Preferred Explanations of Algorithmic Decision-Making Systems in the Dutch Property & Casualty Insurance Industry

Abstract

There are multiple challenges relating to the explainability of algorithmic decision-making (ADM) systems, such as that there is not a clear consensus on what a ‘good’ explanation is and what sort of explanation fit the different types of ADM-systems. Therefore, in this research, the Delphi-method was used to study which explanation would fit the different types of ADM-systems best. This was done in the Dutch property & casualty insurance industry. This report will show that often the data that is used to reach a decision should be included in the explanation because of the privacy perspective. Also, for most ADM-systems in this study, can be said that the general idea behind the algorithm and/or data should be included in the explanation. The main conclusion, however, is that there is not a one-size-fits-all explanation for ADM-systems and that it depends on the type of ADM-system, for what it is used and in which social context.

Date of Award	9 Sept 2020
Original language	English
Supervisor	Deniz Iren (Examiner) & Stefano Bromuri (Co-assessor)