Customer Reviews

Chess Metaphors: Artificial Intelligence and the Human Mind

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

 

 

I bought this book with the expectation of reading something on the mind, AI and chess and ways in which they overlap. In particular how the study of chess might have impacted AI research and how chess programming uses ideas from cognitive analysis of chess players. This book was not about those things...

The book starts out with an overview of the brain physiology. It assumes no prior knowledge and goes through some of the evolutionary differences between humans and other species. It discusses the basics of neurons, how they communicate and some of the ideas behind how the brain might work. In particular some of the original neural network ideas and topological maps of mind and body. The book then moves into the mind and brings up cartesian dualism. It discusses memory, emotions, what makes us self concious. It is a mixture of psychology, neuroscience and philosophy. THere has been little discussion of chess up to this point other than some vague metaphors here and there. They might be accurate but they dont follow naturally and they definately dont evolve from the text, they are sort of forced in. The book then goes to discuss computation by computers and cognitive processes, some of the differences and similarities. Nothing too deep, just overviews and the difference between methods, particularly chip deductiveness vs human inductiveness. Memory recognition, patter regocnition. The author is building up the blocks to eventually discuss how excellent chess players seem to be able to compute so quickly, which is attributed to pattern recognition rather than superior backwards induction. This is backed up through scientific experiments

The book moves into chess more completely, discussing the history first. It then gets vague again discussing the beauty of chess and the intuition that is seen as beautiful when moves are made with incomplete information inferring the human insight and feel based on implicit pattern recognition. The book gives some quick overview of the expectation of a game in terms of duration and the work of de groot who did work with great players under observation recording the neuroscience results. The parts of the chapter dont fall together that well, some statements are made which are stretched about chess composition is equivalent to theorem proving (which its not unless considering symbol manipulation theorems). The final chapter goes into the composition of chess programs. The general ideas behind them, considering chess a 0 sum game, the values of players, the backwards induction, optimized induction based on database of postions. This is the first, AI/chess that is in the book, unfortunately it really doesnt get very deep. It describes briefly the programming intent to play moves based on maximization of some n variable analysis.

I bought the book hoping for a treatise on chess, how it has illuminated techniques to study in AI, how chess programming has incorporated aspects of our cognitive processes and the future of chess programming. Only in the vaguest sense that chess programming uses database retrieval and chess masters have vast knowlege on games and positions is there a correspondence. This book serves as a good overview of how the brain works and it has a lot of information on chess. It does not integrate them well. Its discussion of chess is also romanticized rather than quantified, which is inevetable given its complexity, but the author alludes to the beauty of a position too much for the objective audience. Properties of beauty are to be shown, not assumed. All in all, i liked reading parts of the book, they introduced some interesting concepts and some topics I would like to read further on, but I dont think this book really achieves its goals.

The author covers the waterfront since "Deep Blue" soundly beat Garry Kasparov but in a very much hit or miss fashion. Even with the announcement in the title that the book is about "chess metaphors," the organization of the book suffers from too much jumping about and too little apparent actual knowledge about chess thought per se. It is not entirely clear what he expected to accomplish in the book as a whole? The results are a kind of parallel analysis of several loosely connected topics that should they have been better integrated would have made for a much more interesting read: The key topics that were introduced and in some cases glossed over, were: how the brain works, the psychology of competitive board games, and a brief history of how computer chess has evolved, with a little AI thrown in at the end just for good measure.

Having studied, used and taught applications to neural networks, I was particularly interested in how chess-playing algorithms have evolved in this direction since my teaching days, but was not entirely enlightened by this rather uneven treatment of the subject. One thing that has become clear from the author's analysis is that chess-playing computers are still only as good as the humans writing the programs. We are indeed a long way from the era when a computer can simulate human thought processes, and then independently act on them -- which is the most often scenario called up in the imagination when one thinks of a computer playing chess machine.

Instead of a computer that thinks like a top echelon creative chess master, we are told by this author that what we get instead is one that substitutes inelegant power and brute electronic force (its multiple-ply search algorithms can evaluate up to as many as 200 billion moves per second!) for human creativity and intuition. This number-crunching behemoth that (again) we can imagine crashing along the chessboard like a medieval armored Knight, with over-sized clanging metal feet is not exactly what we had in mind as the replacement for the world's greatest chess masters.

