What do you think?
Rate this book
Blondie24: Playing at the Edge of AI
Blondie24 tells the story of a computer that taught itself to play checkers far better than its creators ever could by using a program that emulated the basic principles of Darwinian evolution--random variation and natural selection-- to discover on its own how to excel at the game. Unlike Deep Blue, the celebrated chess machine that beat Garry Kasparov, the former world champion chess player, this evolutionary program didn't have access to strategies employed by human grand masters, or to databases of moves for the endgame moves, or to other human expertise about the game of chekers. With only the most rudimentary information programmed into its "brain," Blondie24 (the program's Internet username) created its own means of evaluating the complex, changing patterns of pieces that make up a checkers game by evolving artificial neural networks---mathematical models that loosely describe how a brain works. It's fitting that Blondie24 should appear in 2001, the year when we remember Arthur C. Clarke's prediction that one day we would succeed in creating a thinking machine. In this compelling narrative, David Fogel, author and co-creator of Blondie24, describes in convincing detail how evolutionary computation may help to bring us closer to Clarke's vision of HAL. Along the way, he gives readers an inside look into the fascinating history of AI and poses provocative questions about its future.
406 pages, Paperback
First published January 1, 2001
5 people are currently reading
78 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 8 of 8 reviews
Read
March 13, 2011Excellent overview of a new paradigm: An excellent laymans' guide to a 'new' paradigm - I've bought two copies already (one given away).
Deep Blue beat Gary Kasparov but only after a multi-year effort involving special hardware and was a program that was specifically designed to play chess.
Blondie24 'evolved itself' using non-special hardware (Pentium 400MHz) and in a matter of months beats 99% of all human players - WITHOUT KNOWING ANYTHING ABOUT DRAUGHTS.
Part 1 of the book is an overview of the two philosophies of AI embedded in Deep Blue and Blondie24 respectively. The argument is well made that Hal2001 could never be programmed in a conventional fashion.
Part 2 is the story of Blondie's creation - if anything here the most impressive item is the author's enthusiasm which is contagious.
Three qualifications:
1) on a technical quibble front Blondie24 is actually a co-evolved board valuation function which can only play draughts as it is embedded in a conventionally programmed mini-maxing game playing program. However one can hardly quibble with an author who includes an appendix of criticisms of his contentions!
2) at a more fundamental level Blondie24 is not a learning program (as far as I can see it learns nothing from its games) but rather the output of a learning system. This may be an interesting area for future study as, having evolved a 'good solution' it seems a pity to then 'waste' its future experiences.
3) like most highly skilled people Fogel makes it look easier than it is. If, sucked in by enthusiasm, one is tempted to try for oneself such an approach one finds a great number of pitfalls - it ain't as easy as he makes it look :-)
Overall no hesitation in suggesting that you buy this book if you have a lively, wide-ranging mind. If you liked reading about any of Chaos, Complexity, ALife etc as new paradigms then you have to buy this one.
MRB
March 11, 2014
I would like to rate this book between 3 and 4 stars. It didn't blow me away, but the underlying ideas are very interesting and well presented. The book is presented as a story [no spoilers].
The main point he makes about AI in computer learning is important and profound. We have computers like Deep Blue to play chess, but these computers haven't actually learned anything. They are hand fed the human knowledge. Deep Blue can play chess but out of the box would fall flat on it's face in a game of Connect Four. Fogel well illustrates this failure in the advancement of computer learning, as well as why it's so challenging.
This book sits between technical and accessible, and balances pretty well. Everything is presented in a narrative, with some fun suspense and gradual introduction of ideas. He introduces some algorithms at a high level with optional footnotes for further reading. I didn't usually find the footnotes useful. As I have a technical background, I would have liked more in depth explanations of some things, leaning towards implementation details. The concept of machine learning is rooted in a technical background. I don't think a lay person would pick up this book with no prior computing experience. Fogel could have taken more liberty with the algorithm descriptions.
