Introduction

Commonsense knowledge is information that humans typically have that helps us make sense of everyday situations. As such, this knowledge can generally be assumed to be possessed by most people, and it is typically omitted in (written or oral) communication. The fact that commonsense knowledge is often implicit presents a challenge for automated methods in natural language processing and question answering as the extraction and learning algorithms cannot count on the commonsense knowledge being available directly in text. As such, commonsense knowledge and reasoning have been considered the “black matter” of AI, raising concerns about the trustworthiness and applicability of AI methods in automated and hybrid applications, especially social good applications in misinformation, traffic, health, and education. While large models are often hypothesized to have acquired various commonsense skills implicitly during their training on massive amounts of data, further experiments show that such skills are seldom robust, controllable, and communicable to users consistently, causing a lack of trust. This chapter provides a brief introduction and history of the research in AI on the topic of common sense. It unpacks commonsense AI through dimensions of knowledge and reasoning. It summarizes efforts to evaluate common sense abilities in AI models and discusses their present level of success. The chapter describes why common sense is still an open challenge and introduces four requirements for AI to be human-centric, i.e., adequate to augment humans on complex tasks. The chapter concludes with an overview of the organization of this book, which aims to bridge the gap between commonsense reasoning research and the requirements of human-AI teaming.