Why are Deep Learning Predictions Greek to me? From Explainability towards Ethical AI

Abstract

With a rapid proliferation of Artificial Intelligence (AI) and advanced shift from Machine learning (ML) towards Deep Learning (DL), its reliance on critical decision-making has increased exponentially. The benefits of AI are colossal particularly in education, finance and healthcare; however, the black-box nature of its deep learning algorithm has raised manifold questions regarding explainability of its decisions. This in turn, casts doubt upon authenticity and accuracy of its decision-making process. Explainable Artificial Intelligence (XAI) offers a solution to the black-box problem of the deep learning algorithm. Explainable AI has evolved with the transition from rule-based learning towards more complex neural network predictions. Enormous contribution has been rendered by academics and researchers for the development of numerous Explainable AI techniques as a post hoc explanation mechanism. However, these techniques exhibit profound limitations and impediments. It is imperative to ponder upon as why humans need explanations in the first place. An exploration of the topic further illustrates the direct relationship between core ethical values such as transparency, accountability and reliability with Explainable AI.  Arguably, Explainable AI supports and upholds the necessary elements for formation of an Ethical AI system. Future directions should therefore focus on an interdisciplinary approach. The collaborative endeavor of designers, experts, developers, users, academics, researchers, ethicists, philosophers and policymakers may conjointly address the problem. Besides providing solutions of technical nature, Explainable AI may be considered as one of the enabling factors for an Ethical AI Policy and Governance framework.