ProjekinAja. Clear Thinking Stage 2 · Applied Judgment Lesson 2.2
● Stage 2 · Lesson 2.2

Game Theory: Thinking in Interactions

Most decisions are not made in isolation — they depend on what others do. Game theory is the vocabulary for thinking clearly about strategic interaction.

⏱ ~20 min read 🧩 Core Concept 🕹 Prisoner's dilemma game

🎯 By the end you'll understand

The three components of any game (players, strategies, payoffs); how to spot dominant strategies and Nash equilibria in 2×2 tables; the difference between zero-sum, positive-sum, and negative-sum games; why the prisoner's dilemma captures so much real social dysfunction; how repeated interaction, tit-for-tat, bargaining, and information asymmetry change the picture; and exactly which premise in the running argument fails under game-theoretic scrutiny.

FOUNDATIONSPlayers, strategies, and payoffs

Every game theory model — no matter how complex — is defined by exactly three things:

ComponentDefinitionConcrete example
Players The decision-makers in the interaction Two firms competing for the same customers; two countries in a trade negotiation; two suspects being interrogated separately
Strategies The complete menu of actions available to each player — a full plan specifying what to do in every possible situation Price high / price low; cooperate / defect; invest in R&D / cut costs; accept offer / reject and counter
Payoffs The outcome each player receives for each combination of strategies chosen — expressed in whatever unit captures their interests Profit in dollars; years in prison; points in a computer tournament; probability of survival

Once you know who is playing, what they can choose, and what outcomes follow from each combination of choices, you can reason systematically about what rational players will do — and why the results are sometimes collectively terrible even when every individual is behaving rationally.

💡 The robotics parallel

In multi-agent systems: players are agents; strategies are policies (mappings from observations to actions); payoffs are reward functions. Game theory is essentially the study of how rational agents with reward functions behave in shared environments. The same mathematics underlies both fields.

CORE TOOLDominant strategies and reading a payoff matrix

A payoff matrix is a compact table showing the outcome for every combination of strategies. For a two-player, two-strategy game it looks like this:

Player B:
Strategy X
Player B:
Strategy Y
Player A: Strategy X (A: 3, B: 3) (A: 1, B: 4)
Player A: Strategy Y (A: 4, B: 1) (A: 2, B: 2)

Each cell shows (A's payoff, B's payoff). Player A chooses rows; Player B chooses columns. Both choose simultaneously without knowing the other's choice.

To find the rational choice for each player, look for a dominant strategy — a strategy that gives the highest payoff regardless of what the other player does.

🔑 How to spot a dominant strategy

For Player A: look at each row and ask "is strategy Y always better than strategy X, no matter which column B picks?" In the table above: if B picks X, A gets 4 (strategy Y) vs 3 (strategy X) — Y is better. If B picks Y, A gets 2 vs 1 — Y is still better. So Strategy Y is dominant for A. Run the same check for B. If both players have dominant strategies, the game's outcome is predictable: both will play their dominant strategy.

KEY CONCEPTNash equilibrium

A Nash equilibrium is any outcome where neither player has an incentive to unilaterally change their strategy, given what the other is doing. It's a stable resting point — but not necessarily a good one.

John Nash (Princeton, 1950) proved that every finite game has at least one Nash equilibrium (possibly in mixed strategies). The concept gives us a way to predict outcomes without requiring one player to know the other's move in advance — they just need to know that the other is rational.

⚠ Nash equilibrium ≠ optimal outcome

This is the most important thing to understand about Nash equilibria: they describe a stable state, not a good state. Two countries locked in an arms race are at a Nash equilibrium — neither benefits by unilaterally disarming — but both are poorer and less secure than if neither had armed. Stability and optimality are completely different concepts.

When a game has a dominant strategy for both players, the cell where both play their dominant strategy is a Nash equilibrium. When no dominant strategy exists, finding Nash equilibria requires checking each cell: can either player improve by switching, holding the other's strategy fixed? If neither can, it's a Nash equilibrium.

A game can have multiple Nash equilibria (common in coordination games), or none in pure strategies (requiring mixed — probabilistic — strategies). For this course, 2×2 matrices with pure-strategy equilibria are sufficient to develop the core intuition.

▸ Source note

The treatment here follows Ben Polak's Game Theory (ECON 159, Yale Open Courses) — specifically his lectures on dominance, Nash equilibrium, and commitment. The MIT OpenCourseWare game theory materials (14.12) cover advanced applications including bargaining and mechanism design.

