In this post, Behailu Bekera, a 2nd-year PhD student in the SEED group discusses the role of robust decision making under deep uncertainties. This post was inspired by a reading of Louis Anthony Cox's "Confronting Deep Uncertainties in Risk Analysis" in a recent issue of Risk Analysis.
There is no one good model that can be used to assess deep uncertainties. Hence our decisions about complex systems or decision contexts are typically made based on insufficient knowledge about the situation. Deep uncertainties are characterized by multiplicity of future events and an unknown future. So, being able to precisely anticipate undesired events in the future and conducting the necessary preparations would be an example of a decision context with deep uncertainty. In this article, Tony offers recognition to ideas from robust optimization, adaptive control and machine learning that seem promising for dealing with deep uncertainties in risk analysis.
Using multiple models and relevant data to improve decisions, average forecasting, resampling data that allows robust statistical inferences despite model uncertainty, adaptive sampling and modeling, and Bayesian model averaging for statistical are some of the tools that can assist in robust risk analysis involving deep uncertainties.
The robust risk analysis techniques shift the focus of risk analysis from addressing passive aspects of it, such as identifying likelihood events (disruptions) and their associated [estimated] consequences to more action-oriented questions. Active questions such as how we should act now to effectively mitigate the occurrence or consequences of events with highly undesirable effects in the future.
Robust decision making, for instance, is used by developing countries to identify potential large-scale infrastructure development projects and investigate possible vulnerabilities that require profound attention of all stakeholders. Additionally, adaptive risk management may be used to maintain reliable network structure to ensure service continuity despite failures. This sort of techniques can be considerably important in the areas of critical infrastructure protection and resilience.
Through these emerging methods, Dr. Cox, makes important suggestions for making robust decisions in the face of extreme uncertainty in spite of our incomplete or inadequate knowledge. This will be an important paper for those looking to advance the science of robust decision-making and risk analysis.