Advanced Decision Science: Game Theory, Bias-Proofing, and Sharp Judgment

You've already met the core machinery. You know that a good decision is not the same as a good outcome. You know how to value a choice with expected value, how to update a belief with evidence, and why your brain takes shortcuts that sometimes misfire. This final chapter pushes into the deep end — the place where decisions stop being a private calculation and become a contest of judgment against other people, against your own ego, and against the clock.

Three big questions sit at the heart of advanced decision science:

1. What do I do when the right move depends on what someone else does? That is game theory.
2. Knowing my brain is biased, how do I actually build a process that defends against it? That is bias-proofing.
3. How do I get measurably better at judgment over a lifetime, not just feel better? That is calibration and the personal decision operating system.

We'll take them in that order — and keep returning to the spine of the whole discipline: be right when you can (normative tools), understand why you're wrong (descriptive truth), and build habits that close the gap anyway (prescriptive practice).

30.1 When your best move depends on theirs: game theory

So far we've treated the world as something you bet against — a coin, a market, a disease. But many of life's biggest decisions are made against other thinking people who are also trying to win. Your competitor sets a price knowing you'll react. Your counterpart in a negotiation hides their bottom line. Two drivers approach a merge.

Game theory

The study of decisions where your best choice depends on what others choose — and theirs depends on yours. "Game" just means any situation of interdependent choice; it does not mean it's trivial or fun.

The dominant strategy — the easy case

Dominant strategy

A move that is best for you no matter what the other side does. When you have one, the other player's choice doesn't matter — just take it.

Dominant strategies are wonderful because they collapse a hard interdependent problem into a simple one. The trouble starts when the dominant move for each person leads everyone to a bad place. That's the most famous result in the field.

The Prisoner's Dilemma

Picture two rival coffee shops on the same street. Each must decide: keep prices normal ("cooperate") or slash prices to steal the other's customers ("defect").

              Shop B: Normal     Shop B: Cut price
            +------------------+------------------+
Shop A:     |  Both do well    |  A loses lots,   |
Normal      |  A: +5  B: +5    |  B wins customers|
            |                  |  A: -2  B: +8    |
            +------------------+------------------+
Shop A:     |  A wins customers|  Price war:      |
Cut price   |  A: +8  B: -2    |  both thin margin|
            |                  |  A: +1  B: +1    |
            +------------------+------------------+

Look at it from Shop A's seat. If B keeps prices normal, A earns +8 by cutting versus +5 by staying — cutting wins. If B cuts, A earns +1 by cutting versus −2 by staying — cutting still wins. So cutting prices is A's dominant strategy. By the same logic it's B's dominant strategy too. Both cut. Both land in the bottom-right box (+1, +1) — worse than if both had simply stayed normal (+5, +5).

Nash equilibrium

A state where no single player can do better by changing only their own move, given what everyone else is doing. It is stable — nobody wants to be the one to deviate.

(Defect, Defect) is the Nash equilibrium of this one-shot game. Notice the lesson that trips up almost everyone:

Why does this matter for you? Because once you can spot a Prisoner's Dilemma, you can sometimes change the game instead of just playing it badly. You can sign a contract that punishes defection, build a reputation that makes cooperation believable, or turn a one-time deal into a repeated relationship — which leads to the most hopeful result in the field.

Repeated games and Tit-for-Tat

The grim logic above assumes the game is played once. Most real relationships repeat — with suppliers, colleagues, customers, neighbors. When the game repeats indefinitely, cooperation can emerge and survive, because today's betrayal can be punished tomorrow.

In 1980 the political scientist Robert Axelrod ran a famous tournament: he invited experts to submit computer strategies to play the repeated Prisoner's Dilemma against each other, thousands of rounds. The winner was almost embarrassingly simple. It was called Tit-for-Tat.

Tit-for-Tat

Cooperate on the first move. After that, simply copy whatever the other player did last time. Be nice first; punish a betrayal once; forgive the moment they cooperate again.

Axelrod found that the strongest strategies shared four traits:

A practical game-theory checklist for negotiations

• Ask: is this one-shot or repeated? Your generosity should scale with how often you'll meet this person again.
• Find your dominant move first. If a choice is best regardless of their response, you've saved yourself a lot of mind-reading.
• Beware the anchor. The first number named in a negotiation drags the final price toward it. Decide your walk-away number before the talk so their opening offer can't reset your reference point. (We'll return to anchoring below — it's a bias, not just a tactic.)
• Change the payoffs, not just your play. Contracts, escrow, warranties, and public commitments all turn "defect" from tempting into costly. That's how you upgrade a bad equilibrium.

30.2 The descriptive truth, sharpened: prospect theory and the broken ideal

You met the normative gold standard earlier: Expected Utility Theory (EUT) — the idea that if your preferences are internally consistent (obeying the four von Neumann–Morgenstern axioms), you behave as if you maximize expected utility, and you can't be tricked into a money-losing loop of trades. It's the rulebook for an idealized rational agent.

Here's the advanced point: EUT is a beautiful description of how we should decide and a poor description of how we do decide. Real humans systematically and predictably violate it. Two famous puzzles proved this beyond doubt.

