As the other review comments provide good summaries of the book, I will focus on what I most enjoyed about this book. Erik's explanation on the nature of inference is a joy to read. And you don't need to be a philosophy major to follow his narrative or enjoy it. As familiar as I already was with the basic concepts of deductive reasoning and inductive reasoning, I found it very helpful to read Erik's chapters on these topics. He beautifully illustrates how deductive reasoning was (and still is) the core of symbolic AI (Theorem proving, Rule-Based Systems, Logic-Based Systems) and how inductive reasoning is the core of contemporary data-driven machine learning (large data sets + deep learning methods). With this backdrop, he clarifies the meaning and importance of the third type of reasoning, abductive inference, and provides a historical origins of this form of reasoning.Erik takes the reader through the journey of understanding that abductive inference really is a third type of inference and cannot be realized by further scaling and enhancing inductive approaches (current and ever improving deep learning networks with ever increasing data sets), or by combining induction with deduction. His key point is that without the ability to do all three of these types of inference, the extent to which AI-enabled machines are able to get more capable and more "intelligent" will be inherently limited because they lack the ability to do abductive inference. He explains why this missing type of inference capability results in important limitations in the journey towards increasing degrees of human-like intelligence. I was not aware that the origins of abductive inference goes back to the 2nd half of the 19th century and the work of the American philosopher of science (and polymath) Charles Sanders Peirce. I very much enjoyed learning about this history. Larson's recounting of Peirce's history (and eccentric character) and contributions is sufficient reason by itself to read the book, or selected chapters.The book has three main parts. The content on the nature of inference is the 2nd part. There is also a beginning part where he talks about the history of the concept of achieving intelligence with machines-- going back to Bletchley Park and Alan Turing. There is a 3rd part that deals with the problems of believing that it is a "no-brainer" assumed inevitability that machines will soon or soon enough exceed human-like intelligence. These chapters are also well done, but my favourite part of the book was the 2nd part on the nature of inference and how that links to intelligence.In several parts of the book, Erik makes the point that now the common societal wisdom pendulum seems to have swung in the direction of believing that there are unlimited possibilities (in terms of degrees of intelligence) for machines, but restricted horizons for us humans. Given his observation that as of now and for the foreseeable future that only humans have the capability for abductive inference, he cautions that this type of increasingly common societal belief is not only misinformed, but dysfuntional, if not dangerous.Erik has built and continues to build AI-enabled software systems. He has a very sound understanding of what systems can and cannot do, and he is a tremendous fan of the remarkable progress that AI has made in very recent years. He clearly acknowledges this. At the same time, he helps the reader to understand that this truly remarkable recent progress has its limits, and the inability of current systems to go beyond whatever way of performing and combining deductive and inductive inference results in an inherent limitation on their ability to achieve certain aspects of human-like intelligence.