A Comprehensive Survey on Explainable Artificial Intelligence: Methods, Challenges, and Future Directions
Karthik Wali · International Journal on Science and Technology · 2020
XAI has come to be an important research field when designing artificial intelligence systems that are oriented toward interpretability and explanation. As AI applications become used in critical areas like health care, finance and self-driven vehicles, people have realized the importance of having explainable AI. It aims to provide an overview of XAI in terms of its definition, methods, challenges and potential opportunities for future development. The methods of providing interpretability of the results are described, including model-specific and model-agnostic approaches and their strengths and weaknesses. We group them according to the type of machine learning model to which they can be applied and explain how the interpretability is improved. This paper also focuses on some main issues of XAI, such as the compromise between the forecast, interpretability and quality, the evaluation from a human perspective, and the governance regulations. Further, results will be discussed alongside the comparative study in other existing XAI frameworks and tools. Last of all, the weaknesses present in the current work and potential directions for future research are identified to advance the field for more interpretability and trustworthiness of AI systems.