Annie Duke on Poker, Probabilities, and How We Make Decisions

Guest:
Annie Duke — Former professional poker player and decision strategist; author of Thinking in Bets and How to Decide
Host:
Tyler Cowen
Source:
Conversations with Tyler · 1 July 2020

Annie Duke on Poker, Probabilities, and How We Make Decisions

Annie Duke — former professional poker player turned decision strategist, who left a PhD programme in cognitive science for the table — joins Tyler Cowen in Ep. 99 to explain why a good decision can produce a bad outcome, when probabilistic thinking helps and when it hurts, why we refuse to hedge our marriages but happily insure our houses, and what separates the handful of poker players who stay happy and keep improving from everyone else.

Key ideas

  1. A goal and an expected value can live in the same head. People treat ambitious goals and cold probability as a false dichotomy. Duke’s correction: set a high goal and run a running tally of expected value — the probability-weighted average of the payoffs, what a bet is worth on average once you price in both the odds and the stakes — at the same time. The expected value may sit well below the goal; aspiration and a sober estimate of the odds need not crowd each other out.
  2. We won’t hedge against our identity. Insuring a house against fire feels clean because a fire is bad luck that happens to us — we can offload it to chance. A prenuptial agreement feels like predicting your own marriage will fail, and worse, in a magical-thinking way (the unfounded sense that imagining a bad outcome helps cause it) like willing the failure into being. Only about 5 per cent of couples sign one. The good decision-maker thinks about the negative routes a life can take and gets out in front of them regardless.
  3. Process is the route to happiness under uncertainty. Poker forces loss on you in the most personal way — you hand the money across the table, face to face, in amounts that matter — so loss aversion is on constant display. The small group of players who stay happy fix their attention on solving the problem of the game, on building a good model of the opponent and choosing the right line, and treat the chip count as a scorecard for the process, not the prize.
  4. Skill in one domain need not transfer. A player who crushes the poker table will walk into the same casino and dump money at craps — negative-expectancy by design. The lesson Duke draws is that competence is often sharply domain-specific; do not assume a brilliant performer in one arena makes good decisions elsewhere.
  5. Open-mindedness, not intelligence, is the scarce ingredient. Thinking probabilistically is rarer than it looks — it is not taught in K–12, and most people read a 35 per cent forecast that comes true as having been ‘wrong’. Beyond that, the top player must be hungry to collide with corrective information, hold beliefs loosely, update constantly, and ponder daily where they might be wrong — against a brain built to insist ‘I’m right, I’m right, I’m right’.

Content

Resulting, and when probability stops helping

Cowen opens by asking when thinking probabilistically makes decisions worse. Duke struggles to find many such cases — the closest is the sensitivity-versus-specificity trade-off, the choice between catching every real threat (and tolerating false alarms) versus flagging only the certain ones (and missing some). A human on the savannah who hears rustling in the grass should not pause to estimate the probability of a lion; the magnitude of the downside — death — makes a categorical yes-or-no the survival move. The point generalises: probability is almost always the right frame, but what you do with the estimate depends on the asymmetry of the payoffs.

Underneath the whole conversation sits the idea Duke is best known for, resulting — judging the quality of a decision by how it happened to turn out rather than by whether it was sound given what you knew. A good decision can lose; a bad one can win. The savannah, the date, the marriage, and the poker hand are all worked examples of holding decision quality and outcome quality apart.

Goals versus expected value

Pressed on whether a dater should run a ‘mental probability ticker’ — ‘the chance I marry this person is 2.3 per cent’ — Duke says it depends on the goal. If the aim is a genuine match, thinking that way is a feature, not a bug: it filters toward people compatible with that cast of mind, raising the quality of the match even as it lowers the count. Her larger move is to dissolve the apparent conflict between having a high goal and thinking in expected value. The two can coexist; insisting on one or the other is the false dichotomy. Cowen offers a parallel from business — a hurdle rate, the minimum return an investment must clear to be worth funding, deliberately set high because gains get overstated and costs understated along the way — and Duke agrees that biasing your thresholds can be a sensible correction against fooling yourself.

What we will and won’t bet on

The bet you are willing to make is, for Duke, a question about the relationship as much as the wager. Among friends who think this way — her brother and brother-in-law literally made a market on whether she would marry her husband before the first date, bidding 23 and 22 — a bet is the fun of the friendship. With a tennis friend, she would never take the other side of a bet on whether their child gets into a given university, because a market that settles to one or zero reads as a prediction: take the low side and you have told them you expect failure. Different friends serve different purposes, and not everyone should be drawn into this kind of thinking.

