Knowledge-Infused Learning : Neurosymbolic AI for Explainability, Interpretability, and Safety

This practical guidebook introduces an emerging field that confronts the opacity of current 'black-box' AI models. Knowledge-infused learning combines machine learning and symbolic AI to make AI safer and more explainable, which is vital for applications in human-critical contexts such as healthcare, law, and finance.

Knowledge-infused learning directly confronts the opacity of current 'black-box' AI models by combining data-driven machine learning techniques with the structured insights of symbolic AI. This guidebook introduces the pioneering techniques of neurosymbolic AI, which blends statistical models with symbolic knowledge to make AI safer and user-explainable. This is critical in high-stakes AI applications in healthcare, law, finance, and crisis management. The book brings readers up to speed on advancements in statistical AI, including transformer models such as BERT and GPT, and provides a comprehensive overview of weakly supervised, distantly supervised, and unsupervised learning methods alongside their knowledge-enhanced variants. Other topics include active learning, zero-shot learning, and model fusion. Beyond theory, the book presents practical considerations and applications of neurosymbolic AI in conversational systems, mental health, crisis management systems, and social and behavioral sciences, making it a pragmatic reference for AI system designers in academia and industry.