Why are Deep Learning predictions Greek to me? Towards Explainable Artificial Intelligence (XAI)

Abstract

With a rapid proliferation of Artificial Intelligence (AI), questions arise about its possible implications in wider sectors of the society. The benefits of AI are innumerable and massive particularly in education, finance and healthcare; however, the black box nature of its deep learning algorithm has raised manifold quarries 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 contributions have 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 to why do humans need explanations in the first place. Explorations in the topic further illustrate 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, responsible and trustworthy AI system. Future directions should therefore focus on an interdisciplinary approach. The collaboration of deep learning experts, designers, developers, users, academics, researchers, ethicists, philosophers and policymakers may conjointly address the problem. Besides providing solutions of technical nature, Explainable AI may be considered furnishing a tipping point for AI policy and governance frameworks.