To Learn or Not to Learn, that is the Question
The last of the four main capabilities of resilience – the fourth cornerstone – is the ability to learn from the past. It only comes last, however, because the four have to be listed one after the other. In practice, the ability to learn is just as important as the ability to respond, monitor and anticipate, as discussed in the Epilogue. A system cannot be called resilient if any one of them is missing and the absence of one cannot be compensated for by the increased quality or quantity of any of the others.
When the importance of learning for safety is considered, it is generally taken for granted that the basis for learning must be things that have gone wrong, such as incidents, accidents and catastrophes. This obviously makes sense from a classical safety perspective since knowledge of why things have gone wrong in the past is essential to prepare for the future. To learn from things that have gone wrong is, however, less reasonable when considered from a learning perspective, since it clashes with the basic principles for effective learning.
In order for learning to take place, three conditions must be fulfilled. The first condition is that there are reasonable opportunities to learn, that is, that situations where something can be learned occur with a sufficiently high frequency. (More precisely, the situations must occur so often that the lessons learned from previous situations have not been half-forgotten.) The second condition is that the situations are sufficiently similar to allow generalisations to be made, that is, they must have something in common or be comparable in some sense. People and organisations must be able to recognise that there is something in situation A that also can be found in situation B, not just in the manifestation of the outcomes but also in the reasons or causes. Finally, the third condition is that there must be sufficient opportunity to verify that the right lessons have been learned. This can be seen as a kind of combination of the first and second condition, in the sense that a comparable event must happen before the lessons have been forgotten – and hopefully well before the lessons have to be used in an actual event.
If we consider accidents, or even more extreme events such as emergencies and catastrophes, it is clearly important to find out why they happened but also clear that they do not offer the best basis for learning. Accidents do not happen very frequently, at least if the domain of activity is reasonably safe. Accidents are furthermore usually different from each other, and the differences are often proportional to the magnitude of the outcomes. Lastly, because accidents happen so rarely, there is little opportunity to check whether the right lessons have been learned. Accidents therefore do not provide good conditions to learning, common stereotypes notwithstanding.
It follows from these arguments that learning can be more effective if it is based on events or conditions that happen more frequently and that – almost by virtue of that fact – are less extreme and less dissimilar (e.g., Herrera et al., 2009; Woods and Sarter, 2000). Indeed, it is more efficient to learn from what goes right than to learn from what goes wrong (cf., the Prologue), because the former happens far more often than the latter. This position is also consistent with the basic principle of Resilience Engineering that failures are the flip side of successes and that they both have their origin in performance variability on the individual and systemic levels (Hollnagel et al., 2006: xi).
The four main capabilities of resilience are equally necessary and therefore equally important. Taking learning as a starting point, it is easy to argue that the ability to respond would be of little value without the ability to learn. A nation, a system, an organisation or an individual can, of course, adopt a set of predefined or stereotypical responses – and sometimes do. As long as the characteristics of the environment do not change, as long as nothing unexpected happens, the set of responses may be adequate. But unless the environment remains stable, or unless the environment can be completely controlled so that nothing unexpected happens, the pre-defined responses will sooner or later become inadequate. In other words, it will be necessary to learn new ways to respond, as discussed in Chapter 1. And it is by observing and evaluating the efficiency of the responses that the system (nation, organisation, individual) can learn.
A similar kind of argument can be made for the relation between learning and monitoring. The choice of which indicators to monitor can be based on formal or a priori models but it is ‘rare to find explicit models that provide a formal basis for identifying measures’, as argued in Chapter 5. It is mainly by learning through practice that the proper basis for monitoring – the indicators that must be watched – can be established. Yet simply adding new indicators whenever something has happened is not efficient in the long run. (A handy illustration of that is the way that most anti-virus software relies on a list of virus signature definition.) The efficiency of monitoring depends on the efficiency of learning, just as the efficiency of learning depends on looking at the right kinds of experiences.
