not necessarily act as a mindless mass if given certain information. What this type of analysis does do is to help to clarify what entities are involved, how they are related and how they use these relationships to try to align the whole network to their benefit.


Figure 2 Expanding the network: VAAC and engine manufacturers

Hazards: Rational Choices

The dominant approach to the study of hazards doesn’t mean that the action of people can’t be studied, just that a particular type of behaviour is often expected – rational behaviour. Given a set of choices, you will behave in a certain way and numbers can be put to that predictability. It may be that this behaviour will be modelled statistically, 95% of the time you will choose A, 5% of the time B, or 95% of individuals in a particular set of circumstances will choose A, only 5% will choose B. you get the idea. Assuming the behaviour of individuals is predictable given a context, means that responses can be modelled and planned for.

An earthquake hits a major urban area in the US. You don’t really want the authorities to spend time trying to second-guess what people will do, you really want them to use their experience and insights from experts to rapidly rescue people. A plane shudders to a halt during take-off and the smell of burning fills you nostrils. Aren’t you glad experiments and computer models gave engineers the answers as to how people behave in such a situation and so where to put the exits to try to get as many panicking people out as possible.

But if people are rational then why are seemingly irrational choices made everyday and everywhere? Why do farmers persist on farming on the flood plain in Bangladesh, why are houses still built on the flood plain in Britain, why do fishermen set to sea when the weather forecasts a hurricane? These decisions can still be viewed as rational. Take the fisherman and the hurricane, a classic problem in rationality used by Burton, White and Kates back in the 1970s to illustrate the behaviour of individuals in hazards.

They suggest that understanding rational behaviour may be better understood if it is represented as a series of possible choices or alternatives given a particular or expected state of nature (Figure 1). For each state of nature and alternative action available, the individual will judge the consequences of their actions and choose the most rational option. The individual has to appraise what the state of nature is, not the easiest thing to do, as well as be aware of the range of alternative actions available. Assuming that the researcher can limit this choice depending on the individual’s experience and circumstances there are still problems with the simple application of this model.



Figure 1 Hazards, choices, state of nature

The individual and even the researcher may be operating in complete ignorance of how nature operates so there is complete uncertainty. In this case Burton, White and Kates suggest that ‘expected utility’ will be the rational mode of decision making – thus giving away the debt to economics and economic reasoning that this view of individuals has (Figure 2). In this case the numbers in brackets represent the payoffs of each alternative action. Rationally, the fisherman will be expected to remain at sea and fish.


Figure 2 Rational chocies for fisherman

Where there is a known probability of an event occurring then the payoffs can be altered to reflect this as in Figure 3. Here the 40% probability of a hurricane (0.4) changes the likely payoffs of each alternative action. (For the remain, no hurricane cell, for example, the new payoff is now the probability of no hurricane or 0.6 multiplied by the old payoff of +2 which gives a new payoff of +1.2). The remain option is still the rational one but the difference between it and evacuating is now a lot less, particularly if the payoff of 0 for the remain option is considered if the fisherman is proved to the wrong and a hurricane does happen.


Figure 3 Choices based on probabilities of events and alternative actions

The fisherman may not think in terms of scientific probabilities but may apply their own experience knowledge and subjective reasoning to the problem. This may alter the payoff again as in Figure 4. In this figure the fisherman has a high expectation of a hurricane, translated to a probability of 0.9. What this is based on is open to quesiton. It may be a general view amongst fisherman that hurricanes are likely at this time of year in this place or it may be more personal - a childhood memory of a major hurricane clouds the perception of an individual. Whatever the cause of this perception it is somehow translateable as a probability. Now it is clear from the matrix in Figure 4 that evacuation is the rational option but based on the subjective probabilities. This type of decision making might not be classed as rational by some experts. If the level of regret is considered, then the fisherman might evacuate at the first hint of a hurricane rather then even consider weighing up options.


Figure 4 Choices based on subjective probabilities

Do people act like this? Do you carefully weigh up the alternatives available to you every time you have to make a decision? Do you consider all the information available to you? Who you are, where you are, what you have been through have no bearing on what decision you make? Being flooded out one year has no bearing on what you do this year? Including or even working solely with subjective probabilities as in Figure 3 may seem a way around this problem of seeming irrationality in decision making but is it really just a fudge? Subjective probabilities still implies that a number, a probability, can be assigned to every alternative and that that number is based on something (and possible even consistent through time). People are a lot more annoying than that – after a decision I am sure virtually everyone can justify having made that choice. Ask them at the time, presuming they have time in a major earthquake with masonry falling all around them as they drag their family to what they hope is safety – and that individual will not be able to tell you why they make a certain decision and not another.

Most decisions are taken with limited information acquired because the individual is who they are and in the circumstances they are. A poor resident of New Orleans has different access to information sources than a rich resident. Information is not action. Individual will interpret and then act (or not) upon information in different ways. These differences could be dependent on their background, their class, their access to resources (real or perceived), their belief systems and a whole manner of complex and interacting factors that might just be amendable to statistically modelling (e.g. people from socio-economic group A are more likely to act in way B as they have access to resources, better information, more insurance, etc) but are unlikely to tell you why a specific individual, in a specific stressed situation made a specific decision. In a disaster the specifics are vital are they as important before and after?

The classic text on hazards by Burton, Kates and White is worth reading to help to understand these ideas.

The Environment as Hazard by Ian Burton, Robert Kates and Gilbert White (1993)