Humans have long aspired to automate both physical and intellectual tasks, but achieving the latter has proven challenging. Over the past seventy years, the study of Artificial Intelligence has seen cycles of optimism and disappointment. The successful development of General Intelligence, characterized by the versatile capabilities humans possess, could lead to a multi-billion dollar industry and broaden the scope of automation. Recent advancements in Machine Learning have sparked speculation that it may be the key to achieving General Intelligence. However, this book argues that the machine learning framework fundamentally conflicts with any reasonable definition of intelligence, risking the loss of vital insights from earlier AI research. A paradigm shift in perspective is proposed, akin to the philosophical transformations in the mid-20th century. The authors present a framework for General Intelligence and a reference architecture that highlights the importance of anytime bounded rationality and situated denotational semantics. They stress the necessity of compositional reasoning, facilitated by symbolic-numeric inference mechanisms rooted in category theory. The text outlines the practical requirements for real-world General Intelligence, critiques machine learning's shortcomings in meeting these needs, and offers a philosophical foundation for the proposed approach, including mathematical details and a research agenda to
