Customer Reviews

Understanding Intelligence (Bradford Books)

5 star:		(1)
4 star:		(1)
3 star:		(1)
2 star:		(0)
1 star:		(0)

 

 

I liked this book because it not only gives a conceptual tour d'horizon on the new field of embodied cognitive science/New AI, but also provides tons of concrete (programming) problems to work with. The companion web site contains additional programming examples for download. Overall, this book gave me the much needed realistic perspective on this new field.

I'm a masters student in general CS doing a semester of school at the University of Sussex in England. One of the courses that I took used Understanding Intelligence as a reccomended text. The book covers in detail a lot of aspects of embodied artificial intelligence and of some of the conflicts between embodied AI and traditional symbolic AI. The book is easy to read and, at least for me, quite worth it. The only problem that I found was that the books website is obscurely documented in the book and is quite low budget. Other than that it's a great introduction to the field; I found it much easier to assimilate than the lectures from my class.

The new book UNDERSTANDING INTELLIGENCE by Pfeifer and Scheier contains many interesting ideas and is well worth reading though rather long (at nearly 700 pages). In many ways it is similar to the claim that "intelligence is a kludge" but this time more of a hardware kludge. There is also a substantial return to behaviorism though the references are missing. Perhaps the core belief of the book is that "intelligence is...a large number of parallel loosely coupled processes" (page 303). But it seems to me that if the modules are too loosely coupled the system can not be successful as a universal approximator and therefore will not be Turing equivalent. It might be possible to decompose the performance system of an agent into such loosely coupled modules (like Norvig and Russell, AI a Modern Approach, page 202) but for it to learn there must be extensive coupling to most if not all modules. While I agree that sensory motor contact with the world is what grounds symbols there is no need for every agent to perform its own grounding of every symbol. In fact there are modules (agents) in a typical subsumption architecture that are not, themselves, connected to either sensors or actuators. I do like the idea that intelligence is related to some niche. Humans and AIs may not occupy the same niche. Machines are better at statistics than humans while humans are better at emotion. The criticism of GOFAI in UNDERSTANDING INTELLIGENCE is often unfounded.