Doctor of Philosophy (PhD), Wright State University, 2022, Engineering PhD
Artificial Intelligence (AI) has seen a surge in popularity as increased computing power has made it more viable and useful. The increasing complexity of AI, however, leads to can lead to difficulty in understanding or interpreting the results of AI procedures, which can then lead to incorrect predictions, classifications, or analysis of outcomes. The result of these problems can be over-reliance on AI, under-reliance on AI, or simply confusion as to what the results mean. Additionally, the complexity of AI models can obscure the algorithmic, data and design biases to which all models are subject, which may exacerbate negative outcomes, particularly with respect to minority populations.
Explainable AI (XAI) aims to mitigate these problems by providing information on the intent, performance, and reasoning process of the AI. Where time or cognitive resources are limited, the burden of additional information can negatively impact performance. Ensuring XAI information is intuitive and relevant allows the user to quickly calibrate their trust in the AI, in turn improving trust in suggested task alternatives, reducing workload and improving task performance. This study details a structured approach to the development of XAI in time-critical systems based on a design thinking framework that preserves the agile, fast-iterative approach characteristic of design thinking and augments it with practical tools and guides. The framework establishes a focus on shared situational perspective, and the deep understanding of both users and the AI in the empathy phase, provides a model with seven XAI levels and corresponding solution themes, and defines objective, physiological metrics for concurrent assessment of trust and workload.
Committee: Subhashini Ganapathy Ph.D. (Advisor); Sherif Elbasiouny Ph.D. (Committee Member); Asaf Harel Ph.D. (Committee Member); Victor Middleton Ph.D. (Committee Member)
Subjects: Artificial Intelligence; Design; Engineering; Experiments; Industrial Engineering