Demystifying XAI: Understanding of Applications, Limitations, Challenges, and Future Perspectives

Artificial Intelligence (AI) appears to be omnipresent, but its usability is currently a pain point as seen with automation bias. The General Data Protection Regulation (GDPR) has started solving this problem. The premise behind Explainable AI (XAI) is described as making AI decisions interpretable, gaining user trust and robustness of applications in healthcare and finance. This paper gives a survey of recent XAI advancements, providing types of algorithms, and examples of tasks. Despite the concept of XAI, we still have the black-box issue of AI, and especially in critical areas like healthcare with sensitive information, we need to protect user privacy. It is proposed that XAI can expose the decision-making process of the AI, to make the AI decision-making more reliable and trustworthy. Evaluation metrics and potential future directions for XAI are discussed, with a guiding principle: AI needs to be ethical, trustworthy, and user-friendly.