Resilient Digital Twins

Business & Information Systems Engineering, vol. 63, pp. 615–619

Abstract

Currently, we can witness contradictory expectations when it comes to IT.On the one hand, there is the belief that Artificial Intelligence (AI) and Machine Learning (ML) will solve most problems because of the abundance of data and sophisticated algorithms (LeCun et al. 2015).In this editorial, we use AI/ML to refer to machine intelligence, i.e., mixtures of Artificial Intelligence and Machine Learning.AI/ML can deal amazingly well with unstructured data (text, images, and video) as long as there are enough training data.On the other hand, the COVID-19 pandemic and rapid climate changes (e.g., the floods in Germany in July 2021) show that AI/ML cannot deal with disruptions.When there is a sudden dramatic change, predictive models will fail, no matter how much data was there before.Consider, for example, the impact of the COVID-19 pandemic on supply chains.Especially at the beginning of the global outbreak of COVID-19 in March 2020, supply chains failed because of the unpredicted demand for certain products (e.g., masks and toilet paper) combined with simultaneous restrictions for travel, work, and business.In times of disruption, the ''training data'' are not representative of the actual behavior of people and organizations.Although current AI/ML technologies outperform humans in many areas, tasks requiring common sense, contextual knowledge, creativity, adaptivity, and empathy are still best performed by humans (van der Aalst 2021b).In times of disruption, these aspects are essential.Datadriven algorithms have difficulties dealing with contextual information and data-points that are off-the-scale.Techniques for reinforcement learning can adapt automatically, but are too slow to respond to disruptions.Techniques for ''transfer learning'' only work if there are sufficient similarities.Hence, disruptions like COVID-19 and large-scale flooding show the limitations of machine intelligence.Does this mean that one should simply ignore data and models because they fail when things get tough?The answer is NO for two reasons.The first reason is that if there are enough data, and processes are in steady-state, AI/ ML technologies tend to function well and can take over many duties from humans.The second reason is that also in times of disruption, humans still need data and models to make good decisions.During the COVID-19 crisis, the importance of having reliable data became evident.Data were often incomplete and unreliable, complicating decision-making at all levels (Sa ´ez et al. 2021).The General Data Protection Regulation (GDPR) further delayed actions in the European Union (Karnitschnig 2021).Also, bureaucratic processes that relied on paper forms failed.Actually, COVID-19 helped to drive digital transformation by exposing pre-existing problems.What would normally take years, now only took weeks.Interestingly, also relatively simple models helped decision-makers (e.g., high-

Authors 3

1. Wil M. P. van der Aalst corresponding Aachen Lehrstuhl Für Informatik 9

   RWTH Aachen University

   Affiliation as printed

   Lehrstuhl Für Informatik 9, RWTH Aachen, Ahornstr. 55, 52056, Aachen, Germany

2. Oliver Fast Hinz

   Goethe University Frankfurt

   Affiliation as printed

   Faculty of Economics and Business Administration, Goethe University Frankfurt, Theodor-W.-Adorno-Platz 4, 60323, Frankfurt am Main, Germany

3. Christof Weinhardt

   Karlsruhe Institute of Technology

   Affiliation as printed

   Institute of Information Systems and Marketing (IISM), Karlsruhe Institute of Technology (KIT), Kaiserstr. 89-93, 76133, Karlsruhe, Germany