TAXONOMYZero-sum, positive-sum, and negative-sum games

One of the most practically useful things game theory gives you is the ability to classify interactions by their structure. The classification determines what strategies make sense and what outcomes are possible.

🧩 Interactive · Three game types

Definition: One player's gain is exactly another player's loss. Total payoffs across all players sum to zero (or a constant). The pie is fixed; the only question is how it's divided.

Illustration: A gains 10, B loses 10. Whatever A wins, B loses — not because someone is unfair, but because the structure of the interaction makes it impossible for both to gain simultaneously.

Real examples: Poker (one player's winnings come from other players' losses); currency trading in a fixed-rate world (a gain for one currency trader is a loss for another); most competitive sports (one team wins, the other loses).

⚠ Zero-sum bias: misclassifying positive-sum situations as zero-sum

The most common game-theoretic reasoning error is applying zero-sum logic to situations that are actually positive-sum. Most economic exchange, most employment relationships, most negotiated deals — these are not zero-sum. When you frame them as zero-sum, you leave value on the table and damage relationships unnecessarily. We come back to this in the running example below.

Definition: Trade or cooperation creates new value — both parties can end up better off than they started. Total payoffs sum to more than zero. The pie can grow.

Illustration: A values an item at $20. B values the same item at $5. A buys it from B for $10. A gets $20 worth of value for $10 — a gain of $10. B gets $10 for something worth $5 to them — a gain of $5. Total gain: $15. Neither "won" at the other's expense; the exchange itself generated value that didn't exist before.

This is only possible because people value things differently. If A and B valued the item identically, no exchange would create value — they would simply be trading identical utilities. Difference in valuation is the engine of positive-sum exchange.

Real examples: Voluntary market exchange; most employment relationships (employer values the worker's output more than the wage cost; worker values the wage more than the leisure forgone); international trade where countries differ in relative productivity (comparative advantage — more on this in Lesson 2.3).

💡 Why investment is not inherently zero-sum

When a firm raises capital to build a factory, workers get jobs, customers get products, and investors get returns — all from value the factory creates that didn't exist before. This is positive-sum. We examine this directly in Lesson 2.4.

Definition: The interaction destroys value — both parties end up worse off in aggregate than if the interaction hadn't happened at all. Total payoffs sum to less than zero.

Illustration: Two firms spend $8 million each in legal fees fighting over a contract worth $5 million. The winner gets $5M but spent $8M in legal costs — net loss of $3M. The loser spent $8M for nothing. Total destruction: $11M spent to distribute a $5M prize. Both would have been better off settling.

Real examples: Arms races (both sides spend heavily on weapons neither wants to use, ending up less secure and poorer); prolonged litigation; destructive price wars that crater an entire industry's margins; countries engaged in tit-for-tat tariff escalation.

⚠ Negative-sum traps are often Nash equilibria

The especially treacherous feature of negative-sum situations is that they are often stable. In an arms race, neither party has an incentive to unilaterally disarm — so both continue spending, even though both would prefer mutual disarmament. Being locked in a Nash equilibrium that destroys value is one of the most important failure modes game theory reveals.

KEY CONCEPTThe prisoner's dilemma

The prisoner's dilemma is the most famous game in all of game theory — because it captures a real structural tension between individual rationality and collective welfare. Two suspects are held separately. They cannot communicate. The payoffs in years of prison are:

B: Stays silent B: Betrays
A: Stays silent A: 1 yr, B: 1 yr
(mutual cooperation)
A: 10 yr, B: 0
(A cooperated, B defected)
A: Betrays A: 0, B: 10 yr
(A defected, B cooperated)
A: 5 yr, B: 5 yr
(mutual defection)

Put yourself in A's position. If B stays silent: betray gives 0 years, silence gives 1 — betray is better. If B betrays: betray gives 5 years, silence gives 10 — betray is still better. Betrayal is dominant, regardless of what B does. B faces identical logic. So both betray and both get 5 years — even though mutual silence would give both only 1 year.

🔑 The structural insight

Individual rationality produces collective irrationality. Neither player is stupid or malicious — the payoff structure creates the trap. Understanding this structure explains arms races, price wars, overfishing, tax avoidance, and many forms of social dysfunction. The structure, not the character of the players, drives the outcome.

Real-world prisoner's dilemmas

CONTRASTCooperation games vs coordination games

It is worth distinguishing two types of game that are often confused in everyday language.

