On Intelligence

After the break for my (crazy) 4th of july in New York (have you seen the pics?) I’ve started again reading and I’ve completed “On intelligence” by Jeff Hawkins. This book tries to explain how the brain works and what’s the intelligence; unlike many other texts regarding the same topic, this one is more concerned with neuroscience than computer science. I’ve really liked this approach because, like author himself says, we first need to understand how does the brain works before starting building smart machine (if it’s possible).

The theory proposed is very fascinating, intelligence takes place in the neocortex (the most recent part of the brain) and it’s not based on massive computations (like the classical AI supporters believe) but on a memory framework which store all the patterns it receives from the outer world and simply “retrieve” them when it senses an input that match (at least partially) with them. This idea is called memory-prediction framework because it sustains that the mind always does predictions about the future based on previous experiences and all informations are hierarchically organized from the very specific detail, on the lower levels, to the general concepts, on the higher ones (did you know? You have at least a Bill Clinton cell!).

The arguments shown to support this view are very reasonable (see the “one-hundred step” constraint) and in my opinion this is, among all proposed ideas, the theory which is closest to the way the brain effectively works.

Looking for defects, I can say that the chapter which explains in detail the way the brain works could have been clearer, maybe more schemes and graphs would have helped the reader understanding where and how every function happens (I didn’t understand how invariant representations are stored, for example).
In conclusion the book is a pleasant read, it’s never too technical and (finally!) not very concerned with computer science matters; suggested to everyone who has ever wondered about intelligence.

The Emperor’s New Mind

I’ve read the Roger Penrose book “The Emperor’s New Mind” (finally, I’m so slow at reading!). First of all I’d like to say that it’s, in my opinion, composed by two (quite independent) books.
The first and the last 100 pages (of 600 total) are about the author ideas regarding AI, consciousness and free will. Penrose thinks that intelligence/consciousness cannot be the result of an algorithm, even if very sophisticated, so he affirms that these aspect of mind (not only of human beings but also for other animal species) are the result of non-algorithmic/non-computable events. 
Believing that brain mechanisms do not lie only on classical physics, Penrose introduces some (many in truth) concepts from quantum theory to support this argument. Therefore 400 pages (!!!) of the book are used to explaining fundamental quantum ideas needed involved in his theory. This is the main defect (in my opinion) of this book, to many pages are dedicated to physics lessons (starting from Newton’s mechanics!); maybe it’s my fault because I was only interested in the AI aspect of the book.

In conclusion I think that “The Emperor’s New Mind” ideas are interesting and deserve to be read, especially if you’ve read its rival book GEB (though, personally, I still prefer Hofstadter arguments). My suggestion is, if you’re not interested in a 400 pages overview of physics (although good), to skip that part or to read it at a VERY HIGH LEVEL without caring too much about vectors or functions!

Prey

Under suggestion of my friend Marco, I’ve read Prey, a Michael Crichton book. This book tells (very shortly) about an unemployed AI expert, Jack Forman, called to solve a problem with a program developed by him and applied to a population of nano-agents (the swarm) developed by a company in which works also is wife, Julia. The story is a warning about possible risks related to most active research fields of this years: nanotechnology, distributed artificial intelligence and genetic.
Altough it’s never too technical, the book is interesting also for people with computer science knowledge; I believe that the author has pretty well understood topics narrated, considering also book bibliography.

Definitively a story for both computer science expert and not, written in a very pleasant way.

Gödel, Escher, Bach: an Eternal Golden Braid

I’ve just finished reading “Gödel, Escher, Bach: an Eternal Golden Braid” a book by Douglas R. Hofstadter. The peculiar characteristic of this book is the constant comparison between Gödel’s theorem, Escher’s drawings and Bach’s music, I was surprised of all concepts shared by these works (there’s also a little bit of Zen!). The first chapters are about formal systems and the second part analizes if a set formal rules can express things like high level symbols or human languages. I’ve really enjoyed this book and it’s structure in particular: before every chapter there’s a dialog introducing the topic.

The author thinks that mind processes strictly obey to mechanical rules (at the lower conceptual levels) so they can be emulated by a machine (like a computer) properly designed; arguments proposed to support this theory are very interesting (the dialog about the ant system it’s great!). I agree with Hofstadter opinions in general, but I’d like to read also “The emperor’s new mind” (Roger Penrose) which presents opposite arguments regarding AI realizability. Another interesting topic of the book regards autoreference, especially related to consciousness, seen as a system which has a system for itself (it’s very summarized, GEB explains it much better!). Concerning the autoreference the Hofstadter has also written a law.

Hofstadter’s Law:
It [a task] always takes longer than you expect, even when you take into account Hofstadter’s Law

Anyway this book has also some defects (in my opinion), first of all length, I had the feeling that 800 pages (I’ve read the italian version) are too much and some topics were over-described (like the DNA code, for example). In addiction in final chapters some arguments don’t concern (strictly) with the AI problem (like Bongard problems or the ability of conceive hypothetics scenarios).

In conclusion it’s a worth of reading book, but don’t expect to agree with all ideas presented!