San Diego Reader

“It’s humbling to be beaten by Blondie,” says David Fogel, who many times has stayed up late to play checkers with her.

I tell him I wouldn’t be humbled if Blondie beat me. I don’t want to be any good at checkers.

You shouldn’t be dismissive of checkers, he tells me. Yes, he realizes that the game lacks the snob appeal of chess. He himself characterizes checkers as being one step below chess. But checkers deserves its measure of respect, he says. He quotes Marion Tinsley, who is regarded as the best checkers player who ever lived. (He died in 1995.) “Chess is like looking out over a vast open ocean; checkers is like looking into a bottomless well.”

If a movie were made of 38-year-old David’s life, casting might call for a brainy-looking Tom Cruise. His brown hair slicks straight back. His face is clean-shaven, his complexion creamy. The blue eyes behind the glasses look directly at a questioner. This valedictorian of La Jolla High School’s class of 1981 (who skipped kindergarten) has given a lot of right answers in his lifetime.

David and his parents and younger brother, who all live in La Jolla, run Natural Selection, Inc. The company was founded in 1993 by David and his father, Larry, the year after David received his Ph.D. in engineering sciences from UCSD. Their offices are high up on the cliffs above Torrey Pines beach, close to the Torrey Pines Gliderport. Just as some clans produce successive generations of clerics or chiropractors, the Fogels have produced successive members of an unusual computer-science specialty. The term for what they do is called evolutionary computation. The technique simulates in a computer the twin Darwinian principles of random variation and selection. It’s used to find solutions to complex human problems. The Fogels, for example, have won grants and contracts to improve breast-cancer detection, design new drugs to fight HIV, and devise battle plans for the military.

The technique isn’t new; it’s been around for four decades. The 74-year-old Larry is one of its acknowledged pioneers. In 1960, he conceived of what he called evolutionary programming. But early computers weren’t fast enough to capitalize on the idea. Only within the last decade has it become practical. In 1997, David predicted in his keynote address to a conference on BioComputing and Emergent Computation in Skövde, Sweden, that as desktop computers become even speedier, evolutionary computation will become “routine.”

David is currently the best-known member of the family outside its professional circle. He has become something of a celebrity for inventing a program that uses evolutionary computation to play checkers at the expert level. The program is known as Blondie. That’s short for its official Internet name, Blondie24. If one measure of fame is being part of an answer on Jeopardy! then David has achieved it. “A computer programmed by Dr. David Fogel taught itself to play this game that includes jumping and crowning” was the answer that the Jeopardy! players were told. (And one of them did guess the question correctly.)

Note that the computer taught itself to play. While people have been using computerized algorithms (step-by-step problem-solving procedures) for game-playing since before David was born, no one else’s program has used them without being primed with openings, endgames, good moves, and strategies. David’s earlier ticktacktoe program, which he developed for his doctoral dissertation, produced a “merely” proficient player. Blondie wasn’t even told if she was winning or losing as she went on to become good enough to beat 99 percent of her opponents.

Both Blondie and the ticktacktoe program are products of this so-called “electronic” evolution. As it worked in Blondie’s case, a colony of computer-programmed, checkers-playing “parents” and “offspring,” each slightly different from one another, competed in game after game. Chronic losers were killed off; winners were allowed to reproduce. After thousands of cycles of play, Blondie was the result.

The Blondie persona was fabricated by David and his computer programmer, Kumar Chellapilla, who works as a senior staff scientist at Natural Selection. They competed on the Internet with people who did not know that Blondie wasn’t a real person. David and Kumar imagined a 24-year-old female graduate student in mathematics at UCSD — “single, attractive, and looking for a boyfriend” — and impersonated her when chatting with their Internet checkers opposition. They resorted to this ploy after noticing they weren’t getting as much action as they wanted at www.zone.com — known as “zone” to aficionados — when using the program to play as David1101 and Kumar1201 or even as Obi_WanTheJedi. They not only wanted the action; they needed it in order for their program to excel by attracting high-class competition of a wide variety. In sum, it learned by losing. That is how it evolved new and improved generations of itself.

