Abstract
Explainable Artificial Intelligence (XAI) tackles the rising demand for interpretable and trustworthy AI like systems, where users are able to understand and trust the machine decisions for critical applications. This paper discusses XAI techniques, described architectural designs, and discusses their influence on complex decision making systems in applications such as health care, finance and autonomous systems (e.g., self-driving cars). We survey interpretable models and post-hoc explanation methods. Our approach uses theoretical analysis and case studies to probe the effectiveness of XAI at enhancing user trust and system reliability. Use cases such as medical diagnosis, credit scoring, and self-driving cars underscore the importance of XAI in liability and equity sensitive domains. Experimental results demonstrate that XAI can improve the interpretability of the original network structure without having an adverse-effect to the accuracy of the model up to 90% of user trust in healthcare tasks. Challenges are dealing with explanation versus efficiency and standardizing the evaluation metric. This paper highlights the potential benefit of XAI methods in connecting state of the art AI systems to human comprehension, which becomes the necessity for more accountable decision making systems.