Another minor critique is that some of the stories as part of the narrative are not exciting and perhaps could have been cut out. Most of them are very nice. Some things, like the typed out explanation that "LOL stands for laughing out loud" adds some dating to the book, and might warrant another print.
I would suggest this book to someone with an interest in machine learning. Fogel is a good, clear writer and there are many interesting and important ideas here.
The main point he makes about AI in computer learning is important and profound. We have computers like Deep Blue to play chess, but these computers haven't actually learned anything. They are hand fed the human knowledge. Deep Blue can play chess but out of the box would fall flat on it's face in a game of Connect Four. Fogel well illustrates this failure in the advancement of computer learning, as well as why it's so challenging.
This book sits between technical and accessible, and balances pretty well. Everything is presented in a narrative, with some fun suspense and gradual introduction of ideas. He introduces some algorithms at a high level with optional footnotes for further reading. I didn't usually find the footnotes useful. As I have a technical background, I would have liked more in depth explanations of some things, leaning towards implementation details. The concept of machine learning is rooted in a technical background. I don't think a lay person would pick up this book with no prior computing experience. Fogel could have taken more liberty with the algorithm descriptions.
Another minor critique is that some of the stories as part of the narrative are not exciting and perhaps could have been cut out. Most of them are very nice. Some things, like the typed out explanation that "LOL stands for laughing out loud" adds some dating to the book, and might warrant another print.
I would suggest this book to someone with an interest in machine learning. Fogel is a good, clear writer and there are many interesting and important ideas here.
March 31, 2018
Blondie24 is informative, humorous, and gripping. Fogel begins by covering the history of artificial intelligence both as it currently stands and as Hollywood represents it, showing where Hollywood's vision is slightly inaccurate or too far out of reach. He holds HAL (IBM - 1) from the movie 2001: A Space Odyssey as the ideal artificial intelligence because it is a learning machine that can think and feel for itself, but he says HAL is currently impossible. What we can do is create a learning machine.
Fogel goes on to describe a project he and his partner Chellapilla worked on - a learning checker's program that would eventually become Blondie24. They came up with the Samual-Newell challenge for their checker's program - could the program invents it's own checker features and rely on feedbacks only after a series of games have been played? In opposition to most others chess and checkers playing games before, Blondie24 does not start with pre-programmed rules nor does it solve it's problems by brute force like DeepBlue by calculating every possible boards as far as it can go. Instead, Fogel and Chellapilla used a neural net with very fews features associated with checkers and created random generations that would then compete against each other and evolve to a new better generation. After a certain number of generations, the computer would have developed it's own checker features without human inputs and 'learn' how to play checkers.
I really liked this book not only for the clarity of the topic's presentation, but also it had some technical details and references to more technical papers. This is important to me because I would like to try to recreate the neural net for tic-tac-toe (which Fogel did for his master's) and learn more by doing.
Fogel goes on to describe a project he and his partner Chellapilla worked on - a learning checker's program that would eventually become Blondie24. They came up with the Samual-Newell challenge for their checker's program - could the program invents it's own checker features and rely on feedbacks only after a series of games have been played? In opposition to most others chess and checkers playing games before, Blondie24 does not start with pre-programmed rules nor does it solve it's problems by brute force like DeepBlue by calculating every possible boards as far as it can go. Instead, Fogel and Chellapilla used a neural net with very fews features associated with checkers and created random generations that would then compete against each other and evolve to a new better generation. After a certain number of generations, the computer would have developed it's own checker features without human inputs and 'learn' how to play checkers.
I really liked this book not only for the clarity of the topic's presentation, but also it had some technical details and references to more technical papers. This is important to me because I would like to try to recreate the neural net for tic-tac-toe (which Fogel did for his master's) and learn more by doing.
