Game Theory · GT-09
Two rational actors, doing the individually smart thing, land on the one outcome worse for both.
Each party does better by betraying the other regardless of what the other does — yet mutual betrayal leaves both worse off than mutual cooperation would have. The canonical demonstration that individually rational choices can produce a collectively irrational result.
Formalized in 1950 by Merrill Flood and Melvin Dresher at RAND Corporation as an experiment in strategic behavior; the prison-sentence framing and the name were added shortly after by mathematician Albert Tucker, who used it to explain the result to a lay audience.
The Mechanism
Two coffee chains, deciding whether to discount — click a cell
Discounting dominates for both chains, regardless of what the other does — so both discount, both land in the $4M/$4M cell, and both would have made more ($8M each) by silently holding price together. Neither can get there alone: whoever holds price while the other discounts gets the worst outcome of all ($2M, "the sucker").
01 · DOMINANT STRATEGY
Defect wins the argument every time you run it
Ask "if the other chain holds price, am I better off discounting?" Yes: $11M beats $8M. Ask "if the other chain discounts, am I better off discounting too?" Also yes: $4M beats $2M. Discounting is a dominant strategy — it wins regardless of the other player's move, so a rational actor never needs to guess what the opponent will do.
02 · INDIVIDUALLY RATIONAL, COLLECTIVELY WORSE
Both players reasoning correctly still lose together
Both chains follow the same airtight logic, both land on Discount, and both end up at $4M — worse than the $8M they'd both have gotten by holding price. Nobody made a mistake. That's what makes this different from ordinary bad decision-making: it's the *correct* individual reasoning that produces the bad joint outcome.
03 · ONE SHOT vs. REPEATED
The trap is specific to a single, isolated encounter
Everything above assumes this happens once, with no future and no reputation at stake. Play it repeatedly with the same counterpart, and the calculation changes completely — future retaliation becomes possible, and strategies like Tit-for-Tat can sustain cooperation indefinitely (see Repeated Games & the Folk Theorem).
Where It Fails / Inversion
The trap: importing one-shot logic into a repeated relationship
The classic result is a description of a single, anonymous, one-time interaction with no memory and no future — and it's routinely misapplied to situations that don't meet that description. Real business relationships, marriages, and long-standing partnerships are repeated games with reputational consequences, which is exactly why cooperation survives far more often in practice than the one-shot matrix predicts. Robert Axelrod's 1980 computer tournaments showed simple, forgiving reciprocal strategies (Tit-for-Tat) consistently outperforming pure defection once the game repeats.
The second, subtler trap: treating a genuinely repeated or reputation-bearing interaction as if it were one-shot, and defecting pre-emptively "to be safe" — which can manufacture the very breakdown in trust the theory describes, where none was structurally necessary.
How To Use It
Worked example · the airline fare war pattern
Two airlines serving the same route both do best if both hold fares high; each individually does even better by undercutting the other while fares stay high elsewhere; and if both undercut, both lose margin industry-wide (the $4M/$4M cell). This exact structure has played out repeatedly in aviation, and it explains why price wars are hard to prevent through appeals to "rational restraint" alone — restraint is not the dominant strategy in a single round.
What actually stabilizes prices in real markets that keep repeating this game isn't a handshake agreement (illegal collusion aside) — it's the credible expectation of retaliation in future rounds, transparent pricing that makes defection instantly visible, and a long enough time horizon that the future cost of a price war outweighs this quarter's temptation.
How to use it
Before assuming a partner, competitor, or colleague will "do the smart, cooperative thing," check whether you're actually in a repeated game with visible history and future consequences, or a genuine one-shot interaction. If it's one-shot and anonymous, expect the dominant strategy to win — and if you want cooperation instead, your real lever is changing the game's structure (make it repeated, make defection visible, add a future) rather than appealing to goodwill.
See Also