In a cooperation problem (like the prisoner's dilemma), interests are partially opposed — each player is tempted to defect. The challenge is to align incentives so cooperation becomes individually rational.

In a coordination game, interests are aligned — both players want to make the same choice. The challenge is not defection but synchronisation: knowing which of the mutually-beneficial options the other will pick.

B: Standard A B: Standard B
A: Standard A Both: 1 (compatible) Both: 0 (incompatible)
A: Standard B Both: 0 (incompatible) Both: 1 (compatible)

There are two Nash equilibria (both pick A, or both pick B). Neither is better than the other — what matters is that everyone converges on the same one. This is why focal points (Schelling points) matter: the option that feels natural, salient, or prominent to both players serves as the coordination device.

▸ Coordination games everywhere

Driving on the left or right; which side of a corridor to pass someone; which programming language a team adopts; which messaging platform friends use — all coordination games. Once everyone is on one equilibrium, no individual benefits from unilaterally switching. That's why switching costs are real and standards are sticky.

Practical implication: If you misidentify a coordination problem as a conflict, you destroy value. You try to win when you should be trying to synchronise. Many negotiations that feel adversarial are actually coordination games in disguise — both parties want a deal, the challenge is agreeing on the specific terms.

KEY EXTENSIONRepeated games and the shadow of the future

The one-shot prisoner's dilemma has a clean dominant strategy: defect. But most real interactions are not one-shot. You encounter the same suppliers, colleagues, neighbours, and trading partners again and again.

When the game is repeated, defecting today imposes future costs. Your partner can observe your past choices and adjust accordingly. Robert Axelrod called this the "shadow of the future" — the degree to which the value of future cooperation influences present choices. When that shadow is long enough, cooperation becomes individually rational without any external enforcement.

Axelrod demonstrated this empirically in a famous computer tournament (The Evolution of Cooperation, 1984). He invited game theorists to submit strategies for playing repeated prisoner's dilemmas — every strategy played against every other strategy and against itself over many rounds. The winner was the simplest strategy submitted: tit-for-tat, by Anatol Rapoport.

▸ Tit-for-tat rules
  1. Cooperate on round 1.
  2. In every subsequent round, do whatever your opponent did last round.
  3. That is the entire strategy.

Tit-for-tat won for four reasons Axelrod identified: it is nice (never defects first); retaliatory (immediately punishes defection, so defectors cannot exploit it indefinitely); forgiving (returns to cooperation after one punishment, so cycles of mutual defection don't persist); and clear (simple enough for opponents to understand, which allows cooperative relationships to form).

💡 The practical takeaway

In repeated interactions, your reputation is your strategy. Being known as someone who cooperates in good faith, responds firmly to exploitation, and forgives quickly is not merely ethical — it is the individually optimal strategy in any world where you will encounter the same players again.

APPLICATIONBargaining

Bargaining is what happens when a positive-sum exchange is available but the distribution of the gains is not determined by market price alone — both parties must negotiate how to split the surplus.

The key insight from bargaining theory (Nash, 1950; Rubinstein, 1982) is that bargaining power comes from your outside option — what you will do if the negotiation fails. The party with the better outside option (alternative deal, ability to walk away, lower cost of delay) will capture more of the surplus.

FactorGives you more power when…
Outside option (BATNA)You have a better alternative if talks fail
Cost of delayDelay is less costly to you than to the other party
CommitmentYou can credibly commit to a position ("I won't go below X")
InformationYou know the other party's reservation value; they don't know yours

One game-theoretically interesting finding: sometimes limiting your own options increases your bargaining power. If you can credibly commit to walking away at a certain price (e.g., by making your offer public, or by having a visible constraint), the other party knows negotiating below that point is futile — and they may concede rather than lose the deal entirely. This is the logic behind "take it or leave it" offers and constitutional entrenchment.

⚠ Zero-sum framing in bargaining destroys value

Many negotiations treat every gain for one side as a loss for the other. But most negotiations have multiple dimensions (price, timeline, payment terms, scope, relationship). Often both parties value different dimensions highly — trading on those dimensions creates value that a pure price negotiation would miss. Game theory calls this "expanding the pie before splitting it."

ADVANCED CONCEPTInformation asymmetry

Most real games are played with imperfect information. One player often knows something the other does not. This creates a separate class of strategic problems.

