<![CDATA[Artificial Intelligence (AI) is now viewed as an analyst and decision support in the field of engineering design review and program governance, but the benefit of using AI as such will be contingent upon the level of human dependency to gain from system reliability. This systematic review is a synthesis of 28 sources published from 1997–2023 about human-AI interaction, trust in automation, explainable AI, AI ethics, and algorithmic auditing, and maps these sources to the stages of systematic engineering design. Each source was assigned to one of the constructs of trust, reliability outcomes, types of interventions, and governance functions. From the synthesis, reliability, transparency, and feedback are the main drivers of trust in AI, having been identified as an appropriate design target by over 18 interaction guidelines distilled from well over 150 recommendations, a review of 127 trust in automation studies, and a review of approximately 150 empirical trust in AI studies. While explanations led to higher acceptance of correct and incorrect recommendations, cognitive forcing and confidence displays resulted in better calibration at the expense of usability. Transparency was mentioned in 73 of 84 ethics guidelines and a national risk management framework was recently established, in which governance has four functions. A trust-calibrated review architecture is suggested that combines the level of automation per phase, seven formalized metrics and auditable governance gates, providing measurable criteria for implementing AI in review processes with safety and cost implications.]]>
Copyrights © 2024