The Allais paradox

In carefully designed pairs of gambles, most people switch their preference in a way that no single consistent utility function allows. They reveal that a "certain" outcome has an extra pull that pure EUT can't explain (the certainty effect).

The Ellsberg paradox

People strongly prefer a bet with known odds over one with unknown odds, even when the math is identical. This is ambiguity aversion — we dislike not knowing the probabilities themselves, not just disliking the risk.

Prospect Theory: the model of how we really choose

Kahneman and Tversky built the leading descriptive model to fill the gap. It rests on three pillars.

1. Reference points, not absolute wealth. We don't feel rich or poor in absolute terms — we feel the change from where we started. A €60,000 salary feels like a triumph to someone who earned €40,000 last year and a punishment to someone who earned €90,000. Same number, opposite feeling.
2. Loss aversion. Losses hurt roughly twice as much as equivalent gains feel good. Research puts the "loss-aversion coefficient" λ somewhere around 1.5 to 2.5, with about 2.0 as the canonical figure.
3. Different risk attitudes for gains vs. losses. We are risk-averse when protecting gains (we'll take a sure win over a gamble) but risk-seeking when facing losses (we'll gamble to avoid a sure loss).

   Value (how it feels)
        ^
        |             . . . . .  gains: curve flattens
        |        . '             (each extra € matters less)
        |     . '
--------+--'----------------------> Outcome
     . '|        reference point (0) at the kink
   . '  |
 .'     |   losses fall STEEPER than gains rise
.'      |   (a loss of X hurts ~2x a gain of X)
        v

Probability weighting: why we buy lottery tickets AND insurance

Prospect theory adds one more twist that EUT can't handle. We don't treat probabilities at face value. We overweight tiny probabilities and underweight moderate-to-large ones.

30.3 The advanced bias kit (and why awareness isn't enough)

You've met the headline biases. At the advanced level, three points matter more than memorizing a longer list.

First: knowing a bias does not switch it off. Biases work like optical illusions. You can be told two lines are exactly the same length, measure them yourself, and they will still look different. Anchoring, loss aversion, and overconfidence behave the same way — awareness barely dents them. This is why the rest of this chapter is about process and environment, not willpower.

Second: heuristics are not the enemy. Gerd Gigerenzer's "fast and frugal" research is the important counter-narrative. Mental shortcuts are often ecologically rational — superbly tuned to the real environments where they evolved. A simple rule can beat a complex model when information is scarce and time is short. So the goal isn't to delete intuition; it's to know which situations reward a snap judgment and which demand slow, deliberate analysis.

Third: the most expensive biases are the quiet, self-flattering ones. Here are the ones that wreck big decisions:

Confirmation bias

Hunting for evidence that supports what you already believe and waving away the rest. The cure is to ask, on purpose, "what would change my mind?" and go find that.

Sunk cost fallacy

Throwing more money, time, or effort after a failing project because of what you've already spent. The spent resources are gone either way — they should never influence the next decision.

Overconfidence / overprecision

Being more certain than your accuracy warrants — especially giving a single point estimate when you should give a range.

Optimism / planning fallacy

Systematically underestimating how long something will take, how much it will cost, and what could go wrong.

Hindsight bias

"I knew it all along." After an event, we rewrite our memory to feel we'd predicted it — which destroys our ability to learn what we actually got wrong.

Resulting (outcome bias)

Judging the quality of a decision by the quality of its outcome. Annie Duke's term, and the most important reframe in the entire discipline.

30.4 Bias-proofing: building a process that's smarter than you are

Since awareness fails, we engineer the decision environment instead. These are the prescriptive tools — the "what to actually do Monday morning" layer. Each one targets a specific bias.

1. The premortem — beating optimism before you commit

Premortem (Gary Klein's technique)

Before you act, imagine it's a year from now and the decision has already failed spectacularly. Then ask everyone: "Why did it fail?" You generate the autopsy before the death, while you can still prevent it.

2. Checklists — beating memory under pressure

Checklist (Atul Gawande, The Checklist Manifesto)

A short, pre-written list of must-do or must-check items for a recurring high-stakes decision. Not because you're forgetful — because everyone's memory fails under stress, fatigue, and time pressure.

Pilots run a checklist before every takeoff. Surgeons run one before every incision. These are among the most skilled professionals alive, and they still use a piece of paper, because the cost of forgetting one obvious step is catastrophic. Your version: a hiring checklist, a "before we sign" checklist, a launch checklist.

3. Reversibility — matching speed to stakes

Not every decision deserves the same effort. Jeff Bezos's framing: sort decisions into one-way doors and two-way doors.

4. Group decision tools — beating groupthink

Teams have a special failure mode: everyone anchors on the first loud voice (often the boss), doubts go unspoken, and the group convinces itself of a consensus nobody truly holds. Three fixes:

• Anonymous estimates first. Have everyone write their number or vote privately, before discussion. This stops the room from echoing the first speaker. (A version of the Delphi method.)
• Assign a devil's advocate. Make it someone's explicit job to argue the opposing case, so dissent is a duty rather than an act of courage.
• "Disagree and commit." Air the disagreement fully, then unite behind the decision so the team doesn't relitigate forever.

