Abstract
Global climate changes increase the risk of extreme weather events, posing a threat to our society. To reduce the impact of such events, efficient emergency management is crucial. Extreme weather events are characterized by a myriad of information from various sources which emergency managers must systemize and understand to make the right decisions. Automaton can be utilized to ease the demands on emergency managers by developing a decision support system. However, automation of prior human tasks is challenging and trust in automation is essential to deal with these challenges. The aim of this thesis is therefore to explore (1) emergency managers’ reflections about a decision support system in the context of extreme weather and (2) how learned trust in the model from K. A. Hoff and Bashir (2015) works in this context. As there are different ways to develop decision support systems, the differences between a system based on machine learning and a traditional system has been investigated within these two aims. Ten participants tested a prototype of a decision support system, half tested a traditional system while the other half tested a system based on machine learning. To investigate the first aim, they were interviewed in a semi-structured approach using the SWOT format which was analyzed inductively. The second aim was investigated using a specific interview guide to capture learned trust, which was analyzed deductively. The themes from the inductive analyses were (1) aspects and characteristics of the system, (2) users of the system, (3) operational context, (4) interaction with the system and (5) decision making. The results provide valuable insight for further development of decision support systems. Moreover, the deductive analysis indicates that learned trust from the model of trust by K. A. Hoff and Bashir (2015) is relevant in this context, with some suggested adjustments that should be further addressed. There were no major differences between the participants testing the two different systems. This thesis is an early, explorative approach based on a limited sample. Nevertheless, it contributes with insights to the field of new and complex automated decision support systems based on machine learning.