The beating, although inevitable, wasn’t unrelenting. I’d successfully slow-played my pocket kings, and apparently not broken into a giveaway grin when I’d hit trip threes on the flop.
If you’re not confident about the full meaning of that last sentence, well, neither am I. There’s a whole lexicon and grammar of poker that I have only a vague, childish grasp of. Even when I play my hand roughly in line with “proper poker”, I struggle to accurately describe the outcome. It’s like hitting a four in cricket, but not knowing the difference between “extra cover” and “deep midwicket”.
Experience – lots of experience – clearly counts. Before this challenge, I’d played three games over the last ten years, against similarly naïve opponents. Last-minute revision was urgently required. The morning before the showdown, I tested my memory for the ranking of poker hands, and was concerned to realise that two pairs did not in fact beat three-of-a-kind. Rather in desperation, I downloaded an app specifically for Texas hold ‘em, the “heads-up” version (i.e., two players rather than the typical seven or more). I spent the next few hours playing against my phone, with essentially a 50% success rate. This outcome seemed to reinforce the idea that poker is indeed a game of chance.
Not all encounters with virtual opponents are so haphazard, however. My revived poker interest was prompted by reports of Cepheus, a program that could apparently beat anyone at heads-up hold ‘em, given enough hands to iron out the vagaries of the deal. For me, this raised the whole question of expertise, and what it means to be really, really good at something.
I once played chess against a former schoolboy prodigy, who had himself played Gary Kasparov when reigning world champion. In contrast with what – I assume – was Kasparov’s attitude, I accepted in advance that I would lose and was primarily fascinated to see how long I could stay in the game. Ten moves in I was floundering, not sure whether to attack or defend. Three moves later, any tactical nuance was obviously superfluous: my pawns were falling over themselves in confusion and my king was surveying the shattered remnants of his defences.
My poker beating, though equally decisive, was slower and harder to pin down. For one thing, we just kept note of the score, and despite the losses on paper, the comforting stacks of chips in front of me never diminished. More significantly, I came out ahead in plenty of hands. In heads-up poker, it’s generally foolish to bother with an obviously weak opening hand. Your two of hearts and seven of clubs might very occasionally turn into a full house on the flop, but most half-decent players wouldn’t take those cards further if it cost them any chips. This means you keep winning small amounts easily, psychologically offsetting the larger losses and maintain an illusory feeling of achievement.
Despite that, I never deluded myself that I had a chance over the whole game. Pinning down the nature of my opponent’s edge was difficult, however. I quickly gave up trying to read his expression as he checked his cards – the cliché of inscrutability is doubtless mandatory in any seriously successful player. More significantly, it became clear in our post-match debrief that he knew the odds of any hand transpiring at any point in the deal, and the odds of an opponent having the cards to beat it.
Given those necessary skills, computers ought to be good at poker. The required statistical database is trivial to program, whilst facial expressions don’t come into it. Furthermore, the best programs, like the best players, rely on game theory to work out the optimal strategy, based not just on the odds of a winning hand at any stage, but according to how an opponent is likely to play.
This is where the distinctive fascination of poker lies. Every bet on every hand gives you information about your rivals, sometimes helpful, often misleading. A few times in our heads-up game, I paid to see my opponent’s hand, even though I correctly suspected his cards were better. One time I called him, for example, he turned over a pair of eights: seeing that, I could then calibrate what a similar opening bet might mean next time.
Of course, he knew that I knew, and the next time he raised to 500 he might have nothing or a pair of aces. The target is always moving. Good players absorb everything they can about their opponent’s strategy whilst minimising their own predictability, and staying emotionally robust to setbacks. Your ace-high straight might lose to a flush on the river card, but – with Kiplingesque detachment – you need to take the beating and treat the next hand with a similar level of cold calculation.
Of course, the computer will not go on tilt (poker slang for losing your cool). But nor will it recognise the signs that an opponent has overheated and started throwing chips around with nihilistic abandon. This encapsulates why, unlike for the strictly calculable game of chess, looking for optimal performance from an algorithm misses one of the key ingredients that makes poker so compelling.
The fascination with the game, based on my very brief and unprofitable career to date, is in mastering a combination of specialist knowledge, responsive strategy and the vagaries of the human ego – your opponent’s and your own. A truer mark of artificial intelligence in poker might arrive when a computer plays like a decent human player, knowing the odds, reading the game, and just once in a while, losing it completely.
Thanks to Tom Sambrook.