POKER-FACED

Of the millions of gamblers who have rushed to play Texas Hold ‘Em and other fast-growing poker games online, Roger Gabriel isn’t the most intimidating.

The 30-year-old Newport Beach engineer started playing for money only a month ago. He lurks online at the tables for the chicken-hearted; even there, where the biggest ante is 4 cents, he can’t win consistently.

But Gabriel has a potentially powerful alter ego. In his spare time, he’s perfecting a computer program to go online and play the game for him.

His BlackShark software is still a work in progress, but Gabriel has no doubt that such programs eventually will be championship quality. “In the future,” he said, “robots are going to take over.”

Gabriel is one of an increasing number of computer professionals who design poker robots, or “bots,” that pose as human gamblers but can play endlessly without tiring or losing concentration -- for real money.

Though not yet good enough to beat skilled humans consistently, these programs are seen as a threat by online casinos -- all based outside the U.S. and out of the reach of American laws -- and the gamblers who spend billions of dollars chasing big pots.

“There are already lots of robots playing online, and that’s definitely unethical. They should identify themselves,” said Paul Magriel, a veteran professional poker player.

The march of the machines will be celebrated in Las Vegas next month with the world’s first money tournament for robots -- and the $100,000 prize is drawing a handful of coders out of anonymity.

The emerging technology does more than raise the stakes for real people and online casinos. It also raises fundamental questions about how far computers have come in mimicking and improving on human behavior, and about how far they can go in the future.

Computer programs have conquered checkers, chess and, most recently, backgammon. By rapidly evaluating plays more moves ahead than a person can, computers routinely beat the strongest human players in those games.

This was demonstrated most dramatically in the classic 1997 match between world chess champion Garry Kasparov and Deep Blue, a 1.4-ton supercomputer built by IBM. The machine’s victory marked Kasparov’s first professional loss, and many took it as a depressing event for mankind. Even Gabriel, then studying artificial intelligence at UC Irvine, had been rooting for Kasparov.

Backgammon programs, which had to adapt to the random element of dice, grew so good by the late 1990s that they changed strategic wisdom built up over 2,000 years, influencing how the best humans play the game.

But poker -- popularized recently by televised tournaments for pros and celebrity amateurs -- is a far more human game, one in which psychology matters as much as probability.

That’s why in poker there’s no such thing as an absolutely correct play, except in retrospect. If someone, or something, bets heavily with a lousy hand and everyone else folds, that was the right bet.

This makes poker bot design fascinating to academics like Jonathan Schaeffer, a computing science professor at the University of Alberta in Edmonton who for 14 years has headed a project to build poker programs.

Schaeffer said cards were more likely than chess to produce computing approaches useful in the real world because poker players must deal with incomplete information. But before such research can contribute dramatically to solving real-world problems, Schaeffer said, it has to solve the challenge of poker -- and that’s several years away.

For now, only the poker players with the poorest skills -- people like Gabriel, for instance -- have much to fear.

Typically, a user signs on to an online game site manually, launches the poker bot and lets it run. Gabriel’s BlackShark, for example, displays a window on his computer that collects information from the poker site and then calculates odds before making a bet.

Like most of his peers, Gabriel, who is working five nights a week to get BlackShark ready for the Las Vegas tournament, is an engineer first and a poker player second. He said his poor game skills are his biggest handicap.

“The hard part is: What if I’ve got two 10s? What am I going to do?” As he scans poker books for strategy tips, Gabriel is laboring to add an enormous set of rules telling the machine what to do with different cards and how to react to the frequency with which other players fold, call and raise.

Other robot designers, such as Ken Mages of Evanston, Ill., are further along. But though their electronic progeny may win at small-stakes tables, they usually fall apart when the human competition is stiffer.

After two weeks of programming, Mages said, “I could sit down at a 50-cent table, put 50 bucks in the account, go to bed and wake up with at least $75.” The most Mages said he won that way was $250; he never lost.

