A Comprehensive Review of Interpretability in AI and Its Implications for Trust in Critical Applications
Ujjwal Singh Kathait, Anamika Rana, Rahul Singh Chauhan, Ruchira Rawat · 2024
With the advent of deep learning (DL) and other advancements in machine learning, artificial intelligence has recently taken the globe by storm. The requirement for openness and interpretability has arisen as computing power has reached superhuman capability; its absence is a serious problem in industries like healthcare and finance, where trust is crucial because of its enormous impact on human lives. This necessitates more interpretability, which often means understanding the algorithmic system. Unfortunately, there are still a lot of unanswered questions regarding the DL. Our knowledge of machine decision-making is still incomplete. We provide an overview and categorize the interpretabilities provided by different studies. However, there is a significant disadvantage with AI: people would see it as a “black box,” which would erode trust in its trustworthiness. In a field where decisions can literally mean the difference between life and death, this is a severe issue. Considering different application domains and activities, this study offers a thorough overview of the literature on the latest developments in XAI approaches and assessment metrics.