Finally, learning is also necessary for anticipation. As Chapter 9 points out, one of the patterns of anticipation is that ‘resilient systems are able to recognise the need to learn new ways to adapt’. In relation to anticipation, learning is essential to produce a realistic, or even adequate, model or understanding of what may possibly happen in the future. This highlights the importance not only of learning, but of learning the right lessons, that is, of understanding what has happened in a way that is useful for the future functioning of the system. The worlds of industry, to say nothing of the world of politics, business, and finance, provide ample examples of how difficult this is (Touchman, 1985).
Learning is not a mechanical function, and cannot be reduced to data collection or statistics. For every use of learning it is crucial to learn ‘the right thing.’ But what exactly does that mean?
In relation to safety it is frequently pointed out that it is important to be thorough in learning. Given that learning traditionally has been based on things that have gone wrong, one advice has been to look for second stories beneath the surface (Woods and Cook, 2002). Unlike the traditional search for ‘root causes’, this does not mean that one should go as far back as possible, but rather than one should consider possible alternative explanations. Another advice has been captured by the phrase ‘what-you-look-for-is-what-you-find’ (WYLFIWYF) (Lundberg et al., 2009). The WYLFIWYF principle means that explanations of accidents are strongly influenced by assumptions about how different factors interact, that is, the accident ‘‘mechanisms.’ And since it is impossible to learn what has not been found, the corollary of the principle is that ‘what-you-find-is-what-you-learn’ (WYFIWYL).
The four chapters in this section each contribute valuable advice on how learning can be improved. Chapter 15 focuses on the obvious importance that the factual functioning of the system (where the events took place) must be transparent. This means that it is important to gather evidence about how the system functions, over and above looking for direct causes. In cases where things have gone wrong (accidents) or may go wrong (risk assessment), this is relatively straightforward and is often practised as an institutionalised process. In other cases, for instance when it is uncertain whether an organisation is sufficiently safe, the gathering of evidence is more difficult and may require a protracted process and investment of time and effort. Yet without the evidence, the situation cannot be properly assessed, and without that learning cannot take place.
Chapter 16 emphasises the importance of going beneath the ‘surface’, by illustrating how things may go wrong because of communication failures. It is argued that it is crucial to study coordination mechanisms in order to understand the resilience of socio-technical systems. This is therefore also an argument for extending the study beyond accidents and failures. The ways in which various entities of an organisation, in this case a health-care system, coordinate their activities and in particular how practitioners adapt to the unexpected using emergent coordination mechanisms, can most easily be seen in cases where the coordination succeeds and where therefore nothing goes wrong. Coordination is a function rather than a failure, but it is a function that can and must vary to match the working conditions.
Chapter 17 takes a closer look at the way data can be obtained, in this case from incidents. Incident reporting is not something that can be established by an edict, and successful incident reporting requires more than a simple set of tools or procedures. One issue is the pass criterion, that is, how easy it is to distinguish between important and negligible events. A second is the degree of standardisation or regularity of the work situation, since this determines how much information it is necessary to gather. A third issue is the visibility of events, and the question of whether sharp-end operators always are the best source of information. A fourth is how the heterogeneity of the community affects the scale of the reporting system. And a fifth issue is the current safety culture, which may determine whether reporting can be anonymous or not. All five issues are important for the basis of learning, whether it is narrow or extensive, how subjective and reliable it is, how complete it is, and when incident reporting will work or when it will not.
Chapter 18, finally, takes a closer look at how differences in cultures and occupations may affect what is considered safe and what is considered risky, hence affect what is learned. The chapter shows that there are clear variations in perspectives according to the national culture and occupation of respondents. People from different cultures and different occupations may therefore not learn the same lessons from what ostensibly are the ‘same’ events. This variety in outcomes is, however, not necessarily a weakness but may, following the arguments of Woods and Cook (2002) in fact be a strength, if only the organisation is able to make use of it.