30.5 Sharp judgment over time: calibration and forecasting

Everything so far improves a single decision. The crown jewel of advanced decision science is getting measurably better over a lifetime. That requires turning your vague confidence into numbers you can score.

Calibration

The match between your stated confidence and your real hit rate. You are well-calibrated if, across all the times you said "70% sure," you turned out right about 70% of the time.

The Brier score — keeping honest score

Brier score

A single number that measures how good your probability forecasts are. Lower is better: 0 is perfect prediction, 0.5 is no better than random guessing on a coin-flip question. It rewards being both accurate and appropriately confident.

Philip Tetlock's Good Judgment Project ran the landmark experiment: it had thousands of ordinary, trained volunteers forecast real geopolitical events and scored them. The headline finding is the most encouraging result in the field:

• Calibration is learnable. The best volunteers — Tetlock's "superforecasters" — achieved Brier scores around 0.20 to 0.25 and routinely out-forecast professional intelligence analysts who had access to classified material.
• They weren't geniuses with secret data. They had better habits.

What did the superforecasters actually do?

1. Start from the base rate (the outside view), then adjust. Before predicting "will this startup succeed?", they note that ~90% of startups fail — and only then move toward the specifics of this case. Base rate first, evidence second, in that order.
2. Update incrementally and often — but not wildly. Many small Bayesian updates as new information arrives beat both stubbornness (never updating) and over-reaction (flip-flopping on every headline).
3. Think in probabilities and fine degrees — "23%," not "probably not." Precision forces clear thinking and makes scoring possible.
4. Break big questions into smaller ones they can actually estimate, then recombine.

The decision journal — the highest-leverage habit

You cannot improve a process you can't see. The decision journal makes your decisions visible and scoreable, and it defeats the two biases that otherwise make learning impossible: hindsight bias ("I knew it all along") and resulting ("it worked out, so it was a great call").

For every meaningful decision, write down — at the time, before the outcome is known:

1. What you decided.
2. Why — your reasoning and the key assumptions.
3. Your probability estimate ("I think this has a 65% chance of working").
4. What you expect to happen, and by when.
5. How you felt (tired? rushed? emotional? — these predict bad decisions).

Then, later, review. Now you can ask the only fair question:

30.6 Putting it together: your personal decision operating system

Advanced decision science isn't a collection of party tricks. It's an integrated way of working that you can run on every important choice. Here is the whole arc of these three chapters compressed into one repeatable loop.

  +-----------------------------------------------------+
  |  1. FRAME    Is this reversible? (one/two-way door) |
  |              Match effort to stakes.                |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  2. OUTSIDE VIEW   Start from the base rate.        |
  |                    What usually happens to choices  |
  |                    like this one?                   |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  3. THINK IN BETS  Give a probability + a range,    |
  |                    not a certainty. Estimate EV /   |
  |                    expected utility, not just $$.   |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  4. WHO ELSE?      Is anyone else deciding? Map the |
  |                    game. One-shot or repeated?      |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  5. BIAS-PROOF     Premortem. Checklist. Anonymous  |
  |                    estimates. Reframe gain<->loss.  |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  6. JOURNAL        Record decision, reasoning, odds |
  |                    BEFORE the outcome.              |
  +-----------------------------------------------------+
                          |
                          v
  +-----------------------------------------------------+
  |  7. REVIEW & SCORE Judge the decision, not the      |
  |                    outcome. Update your calibration.|
  +-----------------------------------------------------+
                          |
                          +------> (feeds back to step 2:
                                    better base rates next time)

Notice how the loop unites the three lenses. Steps 2–3 are normative (base rates, EV, expected utility — how to be right). Step 5 acknowledges the descriptive truth (we're biased, so we engineer against it). Steps 1, 4, 6, and 7 are prescriptive (reversibility, game-mapping, journaling, scoring — how to do better anyway). And the whole thing feeds back on itself: every reviewed decision sharpens the base rates you bring to the next one.

30.7 The through-lines, one last time

If you forget every formula in these chapters, keep these four sentences. They are the entire discipline distilled.

1. A decision is not its outcome.

You make decisions with foresight under uncertainty; outcomes are judged in hindsight. Judge the process, because the process is the only thing you control. A good 70% bet that loses was still a good bet.

2. Think in probabilities and ranges, not certainties.

"What odds would I give?" beats "will it work?" every time. Confidence is cheap; calibration is the real skill, and it's trainable.

3. Base rate first, then update.

Start from how things usually go (the outside view), then move toward your specific evidence in small, honest steps — without forgetting where you started.

4. The process is the product.

You cannot control luck, other players, or the future. You can control your framing, your base rates, your bias-proofing, and your journal. Build the operating system, run it on every important choice, and let good outcomes accumulate as the long-run reward of good decisions.

That is where the journey ends and your practice begins. You now know how an ideal agent should choose, why your own mind reliably deviates, how to defend against those deviations, and how to keep honest score so you get sharper every year. Pick your next real decision — a job, a purchase, a launch, a negotiation — and run it through the loop. The first time it will feel slow and mechanical. By the tenth time it will feel like clear sight.