A Review of Explainable Artificial Intelligence: Taxonomies, Challenges, Implementation Frameworks and Future Directions

Abstract

The recent surge of interest in eXplainable Artificial Intelligence (XAI) within the research community has led to an increasing number of publications in this field. This interest stems from the critical need for interpretability of machine learning models, particularly in sectors such as healthcare and autonomous driving. Consequently, there is an urgent requirement for an updated review to keep up with the current trends in XAI. However, existing literature reviews pose challenges for newcomers, especially those without a computer science background, as they struggle to comprehend complex taxonomies while lacking information about available libraries and frameworks for practical implementation. In this paper, we aim to address these limitations by presenting a simplified taxonomy of XAI that is accessible to readers with diverse backgrounds. By breaking down complex concepts into more understandable components, we provide a comprehensive overview of XAI and its implementation libaries. Moreover, we discuss the challenges faced by XAI, including data privacy and security concerns, the complexity of implementation, bias and fairness issues, and the integration of AI systems with existing infrastructure and finally, we suggest future directions for XAI research, emphasizing the need for responsible AI development, interactive machine learning, AI-assisted education, and further advancements in healthcare applications.

Ähnliche Arbeiten