George Akerlof's famous 1970 paper "The Market for Lemons" showed that when sellers of used cars know quality but buyers don't, the entire market can unravel: buyers assume average quality, offer average prices, sellers of high-quality cars exit (the price is too low), average quality falls, buyers revise down further, and so on until only low-quality ("lemon") cars remain.

Two key concepts from information game theory:

🔑 Why information asymmetry matters for reasoning

When someone has private information that affects your decisions, naive analysis of their stated positions is not enough. You need to ask: What would they say if the information were bad for me? What actions would they take? Good decision-making in asymmetric-information environments requires modeling the other party's incentives, not just listening to their claims.

INTERACTIVEPlay the prisoner's dilemma

You will play 8 rounds against a tit-for-tat bot. You are a supplier; the bot is a competing supplier. You can cooperate (hold your price) or defect (undercut). Watch how the bot responds and observe your cumulative score.

Payoffs: both cooperate = 3 pts each; both defect = 1 pt each; you defect / bot cooperates = you get 5, bot gets 0; you cooperate / bot defects = you get 0, bot gets 5.

🕹 Interactive · Prisoner's dilemma vs. tit-for-tat
Your score
0
Bot score
0
Round
1 / 8

RUNNING EXAMPLEBack to the argument: where exactly does P2 break down?

The running argument across this course is:

P1 "It is wrong to profit when another person loses."
P2 "Whenever someone profits from investment, another person loses."
C "Therefore, investing is wrong."

Stage 1 gave us the tools to identify the logical structure: P1 and P2 are premises; C is a conclusion; the argument is valid (C follows from P1+P2). So to defeat it, we need to show P1 or P2 is false or dubious. Game theory gives us the vocabulary to examine P2 precisely.

P2 presupposes that investment is a zero-sum game: every dollar of gain for an investor comes directly from a dollar of loss for someone else. This is the zero-sum assumption in its plainest form.

But the taxonomy above gives us the tools to ask: is investment actually zero-sum? The answer depends on which type of investment we are talking about:

Investment typeGame structureWhy
Poker, currency speculation in a fixed-sum market Zero-sum One party's gain is another's exact loss. P2 holds here.
Short-selling, futures markets Zero-sum (among traders) For every long position profit there is a short-position loss. P2 holds here too.
Equity investment in a productive firm Positive-sum The firm uses capital to create new value (products, services, employment). Returns come from value created, not from another investor's loss.
Prolonged hostile takeovers, patent trolling Negative-sum Legal and strategic costs exceed the value redistributed. Both sides can end up worse off than a negotiated deal would have achieved.

P2 treats all investment as if it were poker — a fixed total being redistributed. This is a false generalisation. The premise fails because it ignores the positive-sum nature of productive investment. The quantifier "whenever" (the universal quantifier) is doing too much work: it turns a sometimes-true claim into an always-true claim, and that extension is precisely what game theory disproves.

🔑 Lesson 1.5 + Lesson 2.2 combined

Lesson 1.5 showed us that "whenever" is a universal quantifier — and universal claims are defeated by a single counterexample. Game theory gives us the conceptual framework to construct that counterexample: productive equity investment is a demonstrably positive-sum game, which is exactly the counterexample that falsifies P2.

WEAKNESSCommon mistakes to avoid

✗ "Game theory says always defect."

Game theory says defect in a one-shot game with a stranger where defection has no future consequences. In repeated games with identifiable players who observe past behaviour, cooperation is often the individually optimal strategy. The recommendation depends entirely on the context — one-shot vs. repeated is one of the most important distinctions in applied game theory.

✗ Treating Nash equilibrium as the optimal outcome.

A Nash equilibrium is a stable state — no player wants to deviate unilaterally. It says nothing about whether the outcome is good for anyone. Mutual defection in the prisoner's dilemma is a Nash equilibrium and it is terrible. Arms races are Nash equilibria. Equilibrium ≠ optimum.

✗ Misidentifying coordination problems as conflicts.

Not every multi-player situation is adversarial. Coordination problems have aligned interests but unclear expectations. Treating them as conflicts causes both parties to choose mismatched strategies — and both lose. Asking "do we both want the same outcome here?" before assuming adversarial structure can prevent enormous amounts of unnecessary friction.

✗ Assuming all investment is zero-sum (the P2 error).

Zero-sum logic applies cleanly to redistributive games like poker or currency speculation. It does not apply to productive investment, mutual employment, or voluntary exchange where value is being created, not merely redistributed. The cost of this error is refusing to participate in genuine positive-sum opportunities out of a misapplied intuition about fairness.