Read
October 1, 2021A great collection of anecdotes that resulted in a state of the art shift towards deep learning. The main focus was talking about the ability to evolve a model over generations and the book painted that process in a really informative way. A bit dry around actual action and content but an interesting window into how a great mind thinks laterally in a way that pushes innovation forward.
April 3, 2019
AI, board games, psychology, evolution, natural history and scrappy product building - a good read! One of the few build I've read and enjoyed the appendix as much as the main book!
November 18, 2009
What is thinking? What is learning?
The story of a project with the aim of developing a learning artificial intellegence. Blondie24 is a program with a neural net and an evolutionary algorithm, which without much guidance learned to play checkers and crack a rating of 2000.
It never reached the high performance of programs that have been explicitly programmed with rules, but the way it reached its high function was extraordinary.
A set of weights were assigned to programs on a random basis, and the worst performing were culled; the remnants were used with slight random variation, to parent the next generation and so on.
It is interesting evidence for biological evolution and against the complexity argument of intelligent design, as the program was set up with the most elementary model of what checkers was. But I digress.
The programs were not "told" how good any move would be, or that they won or lost any game, but were selected on how they performed on a set of five games. Even so, natural selection produced a superior model of play. Rather than giving the programs rules, the rules were generated by inheritance from their slected ancestral lines over many iterations.
(BLONDIE??? The program was entered on a checkers game site, originally with a Obi-Wan Jedi name, but they found that males tended to get very nasty if they lost, so they switched their identity to a female. (They got some undesirable responses there too, but they got asked out on a lot of dates.) To play they would manually run the program to whatever depth it could go in the time limit, and enter that move against the human.)
The story of a project with the aim of developing a learning artificial intellegence. Blondie24 is a program with a neural net and an evolutionary algorithm, which without much guidance learned to play checkers and crack a rating of 2000.
It never reached the high performance of programs that have been explicitly programmed with rules, but the way it reached its high function was extraordinary.
A set of weights were assigned to programs on a random basis, and the worst performing were culled; the remnants were used with slight random variation, to parent the next generation and so on.
It is interesting evidence for biological evolution and against the complexity argument of intelligent design, as the program was set up with the most elementary model of what checkers was. But I digress.
The programs were not "told" how good any move would be, or that they won or lost any game, but were selected on how they performed on a set of five games. Even so, natural selection produced a superior model of play. Rather than giving the programs rules, the rules were generated by inheritance from their slected ancestral lines over many iterations.
(BLONDIE??? The program was entered on a checkers game site, originally with a Obi-Wan Jedi name, but they found that males tended to get very nasty if they lost, so they switched their identity to a female. (They got some undesirable responses there too, but they got asked out on a lot of dates.) To play they would manually run the program to whatever depth it could go in the time limit, and enter that move against the human.)
January 4, 2012
David Fogel is a son of Lawrence Fogel, a pioneer of evolutionary computation. In 1993 father and son founded a company that applies evolutionary methods to various problems such as detecting cancer in mammograms, a "dual-use" project funded by the US DoD. In the late 1990s Fogel and a coworker decided to try to evolve a program that plays a strategy game: they deemed chess too complicated, but checkers just right. There was a famous checkers-playing program in the late 1950s which played against itself to decide on the weight of various features, but the features themselves were hard-coded (i.e. a checkerboard is worth 11 points more if there is a particular combination of checkers in the last two rows; the number 11 is adjustable but not the combination). Fogel decided that this is not really artificial intelligence, since the features come from human experts, so he and the coworker evolved a neural network that chose its own features. They gradually improved the evaluation procedure and made the neural network ever more sophisticated, all the while avoiding hard-coding the features. Several times did they run the evolution process for a few hundred generations, and each time made the evolved neural network play against presumably human opponents on the MSN gaming site. Their latest and greatest network played under the nickname Blondie24, and earned an expert-level rating.
August 12, 2015
A must read for anyone new to AI. Exciting, fast paced, fun, clearly written, an all around good read.
Displaying 1 - 8 of 8 reviews