20 February this year a workshop on black swan risk was held at University of Stavanger Norway. The concept of black swan has gained a lot of attention recently and is a hot topic in many forums that discuss safety and risk. Also in the scientific community it has been focused in the aftermath of Nassib Taleb’s best-selling book The Black Swan (2007). The workshop addressed a number of issues linked to the challenge of conceptualising black swan type of risk, and how to confront it (assess and manage it). Some of the issues addressed were:
- How important is it to improve our hazard/threat identification methods?
- What type of improvements can be made?
- Can the use of alternative uncertainty representations add something to the analysis?
- Can signals and warnings be used to improve our assessment?
- Can red-team approaches be used? Scenario analyses? How?
- To what extent are robustness and resilience the answer to meet black swans?
- What does Taleb's antifragile concept add to this?
- Does risk assessment have a role to play at all in cases of deep uncertainties?
- Can we draw insights from other areas, like High Reliability Organisations (HROs), with its five characteristics: preoccupation with failure, reluctance to simplify, sensitivity to operations, commitment to resilience and deference to expertise?
About 80 people attended the workshop, most participants coming from the oil and gas industry. The workshop was the first in a series of seminars and conferences to be organised by the Black swan project, funded by the Norwegian Research Council and the oil and gas industry. The main aim of this project is to develop concepts, principles, approaches and methods for the proper understanding and assessment of risk, in particular major accident risk, for the Norwegian petroleum activities, that gives due attention to the knowledge dimension and surprises (black swans).
Prepared reflections on the topic were held by Terje Aven (Norway), Andrew Stirling (UK), Ortwin Renn (Germany), Enrico Zio (Italy, France), Seth Guikema (USA), Kjell Sandve (Norway), Malene Sandøy (Norway), Carsten Busch (Norway) and Ullrika Sahlin (Sweden).
Examples of black swan type of events and near black swan events were presented and discussed, both from the oil and gas industry and the public sector. Aven distinguished between three types of black swan type of events: unknown unknowns (a), unknown knowns (b) (events known to some but not others) and known events which are not believed to occur because of low judged probability (c) (all events having extreme consequences). He pointed to the fact that knowledge is the key to meet all these types of risk, and that this justifies further work on improving the practice of risk analysis as a main goal of risk analysis is to strengthening the knowledge dimension. It is however necessary to see beyond the current probability-based approach to risk and give stronger weight to the strength of knowledge that these probability judgments are based on, as well as the potential for surprises relative to the existing explanations and beliefs. How to best do this is a huge research challenge and is one of the key tasks of the Black swan project. Aven outlined some possible directions for the research, linked to inter alia hazard/threat identification methods based on creative thinking, red teaming and scenario analyses, and approaches integrating signals and warnings, adaptive analysis, ideas of the collective mindfulness concept, and robustness & resilience principles.
Andrew Stirling highlighted the importance of the precautionary principle in relation to black swan types of risk: uncertainty requires deliberation about action; deliberation produces more robust knowledge than probabilistic analysis - there is a need for seeing beyond the “science-based” approaches. As a basis for his discussion on how to confront black swans he used a model for structuring risk problems, having four main categories, depending on the two dimensions (axes) knowledge about likelihoods and knowledge about possibilities. The term incertitude is used to reflect the degree of problematic “scores” for these two dimensions - black swans we may experience for cases of moderate and high incertitude. Stirling concluded his reflections by providing some specific comments to the nine issues mentioned above.
Ortwin Renn pointed also to three different types of black swans; (1) the tail ends of a normal distribution (low probability or dependent probabilities); (2) events linked to problems in knowledge and lack of transfer of knowledge (asymmetry of knowledge distribution) and (3) unique events for which there is no historical record for a re-occuring patterns. These three types of black swans demand different and tailored methods of assessment and management. Tail ends phenomena need better assessment methods to characterize and evaluate rare events beyond the two or three standard deviation integral. Asymmetry of knowledge demands better methods to monitor knowledge, access new sources of knowledge and decrease communication hierarchies. Finally, unique events cannot be calculated or assessed but systems may vary to the probability that they can get under stress from unique events. Geological formations may not be very vulnerable, stock markets are probably highly vulnerable. The more vulnerable a system is the more important are management preparations to increase resilience. Vulnerable systems need to be enhanced so that they can sustain their functionality even when they are stressed by unknown events. However, this requires a delicate balance between resilience and efficiency. He underlined that highly efficient systems are usually not resilient. As Stirling he concluded that to make appropriate decisions there is a need for broad value judgments where due weight is given to the uncertainties.
Enrico Zio focused in his talk on complex socio-technical systems operating in an uncertain environment and the use of both deterministic and probabilistic safety assessment (DSA&PSA). His point of departure was that unknown plant vulnerabilities are left unknown in traditional DSA and PSA. Zio reviewed and discussed this challenge in relation to two basic different approaches to safety managment:
Safety I: Anticipation by explanation. The future can be described as a (re) combination of past events and conditions that have been problematic. It is possible to predict and calculate future risk events.
Safety II: Anticipation by imagination. The future has not been seen before. It involves a combination of known performance variability that usually is seen as irrelevant for safety (signals, precursors).
Seth Guikema discussed the use of historical data and related data-driven models in relation to risk and black swans. He underlined that historical data cannot identify black swans of the unknown unknown type (c), but they are still essential for the management of the black swan risk. Data-driven models can identify relationships between system response and hazard inputs. A strongly-validated, data-driven model can help understand the potential for Black Swan types of events if the initiating drivers can be imagined. Guikema also discussed the use of decision-focused, robust risk management planning. The idea is to generate a suitable set of scenarios, assess them using some performance model, and then select alternatives to maximize robustness and flexibility. Then one selects some new creative alternatives and monitor for signals and warnings – make some decision on alternatives and repeat the process.
Kjell Sandve and Malene Sandøy discussed some challenges related to the use of quantitative risk analysis (QRA). They highlighted that these assessments are often used as direct input to the decision-making, without proper reflections on the limitations of the tool producing the QRA results. The analysis is to some degree a black box, the result being that important decisions are in practice taken by technical experts (consultants). To meet this challenge there is a need for research that can bridge the gap between the management and the technical content of the analysis, for example the development of some type of overall assessment based on broad organizational discussions and participations. This type of assessment is seen as an interesting topic for the Black swan project.
Carsten Busch gave an example of a type of black swan event of type c) – judged so unlikely that it is not believed to occur – and discussed various means to confront this type of risk, including measures improving the robustness, resilience and antifragility. His summary of the workshop is available at http://heach.blogspot.no/2014/02/black-swan-workshop-20-february-2014.html.
Ullrika Sahlin drew a line from traditional statistics, via Bayesian analysis to the challenge faced in the project requiring a broader risk perspective which is better able to reflect the knowledge dimension and the unforseen. Inspiration on how to deal with Black swans can be found in climate and environmental management type of problems, in which evidence-based decision making is confronted by the need to evaluate qualitative aspects of knowledge and the knowledge producing process per se, under multiple sometimes conflicting decision objectives.
At the end of the workshop Aven reported what is coming next. He highlighted that different cases and decision-making situations will be studied, and that the project group will decide on which research directions to pursue in the near future. A follow-up workshop/conference is planned next year, where the ambition is to present some initial research results.