GLOSSARYKey terms introduced

Game theory
The mathematical study of strategic interaction — situations where outcomes depend on the choices of multiple decision-makers.
Players
The decision-makers in a game.
Strategy
A complete specification of what a player will do in every possible situation.
Payoff
The outcome (utility, money, years in prison) a player receives for each combination of strategies.
Payoff matrix
A table showing outcomes for every combination of strategies in a two-player game.
Dominant strategy
A strategy that gives the best payoff regardless of what other players do.
Nash equilibrium
A stable outcome where no player can improve their payoff by unilaterally changing their strategy. Stable ≠ optimal.
Zero-sum game
An interaction where total payoffs are fixed — one player's gain is another's loss.
Positive-sum game
An interaction where total payoffs can grow — both parties can gain through exchange or cooperation.
Negative-sum game
An interaction that destroys value — both parties end up worse off in aggregate.
Prisoner's dilemma
A game structure where individual rationality leads to collective irrationality — both players defect even though mutual cooperation would benefit both.
Coordination game
A game where interests are aligned but the challenge is synchronising on the same equilibrium.
Repeated game
A game played multiple times by the same players, enabling reputation and conditional strategies.
Tit-for-tat
A strategy for repeated games: cooperate first, then mirror your opponent's last move. Winner of Axelrod's (1984) tournament.
Shadow of the future
The degree to which anticipated future interactions influence present behaviour.
Focal point (Schelling point)
A salient option that players independently converge on in coordination games without communication.
BATNA
Best Alternative to a Negotiated Agreement — your outside option, which determines your bargaining power.
Signalling
A costly observable action taken to credibly communicate private information.
Information asymmetry
A situation where one player knows something relevant that another does not.

ACTIVE RECALLCheck your understanding

Answer in your own words first, then reveal.

1What is a Nash equilibrium? Give an example of one that is clearly bad for both players.›

A Nash equilibrium is any outcome where neither player has an incentive to unilaterally change their strategy, given what the other is doing. It is a stable resting point — not necessarily a good one. Example: mutual defection in the prisoner's dilemma. Both defect because betrayal is the dominant strategy for each player individually. Yet both would prefer mutual silence (1 year each) to mutual betrayal (5 years each). The outcome is stable and terrible. Arms races are real-world Nash equilibria with the same structure.

2Explain why the positive-sum game illustration (A values item at $20, B at $5, trade at $10) refutes P2 of the running argument.›

P2 claims that whenever someone profits from investment, another person loses — implicitly treating investment as zero-sum. But in the positive-sum illustration, A gains $10 of consumer surplus (gets $20 of value, pays $10) and B gains $5 of producer surplus (gets $10 for something worth $5 to them). Neither gained at the other's expense — the exchange created $15 of new value. Productive investment is structurally similar: capital funds the creation of real goods, services, and employment. The gain comes from value produced, not from another party's loss. Since productive investment is demonstrably not zero-sum, P2's universal claim "whenever" is false — one counterexample is sufficient to defeat a universal.

3How does a repeated game change the prisoner's dilemma analysis? What conditions make cooperation individually rational?›

In a one-shot game, defection is dominant — it gives the highest payoff regardless of the other player's choice. In a repeated game, defecting today has future costs: the other player observes your behaviour and can punish future defection. When the shadow of the future is long enough — many remaining rounds, both players discount the future mildly, and behaviour can be observed — the discounted value of sustained cooperation exceeds the short-term gain from defecting and triggering retaliation. Cooperation becomes individually optimal without any external authority enforcing it. Tit-for-tat operationalises this: cooperate first, mirror the opponent's last move, and cooperation can be self-sustaining.

4Distinguish a cooperation problem from a coordination problem. Why does misidentifying one as the other destroy value?›

A cooperation problem (like the prisoner's dilemma) has partially opposed interests — each player is tempted to defect even though mutual cooperation would benefit both. The challenge is aligning incentives. A coordination game has aligned interests — both players want the same type of outcome — but the challenge is synchronising on which specific equilibrium to choose. If you misidentify a coordination problem as a conflict, you try to win when you should be synchronising: you end up on different equilibria and both lose. Many negotiations that feel adversarial are really coordination games — both parties want a deal, they just need to agree on terms. Treating them as zero-sum conflicts destroys the surplus that a deal would have created.

Progress saves in this browser. Next: the economics behind exchange →