Survey and Special Session of the ESRA Technical Committee (TC) on Decision-Making under Uncertainty (DMU)
More than a look into the crystal ball
– experts get together at the ICSRS in Venice
The Technical Committee (TC) on Decision-Making under Uncertainty (DMU) was established two years ago with the aim of bringing together experts in the field of decision analysis. The starting point of the TC was the accelerating change that mankind is facing worldwide, e.g. increasing threats, environmental disasters, pandemics and scarcity of raw materials. Conflicting requirements in critical decision-making settings have to be balanced against each other, which is difficult in the case of novel challenges for which there is only incomplete knowledge and appropriate ways are lacking to fully account for the parameters influencing the uncertain outcomes to be evaluated.
TC DMU Survey
To help answer urgent questions related to DMU from policymakers, society and industry, the TC called for participation in an online survey. The survey also served as preparation for the special session on "Managing Uncertainty in a Volatile Interconnected World," held on Nov. 23-25, 2022 at the International Conference on System Reliability and Safety (ICSRS, www.icsrs.org) in Venice, Italy. The initiative aimed to capture expert knowledge on DMU through guiding questions, such as "How can we deal with uncertainty in decision making?" or "When can rational decisions be made and when not?".
The survey received contributions from experts coming from several different countries, including the United States, France, Switzerland, Italy, Belgium, Poland, and Germany. Participants to the survey were from academia as well as industry and government, and responses were in most cases encouragingly detailed. When evaluating the results, we found that people working in different fields have similar problems, under the challenges coming from the fact that systems are becoming more complex and environmental conditions more dynamic. To address these problems and challenges, quantitative tools are being developed to support DMU. In this context, for example, the application of data-driven methods and algorithms to assess events in real time is playing an increasingly important role. Scenario analysis, Bayesian networks and Monte Carlo simulation, for example, have been pointed at as promising methods to support DMU analyses. Favorable methods for multi-criteria decision analysis were also mentioned, e.g., PROMETHEE and AHP.
Participants had partly predictive ("What could happen?"), partly descriptive ("Why did it happen?"), and partly prescriptive analysis focus ("What should be done?"). To summarize the survey: decision making in the context of risk analysis is a very complex task because human, technical and environmental factors must be considered in an integrated manner. Knowledge on potential causes and effects is interdisciplinary and should be carefully evaluated. Many participants indicated that uncertainties can be measured and quantified using probability distributions and probability theory. A first step in decision making under uncertainty should be to ensure that the accepted risk value, boundaries and models are well defined during the analysis. There was clear unanimity that this is not an easy task and needs to be well thought out.
Special Session in Venice
For the special session in Venice, the TC on DMU invited Professor Ahti Salo as keynote speaker, a world-renowned and award-winning expert in decision analysis. This year, the Decision Analysis Society (DAS) of the Institute for Operations Research and the Management Sciences (INFORMS) awarded Professor Ahti Salo the most prestigious 2022 Frank P. Ramsey Medal for outstanding contributions to decision analysis.
On November 24, interested experts met at the ICSRS in Venice to discuss how they deal with uncertainty in their fields of research and application. Seven scientists presented their questions on DMU and possible answers. A Q&A session was opened after each presentation. Professor Ahti Salo opened the session with a keynote address, presenting his extensive work on the development of decision analytic methods and their application in resource allocation, risk assessment, reliability engineering, technology foresight, and efficiency analysis. He gave the audience food for thought on DMU: To deal with uncertainty, one must first understand the system and clearly define terms. No one has a complete picture, so collaboration between different disciplines is needed. Scenario analysis is a widely used tool in DMU. Not to be neglected is testing scenarios for consistency.
Lukas Halekotte, then, reported on his research project dealing with resilience at different levels of abstraction from general principles to performance. Dustin Witte continued with a presentation on modeling physical attacks. He focused in particular on the separation of epistemic and aleatory uncertainties. Danko Jerez gave a presentation on the design of cost-effective structural systems that can safely withstand dynamic environmental effects. He explained that reliability-based optimization (RBO) can be achieved through a two-stage Bayesian framework. The proposed approach converts the optimization problem into a sampling problem that ultimately leads to designs that follow an approximately uniform distribution over the optimal solution set. Chidera Amazu reported on the challenges of modern systems and the overwhelming demands on operators to cope with information overload. She presented tools for optimizing and managing process control and showed how the dynamics of the operational tasks can be tracked, e.g., via glasses with sensors. Tim Zander showed how neural network-based models can be well calibrated in terms of their predictive uncertainty. This requires first defining a calibration criterion against which to optimize. In particular, he pointed out the pitfalls of false positive and false negative results. Ingo Schönwand's contribution aimed to broaden the view of uncertainty in resilience management of infrastructures by using a political science approach. Based on the assumption that the embeddedness of infrastructures in society leads to irreducible uncertainty, the theory of policy analysis proposes to structure decision problems and to explicate their uncertainties in order to test possible strategies against a large number of scenarios. The chair of TC DMU, Professor Kai-Dietrich Wolf, concluded the presentations with a final discussion session. The following insights could be drawn from the special session: scenario analysis is an important basis for a structured view on possible futures. Artificial intelligence (AI) can contribute significantly to optimizing decision-making processes so that wrong decisions can be reduced. One issue is testing the validity of models and their adequacy to represent reality. The formulation of concrete decision criteria was also considered important.
Participating speakers were:
- Ahti Salo (Keynote), Aalto University
- Lukas Halekotte, German Aerospace Center (DLR PI)
- Dustin Witte, University of Wuppertal (ISS)
- Danko Jerez, Hannover University (IRZ)
- Chidera Amazu, Politecnico di Torino (MSCA-ITNCISC)
- Tim Zander, Fraunhofer IOSB
- Ingo Schönwandt, German Aerospace Center (DLR PI)
- Kai-Dietrich Wolf (Concluding Summary), University of Wuppertal (ISS)
We would like to thank all contributors and especially ESRA for their support!
Chair: Kai-Dietrich Wolf
Professor of Mechatronics Institute for Security Systems University of Wuppertal email@example.com
Co-Chair: Enrico Zio
Professor of Safety, Security and Reliability Politecnico di Milano and Mines ParisTech firstname.lastname@example.org