Date of Award
Master of Science (MS)
Applied AI has shown promise in the data processing of key industries and government agencies to extract actionable information used to make important strategical decisions. One of the core features of AI-enabled systems is the trustworthiness of these systems which has an important implication for the robustness and full acceptance of these systems. In this paper, we explain what trustworthiness in AI-enabled systems means, and the key technical challenges of specifying, and verifying trustworthiness. Toward solving these technical challenges, we propose a method to specify and evaluate the trustworthiness of AI-based systems using quality-attribute scenarios and design tactics. Using our trustworthiness scenarios and design tactics, we can analyze the architectural design of AI-enabled systems to ensure that trustworthiness has been properly expressed and achieved.The contributions of the thesis include (i) the identification of the trustworthiness sub-attributes that affect the trustworthiness of AI systems (ii) the proposal of trustworthiness scenarios to specify trustworthiness in an AI system (iii) a design checklist to support the analysis of the trustworthiness of AI systems and (iv) the identification of design tactics that can be used to achieve trustworthiness in an AI system.
Arinaitwe, Mark, "Towards Specifying And Evaluating The Trustworthiness Of An AI-Enabled System" (2023). Theses and Dissertations. 5227.