A fast and light read, yet nonetheless important, this book proposes that artificial intelligence (AI) will eventually achieve many, or even all, of the goals promised by developers and the press. But the authors maintain that this will not be achieved on our current trajectory, that is, focusing on just deep machine learning.
The book is not technical, but neither does it talk down to its readers. It would be a perfect book to hand to an overly excited chief executive officer (CEO) or to an enthusiast who believes that perfect self-driving cars will be here immediately. Reading this book will help temper enthusiasm, with a healthy respect for the challenges that remain.
All in all, this is a very good book. Using a balanced and intelligent approach, the authors clearly know the field very well and feature many extremely current examples. The book also includes 50 pages of detailed bibliography and suggestions for further reading.
Even though I agree 100 percent with the authors’ key points, I do have some concerns. Some of their arguments reduce to, “this sounds complicated, so current techniques are not powerful enough.” Not only have I been surprised too many times by the power of deep learning, I can even see that some of the cited failing examples have been made to work in the months since the book was published. The authors also occasionally play with statistics in unfair ways, for example, comparing self-driving cars that need intervention every 10000 miles with human drivers who have fatal accidents once every 100 million miles.
I would have been thrilled, too, had they offered a direct call to arms or suggestions for next steps in AI research. But such prognostication is well beyond the goals of this book, which is more focused on generalities and goes little further than a demand that AI incorporate innate knowledge and multiple domain-specific engines. That said, the book’s arguments and direction are sound--in some cases almost obvious after reading them. I do recommend this book as a quick nontechnical read on the challenges facing AI.
Let me end with a few beautiful quotes that capture the book’s spirit and elegant writing:
- “In the immortal words of Law 31 of Akin’s Laws of Spacecraft Design, ‘You can’t get to the moon by climbing successively taller trees.’”
- ”Start by developing systems that can represent the core frameworks of human knowledge.... Embed these in an architecture that can be freely extended to every kind of knowledge.... Connect these to perception, manipulation, and language.“
- ”The only way out of this mess is to get cracking on building machines equipped with common sense, cognitive models, and powerful tools for reasoning.... That project itself can only get off the ground once the field shifts its focus from statistics and a heavy but shallow reliance on big data.... [T]he royal road to better AI is AI that genuinely understands the world.”
- ”And the best way to make progress toward that goal is to move beyond big data and deep learning alone, and toward a robust new form of AI ... with values, common sense, and a deep understanding of the world.”
More reviews about this item: Amazon, Goodreads