- Shopping Bag ( 0 items )
From Barnes & NobleThe Barnes & Noble Review
Having lived through the actual time period imagined in Stanley Kubrick's classic film 2001, one can't help asking: Will we ever create an intelligent machine like the unforgettable HAL -- one that's flexible enough to adapt its behavior to different environments, solving whatever problems come its way (though, ideally, not as neurotic)?
Moore's law is still working -- processors are doubling in speed every 18 months -- but, says, David Fogel, unless we fundamentally change our approach to artificial intelligence, we may never create HAL, or anything like him. Even Deep Blue, the IBM computer that outmaneuvered Garry Kasparov, is merely an incredibly fast calculator optimized for playing chess.
The alternative? It's straight outta Darwin: evolution. Stripped to its essence, evolution is "a never-ending two-step process of random variation and selection." Turns out you can write algorithms that do the same thing and begin building programs that actually do learn on their own, without being stuffed with knowledge ahead of time.
One of these "evolutionary algorithms," embedded in a program called "Blondie24," has learned how to play checkers well enough to win an online tournament. Blondie24's not at grandmaster level yet, but she's better than 99.61 percent of the 120,000 checkers players at Microsoft's zone.com.
Not bad, when you consider that Fogel and his colleague Kumar Chellapilla told Blondie24 nothing about checkers except the barest rules. Not even the fact that people take turns when they're playing a board game, much less the importance of mobility, the value of kings, or the existence of obscure formations like the "Triangle of Oreo," All that she discovered on "her" own.
If Fogel and Chellapilla are right -- and increasing evidence says they are -- then Blondie24, their new book, may be remembered as a seminal text in the transformation of AI. Yesterday's sterile systems could only reprocess what they'd already been taught. Tomorrow's might learn just about anything on their own -- how to build molecules that never existed before or respond to the sudden appearance of, say, alien monoliths near the moons of Jupiter.
Blondie24 walks you through the entire project: the history of checkers-playing software, the design ideas that led to Blondie24, the tweaking that helped her deepen her skills (without proffering any unfair help), and the exhilaration that came at every milestone -- beating a master, winning a tournament, making an especially prescient move. You can feel the pride. And you can begin to see the future. (Bill Camarda)
Bill Camarda is a consultant, writer, and web/multimedia content developer with nearly 20 years' experience in helping technology companies deploy and market advanced software, computing, and networking products and services. His 15 books include Special Edition Using Word 2000 and Upgrading & Fixing Networks For Dummies®, Second Edition.