Design of a model for assessing accountability in a robotic process automation implementation

Abstract

Artificial intelligence is getting more and more advanced. In the future, robots equipped with AI will behave an act similarly as humans do, and this raises concerns about how they are going to be governed. However, scientists are alerting that not enough attention is paid to AI from a sociotechnical point of view. Intelligent systems can already behave in unexpected ways, make unfair decisions or treat data with bias. The consequences of this behaviour can directly harm humans, and researchers agree that much more research should be done.
One of the applications of AI is combining it with Robotic Automation Solutions. RAS is a form of business process automation that combines both AI and software robots. The simplest application of RAS is known as Robotic Process Automation.
More complex RAS solutions are not yet mature enough to be implemented widely. However, research shows that they may be mainstream in less than five years. This research will focus on RPA to be prepared for the more advanced solutions in the future.
Because of the newness of RPA and the multiple challenges currently not solved, it has been neglected the need to have a proper scientific methodology to model an RPA implementation. The objective of this thesis is to develop a model that allows to analyse the roles involved in an implementation of an RPA project and to assess the accountability of the roles involved.
The model is formed of four main phases. (1) requirements and analysis, (2) development, (3) testing and (4) deployment and governance. Each of these phases comprises different activities. The activities always have two roles assigned; a requestor and an executor. These roles have also been identified, and their responsibilities explained. Special attention has been paid to the accountability relationship between both roles in each activity. The research also covers a discussion of the types of accountability seen in the case studies.
Finally, the research also contributes to the views on how RPA would change with the increase and improvement of AI technologies. An analysis on how the model would change if AI becomes very advanced is introduced.
The practical contribution of the research has been (i) the creation of an IT artefact to assist and facilitate the implementation of RPA projects by detecting the activities and roles required, (ii) a methodology to identify and assess accountability relationships depending on the characteristics of the roles and activities involved and (iii) a discussion on how AI may evolve in the future and, more specifically, how the IT artefact will have to be modified to cope with it. The theoretical contributions have been (i) the implementation of the Action-Based Design methodology to create a model, (ii) an increase of the existing literature and knowledge about RPA and (iii) the use of the Action Design Research methodology instead of a more conventional Design Science Research approach.