Ontology-Based Modeling and Analysis of Trustworthiness Requirements: Preliminary Results

Abstract

The advent of Artificial Intelligence (AI) technologies has made it possible to build systems that diagnose a patient, decide on a loan application, drive a car, or kill an adversary in combat. Such systems signal a new era where software-intensive systems perform tasks that were performed in the past only by humans because they require judgement that only humans possess. However, such systems need to be trusted by their users, in the same way that a lawyer, medical doctor, driver or soldier is trusted in performing the tasks she is trained for. This creates the need for a new class of requirements, Trustworthiness Requirements, that we have to study in order to develop techniques for their elicitation, analysis and operationalization. In this paper, we propose a foundation to develop such techniques. Our work is based on an Ontology of Trust that answers questions about the nature of trust and the factors that influence it. Based on the answers, we characterize the class of trustworthiness requirements. Among other things, this characterization supports the requirements engineer in defining thurstworthiness requirements, identifying the risks presented by the system-to-be, and understanding the signals the system must emit to gain and maintain trust.