The sharper puzzle is why we will insure a house but not hedge a marriage, though both are bets against a bad outcome. Duke’s answer turns on identity. A house fire is luck happening to us, so insuring against it stays out of the emotional space. A marriage is bound up with who we are and what we are committed to, and we treat identity as categorical, not probabilistic. On top of that sits magical thinking — the residue of The Secret and the positive-thinking literature, the notion that picturing failure helps summon it — which makes the prenup feel like both a prediction of failure and a cause of it. The good decision-maker, she insists, must think hard about the negative routes a life can take and get out in front of them, and a hedge is one honest way to do that.

Why poker players bet on everything

Cowen asks why gamblers bet on zero-sum games — shark migration patterns, golf side-bets — rather than the one big positive-sum game, equities. Duke (who says she has not played poker since 2012) explains the mindset: poker players want to state a belief, find someone to take the other side, run the price discovery until there is a number both will trade at, and test the belief in the only way that bites. They become somewhat detached from whether the game is zero-sum; they like the action and the test. The crossover with finance is heavy in both directions — poker players become options traders and vice versa — because the minds work the same way.

Reading opponents, and the limits of tells

Asked whether her psycholinguistics training helps her read opponents, Duke says barely at all. The real work is building a model of the opponent — how often they enter pots, the range of hands they would play from each position — and narrowing the possibilities from there. Physical tells are noisy: a bluffing player shows discomfort, but staring long enough to read the discomfort makes anyone uncomfortable, so the signal is confounded. Tells earn their keep only when a decision is already close. If the pot lays you three to one you need to win just 25 per cent of the time, an easy bar that needs no agonising; it is when your estimate sits right at that margin that a tell can tip you one way or the other. Against beginners the calculus flips — ‘strong means weak, weak means strong’: a player who slams chips down forcefully is usually bluffing, while one who slides them in gently is usually strong, and that tell alone can make your market.

The skill ladder, and what the very top have

What separates a local-casino regular in the top 10 per cent from a top-20 player? Duke reaches for a trading analogy: a trader who thrived when markets were wide (large spreads, plenty of room for error) would not survive today’s tight markets, where the gap between you and your competition has collapsed. A strong amateur can choose the secondary or even tertiary line of play — a good option that is not the best — and still crush a loose game. Against real experts, who pick the primary line far more often, that slack disappears and the small, repeated shortfalls compound. The difference is not one big thing but the steady accumulation of choosing the best option a higher fraction of the time.

Could most smart people do it?

Cowen raises Steve Levitt, who set out to become a top player and did. Duke doubts most clever people could follow. Intelligence alone is not enough; you need a particular kind of mind. First, genuine fluency in thinking probabilistically — rarer than Cowen credits, given it is absent from schooling and that people read a 35 per cent forecast that lands as a forecasting failure. (This gap is why Duke co-founded the Alliance for Decision Education.) Second, and harder, an appetite for being wrong: you must be hungry to collide with corrective information, hold beliefs loosely, update constantly, and keep the long run in view, because the worst outcome is clinging to a belief past its evidence and paying for it on the bottom line. Her single most important sign in a young player is open-mindedness — the willingness to ask daily ‘where am I wrong?’ against a brain built to chant ‘I’m right’. She closes with Phil Ivey, who, having just won a huge tournament as arguably the best player alive, spent the celebratory dinner picking apart the hands he had misplayed and the equity he had left on the table.

  • Annie Duke — speaker; former professional poker player and decision strategist
  • Tyler Cowen — host
  • Philip Tetlock — researcher on calibrated forecasting and belief-updating; kindred work on probabilistic thinking
  • Philip E. Tetlock on Forecasting and Foraging as a Fox — Conversations with Tyler episode on the same probabilistic-judgement terrain
  • Daniel Kahneman — the loss aversion and self-serving-bias literature Duke draws on at the table
  • Steven Pinker — her graduate-school interlocutor on the noun-versus-verb learning debate (semantic versus syntactic bootstrapping)
  • Rationality — concept; probabilistic belief-updating and the discipline of being wrong well
  • Knightian Uncertainty — concept; deciding under uncertainty that resists clean probabilities

See also