They invented the Blondie character for another reason. They were tired of being on the receiving end of expletives that losers sent flaming through the chatbox. While they found that opponents with expert or higher ratings were gracious in either victory or defeat, that was not the case for less skilled players. David and Kumar figured that most of the sore losers were male. Maybe the guys would display better manners if they lost to a woman? Not exactly. Instead of being flamed, the duo’s Blondie started receiving requests for dates. She was also subjected to more than a few raunchy propositions. David and Kumar were so convincing as Blondie, they even chatted with other checkers-playing women, comparing notes on the various jerks they’d encountered on the zone website. “Girl power!” David would regularly remark to his sister sympathizers.

But how did their imaginary Blondie get so good at checkers? David and Kumar needed to come up with a story, so they elaborated on their theme. Not only an ace at math, Blondie surfed and skied. While recuperating after a skiing accident, she had decided to use her time to get really good at checkers. Eventually, Blondie, a work both of science and fiction, earned a spot in the top 500 of zone and won a tournament at another Internet address, www.playsite.com.

Earlier this year, David published a book, Blondie24: Playing at the Edge of AI — i.e., artificial intelligence. That’s the broad term for what David and the other Fogels do. (The book explains, in very readable form, exactly how Blondie was developed and how she works. It’s also a good primer on evolutionary computation in general and gives a concise history of the development of artificial intelligence too.) “To date, however, artificial intelligence has focused mainly on creating machines that emulate us,” David writes. “We capture what we already know and inscribe that knowledge in a computer program.”

IBM’s chess-playing Deep Blue is perhaps the best-known example of traditional artificial intelligence. When it beat world chess champion Garry Kasparov in the historic match in New York on May 11, 1997, it did so by evaluating 200 million chessboards every second. But while Deep Blue is “intensely good” at chess, writes David, it is “brittle.” That’s jargon meaning “good for only one thing.” It can’t do anything but play chess. It can’t make the first move in checkers. It cannot think for itself. It cannot adapt. IBM merely created “an illusion of intelligence,” in David’s words. “That isn’t what the dream of artificial intelligence is all about.”

Knowledge is a wonderful thing, David avers in Blondie24. “But learning is the key element missing from the majority of efforts in what’s routinely called artificial intelligence.” Programs that cannot learn “have nothing to do with intelligence; they instead merely recapitulate things we already know, just like Deep Blue does. Programs that are incapable of learning will never solve the problem of how to solve problems.”

“Where is the intelligence in an automaton like Deep Blue?” he asks. “A system that never learns, and has no capability of ever learning, does not deserve the description of intelligent.”

He regrets that, 50 years ago, pre-programming became the standard approach to creating artificial intelligence. He thinks the seeming triumphs of expert or knowledge-based systems are shallow, and hubristic.

Blondie, by contrast to Deep Blue, is “robust,” another jargon word, meaning “useful across a broad spectrum.”

True, Blondie can’t play at the master or grand-master level. But she could easily be fitted out to do so — by loading her up, as Deep Blue was, with human expertise. But then what? The point of the Blondie research was not to create the checkers equivalent of Deep Blue. The real trick, “the evolutionary thing,” as David is fond of saying, was to create a machine that is itself intelligent. Not only intelligent, but more intelligent than its creators.

It’s an unsettling notion for many people — that a machine could think of a solution that a human couldn’t. Unsettling, but a reality all the same.

“It’s already happened. Already been done, many times,” says David’s younger brother, 34-year-old Gary Fogel, who received his Ph.D. in biology from UCLA in 1998 and joined the company the same year. He would tell me this in the course of my interviews with all four Fogels over a period of days last December. “Even when you begin with the human expertise in a field, it quickly gets superseded by the computer. The evolution finds something better. Almost invariably the humans don’t know it yet. That’s just the way it is.”

< Previous Next >