Computers can fly aircraft, build cars, fire missiles and even calculate your taxes. But for those who fear that we may one day be ruled by machines, reassurance has arrived. It turns out that there is one thing computers still cannot do better than us: bluff.

Proof comes from an historic poker tournament in Vancouver, Canada, where the world’s most sophisticated poker computer was pitted against two of the world’s most talented fibbers: Phil “Unabomber” Laak and Ali “Prince Ali” Eslami, veterans of America’s hugely lucrative professional poker circuit.

The computer, a “pokerbot” named Polaris, is the creation of a team of artificial-intelligence experts from the University of Alberta – and was expected to bring about the kind of humbling victory that Deep Blue delivered against the Russian chess genius Garry Kasparov exactly ten years ago.

But the human players’ bluffing and risk-taking proved too much for the machine’s logical circuitry. It ended the game having lost out on $10,000 (£5,000) and out of chips.

The victory, however, was as slim as silicon: the humans prevailed in the fourth and final game, and by a mere $570. “I’m surprised we won,” an exhausted Prince Ali admitted as he left the table close to midnight on Tuesday after 48 hours of play.

The human adversaries of Polaris were selected for their professional success and their knowledge of computers. The Dublin-born Unabomber, 34, nicknamed after the brand of his trademark hooded sweatshirt, has banked $1.2 million in recent live tournament winnings; Prince Ali, 30, from the Los Angeles suburb of Van Nuys, typically wins $10,000 to $40,000 a tournament. But the computer was also daunting for the professionals: their reflective sunglasses and baggy clothing, designed to disguise their so-called “tells” – the physical ticks used to bluff and double-bluff – were useless against a foe that played using only statistics, and never got tired, hungry or needed the lavatory.

By professional standards the cash prize for beating Polaris was modest: $5,000 for each of the four matches of Texas Hold ’em, or $50,000 for a quadruple win, with 500 hands dealt per match. The format of the game was tweaked to make the play easier to analyse and less vulnerable to a fluke.

To remove the possibility of one side benefiting from the luck of the draw, the exact hands were swapped between humans and computer after each round. The humans were put in separate rooms to prevent collaboration. In the first round the humans were down by $70, a technical draw.

The second round was a win for Polaris. But the humans staged a come-back on the second day as the computer switched from a mode that used a single gambling “personality” to a mode that used multiple personalities – from passive to daring – and drafted them in and out like a team coach. The humans ended ahead by $820.

The final round also went to the humans, but not without drama: Prince Ali was down by almost $1,000 in the first part of the game, for which he blamed Polaris’s ability to learn that when he played certain hands he was typically bluffing about having better cards.

At the last minute Polaris pulled out a royal flush that beat Prince Ali’s three-of-a-kind, but it was too late: Prince Ali finished $460 ahead, with the Unabomber up by $110. The humans won $10,000 for their wins and $2,500 for the draw, with the total split between them.

Man v machine

— Deep Blue, IBM’s chess computer, beat Garry Kasparov in 1997. Kasparov said his opponent played “like a God”

— A virtual wine buff built to analyse the tipple came close to beating an international panel of connoisseurs at predicting the price, region and quality of blindly tasted wines last year

— An unbeatable draughts-playing computer reported by The Times last week compares all 500 billion billion positions possible in the game at every move it makes

Sources: George Mason University; Edinburgh University Chess Club; chemie.de; Times archives

Explore Tech & Web

• Personal Tech

• The Web

• Gadgets & Gaming