Bostrom's introduction into the problematics of Artificial General Intelligence (AGI) is a very good first stop for newcomers to the topic. The material focuses mostly on existential risk factors and mitigation strategies. The book gives also some basic historical context and the barest minimum background on the more technical side of the AGI research, architecture etc.Recommended for general audience as this is an essential topic and the books is overall very competent.That being said... I was very, very disappointed by it... For those that are already familiar with this area there is very little new here. It is a good, but not great summary of the risk/strategy aspects but is deeply lacking on all other aspects (decision theories/interaction with the objective function/Lob's obstacle etc.). Many of the most fascinating issues are completely omitted, I guess understandably given the main aim of the book, but still... As a long time fan of Bostrom's writing this was also a surprisingly slog to read through. The most important parts are essentially re-telling of Yudkowsky's and Hanson's work. There is very little original Bostrom - and as example,his own Orthogonality thesis is essentially just stated, with the argumentation being very cursory and incomplete.As said - I do recommend it generally, but can't say I'm not disappointed. It really should be 3 stars, but I want to encourage its wider reading just because of the import of the topic. In the first few chapters of this book author analyses possible outcome of developing artificial intelligence that ultimately explodes into superintelligence. In most cases outcome is much more frightening than dystopias described in even scariest sci-fi novels.The rest of the book is dedicated to detailed analysis of ways to control superintelligence explosion so that outcome is more favourable.Anyone who associates artificial intelligence with nice and friendly robot from "Short Circuit" movie must read at least initial chapters of this book. And it is a 'must-read' for anyone working on AI field.
What do You think about Superintelligence: Paths, Dangers, Strategies (2014)?