For two weeks this May, Mages sold his software for $60 a copy. After getting deluged with customer pleas for technical help -- and a threat by one who gambled away $10,000 to send him the bill -- Mages sold out to a business associate, Hong Kong engineer Ben Lo.

Mages then struck a deal with Los Angeles public relations executive Darren Shuster to set up the Las Vegas contest -- dubbed the World Series of Poker Robots -- and just after Memorial Day their partnership convinced Antigua-based GoldenPalace.com to put up the prize money.

Even though GoldenPalace bans robots, the publicity-craving virtual casino was a natural target, having spent $28,000 last fall for a cheese sandwich that was said to bear the image of the Virgin Mary. The sandwich is now on tour.

Organizers have further headline-grabbing gambits in mind: They plan to invite the winner of the human poker World Series to go up against the winner of their robot contest, though no one expects the computer code to triumph -- at least, not this year.

Entrants in the robotic-poker tournament so far include Gabriel, Lo, programmers from Florida, Canada and Spain, and Hilton Givens of Lafayette, Ind., who started working on a robot more seriously after he was laid off from his software job.

Most of the confirmed competitors have run their programs on PartyPoker.com, which forbids such activity and confiscates the accounts of those it catches.

The cat-and-mouse game between robots and online game sites is not limited to poker. Whenever any free multi-user computer game gets big enough, cheaters use programs and other means to boost their rankings, collect useful game tools or exact revenge on competitors.

Gabriel, for one, cobbled together an unbeatable Scrabble robot, which he inflicted on Yahoo Games opponents. But the problem is especially acute for sites like PartyPoker, which has a million real-money players registered and so presents a tempting target.

And site parent PartyGaming might soon have to worry about spooked investors as well as spooked players. Gibraltar-based PartyGaming, which reported $350 million in profit last year, is gearing up for a multibillion-dollar initial public offering in London, where Internet gambling is legal. That IPO will be the United Kingdom’s largest in at least four years, underscoring investor enthusiasm for the $8-billon online gaming market.

PartyPoker marketing director Vikrant Bhargava said he wasn’t pleased to learn that many of the poker bot World Series contestants honed their skills on his site, adding that eventually all such cheats get caught. Other sites don’t care whether users are human, he said, because the house takes the same percentage of the pot no matter who’s playing. But Bhargava said PartyPoker has 100 employees looking for robots, collusion among players and other scams.

Gaming companies won’t disclose all their secrets for sniffing out bots, but some of the techniques are simple. Any person playing three tables simultaneously for 48 hours without a bathroom break, for example, or invariably taking exactly one second to bet, is not a person.

Computer gaming experts said the robots have some major hurdles to overcome before they have a chance against the world’s top human beings -- especially in multi-player games with no betting limit, where the psychology is most important and the number of possible bets is much larger.

Bluffing can be programmed: For every 100 basically worthless hands, for instance, a machine might be instructed to bet heavily five times.

A far bigger issue is the need for abstract pattern recognition. Computers are much worse than humans at anything vague, said poker pro Magriel, a 58-year-old former math professor and world backgammon champion.

At such tasks, “computers are basically idiots,” Magriel said. “A computer has an enormous problem recognizing a face. A baby is better.”

The need to recognize patterns comes when anyone new sits down at the table. Good poker players learn from the behavior of their foes and adapt on the fly. Computers can store and process millions of past hands, but they have too little data on each new competitor.

For that reason, Schaeffer’s team has been focused for years on improving a program’s ability to compete one-on-one and learn from as few as 50 hands. After that, the current version does well for a while, until a strong human opponent figures out its patterns. Then the person starts winning.

Magriel once predicted computers would never master backgammon. Now that he knows different, he thinks a better-than- human poker program is inevitable in two or three decades.

“It was a little depressing in chess and backgammon that computers got so good,” he said. “In poker, it won’t really depress me. I sort of expect it at some point.”