As one who not only has watched this spectacle unfold over the last forty years (I still have an operational version of one of the earliest computer chess machine called "Boris," and all of the generations up through Sargon I-IV and had my class program a prototype chess machine using the BASIC computer language), I cannot say that I am pleased with the direction that computer chess has taken, or with the way this book has characterized that progress. Sadly, "brute computer force" seems to be the name, the beginning, and the end of the game.

Nowadays, unless you have memorized the first forty "book moves" for at least several dozen chess openings, you can expect to get soundly trounced every time (and very early out of the starting gate if you are the least bit careless). What I have had to resort to just to keep the games interesting, is pick one or two novel openings and use the machine as a heuristic tutor to deepen my knowledge about the variations in those particular openings. But even when you discover a hole in its logic, the computer still has a tremendous advantage. It can sense your level of play, and then proceed to set "book traps" for you. And then (if it should have erroneously underestimated your strength, which it rarely does), it still has time to recover and angle for a draw, which it can ensure itself 99.9% of the time even in the rare instances when it gets outplayed in the middle game. Risky moves and sacrifices, the very heart and soul of chess, are a "no-no" with chess computers of even the least sophistication. They are so accurate and so brutal and unforgiving that on balance it, and the tendency to seek out "drawish" endings, discourages the desire to continue playing them.

Unfortunately, the computer could care less about style or book theory, except in the openings (and there it plays with such mathematical accuracy that it can be disheartening). According to this author, it accomplishes its superhuman feats by simply using a very tight algorithm to count up the values of the chess pieces at any given position, analyzes a few million moves, and then re-computes them over and over again after each move.

I wish I had that algorithm, especially the positional component of the one most computer programs use. The fact that the computer is entirely free of prejudice and doctrine seems to give new meaning to a famous quip by the great Cuban Grandmaster Raul Jose Capablanca. When asked how many moves ahead he looks, Capablanca answered "only one, the best one."

Four Stars

Rasskin-Gutman hoped to write a book not about how to apply AI to chess, but create a sort of workbook of ideas and suggestions on what might be posssible to do so. "Basically, I have put together in a single volume those ideas, elements, facts, and surprises I would have loved to find in a bookstore 20 years ago.... If I have reshaped chess in new dimensions within the natural world and within the promising scenarios of AI, I have fulfilled my objective."

As a result, it is difficult to disagree with anything in A. Menon's excellent review of the book here on Amazon, and as Kasparov puts it in a superb review in the February 2010 "The New York Review of Books":

"... a book that achieves its goal of being an overview of overviews, if little else. The history of the study of brain function is covered in the first chapter, tempting the reader to skip ahead. You might recall axons and dendrites from high school biology class. We also learn about cholinergic and aminergic systems and many other things that are not found by my computer's artificially intelligent English spell-checking system--or referenced again by the author. Then it's on to similarly concise, if inconclusive, surveys of artificial intelligence, chess computers, and how humans play chess."

As a one time chess enthusiast who has become a patzer with almost no competitive spirit but a love for solving chess puzzles, the book has great appeal. I found it fascinating to see how different approaches have been used first to solve the problem of playing chess by computer, and second by the author's suggestions for how those solutions might be applied to other applications of A1.

The author makes that point in these words; ches offers "an unparalleled laboratory, since both the learning process and the degree of ability obtained can be objectified and quantified, providing an excellent comparative framework on which to use rigorous analytical techniques." It was very interesting for me as a general reader in A1 with a rudimentary knowledge of chess to learn about some of those laboratory results.

Reading the book and Kasparov's fine review (see first Comment) together presents general principles and specific examples that enrich the learning experience. For example,

"In what Rasskin-Gutman explains as Moravec's Paradox, in chess, as in so many things, what computers are good at is where humans are weak, and vice versa. This gave me an idea for an experiment. What if instead of human versus machine we played as partners? My brainchild saw the light of day in a match in 1998 in León, Spain, and we called it "Advanced Chess." Each player had a PC at hand running the chess software of his choice during the game. The idea was to create the highest level of chess ever played, a synthesis of the best of man and machine."

The book, then, comes down to a matter of taste. For A. Menon, the author did not draw out the connections well enough; for me, the "ideas, elements, facts, and surprises" were enough to earn four stars. Your own mileage, of course, is almost certain to vary.


Robert C. Ross 2010