Loading...
L. Jhansi is an Assistant Professor in the Department of Computer Science and Engineering at Narasimha Reddy Engineering College, India. Her contribution focuses on explainable and trustworthy AI applications.
Artificial Intelligence (AI) has emerged as a crucial technology for creating intelligent systems that excel in prediction, categorisation, decision-making, automation, and optimisation. Nonetheless, most contemporary AI models, especially deep learning frameworks, function as intricate black boxes, rendering it challenging for consumers to comprehend the processes behind specific selections. This deficiency in transparency may hinder trust, accountability, safety, and acceptance, particularly in vital sectors like healthcare, finance, autonomous transport, engineering, and security. Explainable Artificial Intelligence (XAI) tackles this issue by creating techniques that yield significant elucidations of AI forecasts and determinations. This chapter elucidates the essential principles of explainability and reliable AI, while examining prominent XAI methodologies such as feature importance, surrogate models, LIME, SHAP, Grad-CAM, saliency techniques, counterfactual explanations, rule-based interpretations, and attention-driven approaches. The section elaborates on global and local interpretations, model-specific and model-agnostic techniques, equity, resilience, confidentiality, responsibility, and human-centric artificial intelligence. The utilisation of XAI in healthcare, engineering, finance, autonomous systems, cybersecurity, manufacturing, and smart cities is explored. Ultimately, the presentation encompasses issues and prospective avenues related to multimodal AI, foundational models, generative AI, causal reasoning, human-AI collaboration, and reliable autonomous systems.