Summary of "The Compleat Strategyst"

Core Idea

Game theory is a mathematical framework for making optimal decisions in situations where the outcome depends not just on your choices but on the choices of an opponent with conflicting interests
The fundamental insight: in conflict situations, the best strategy is often not the one that maximizes your gain if everything goes well, but the one that guarantees the best worst-case outcome — the minimax principle
Deliberately oversimplified models of conflict, like deliberately oversimplified models in physics, can illuminate critical aspects of far more complex real-world situations

A game (in game theory) is any situation involving: (1) two or more parties with opposing interests, (2) choices each party controls, (3) elements neither party controls (chance/Nature), and (4) a payoff structure determined by the combination of all choices
Strategy means a complete plan of action — not just what you do first, but what you do in every possible situation that might arise
Zero-sum games: one player's gain is exactly the other's loss. This is the class of games the book covers, and it provides the cleanest mathematical framework
The key question: Given that your opponent is intelligent and trying to undo you, what is the best you can guarantee for yourself regardless of what they do?

Each player should choose the strategy that maximizes their minimum possible payoff (or equivalently, minimizes their maximum possible loss)
This leads to the concept of a saddle point — when both players' minimax strategies converge on the same outcome, that outcome is the value of the game
Why it works: If you play minimax, no opponent, no matter how clever, can do better against you than the game's value. Any deviation from minimax by your opponent can only help you

Pure strategy: always do the same thing. Works when a saddle point exists — when one strategy is simply best regardless of what the opponent does
Mixed strategy: randomize among strategies with specific probabilities. Required when no saddle point exists — when any predictable pattern would be exploited by a smart opponent
The profound implication: sometimes the optimal behavior is deliberately unpredictable. A poker player who never bluffs is exploitable; one who always bluffs is exploitable; but one who bluffs at the right random frequency is not

A strategy is dominated if there exists another strategy that does at least as well in every scenario and better in at least one
Dominated strategies can be eliminated, simplifying the game — a rational player should never use a dominated strategy
Successive elimination of dominated strategies can sometimes solve a game completely

2x2 games (two strategies each): Solved by checking for saddle points, and if none exists, by a simple arithmetic formula for the optimal mixed strategy
Larger games: Williams walks through 3x3, 4x4, and larger games, introducing progressively more powerful methods
The pivot method (Chapter 6, added in the revised edition): A general algorithm, based on Dantzig's Simplex Method from linear programming, that can solve any matrix game — a major advance over the piecemeal methods used for smaller games

Game theory assumes both players are rational and intelligent — it tells you how to play against the best possible opponent
Against weaker opponents, minimax is conservative but safe. Against strong opponents, it is the only strategy that cannot be exploited
The theory reveals that rationality in conflict is fundamentally different from rationality in isolation — what makes sense when nature is your only opponent may be disastrous when facing a thinking adversary

One of game theory's most counterintuitive results: deliberate randomness is sometimes optimal
This is not chaos or indecision — it is precisely calibrated unpredictability, where the probabilities are computed to make your opponent indifferent among their options
Applications extend far beyond games: military deception, competitive pricing, audit scheduling, and any situation where predictability is a vulnerability

Williams repeatedly emphasizes that game theory, like the Theory of Gravitation, works by heroic oversimplification — the model captures some essential features of conflict while ignoring most of the complexity
Newton replaced planets with mass-points and predicted their orbits for 250 years. Game theory replaces human conflict with payoff matrices and provides strategic guidance
The mortality rate of such theories is high, but the successes of simplified models in physics suggest the approach has merit for human conflict too

It is the best introduction to game theory ever written for non-mathematicians — requiring nothing beyond basic arithmetic with negative numbers
The examples are deliberately whimsical (generals choosing attack routes, couples negotiating evening plans, executives setting prices) but illustrate real strategic principles
Williams's prose is witty and self-aware in a way that makes the mathematical content genuinely entertaining — a remarkable feat for a book about matrix algebra
It demonstrates that strategic thinking is a learnable skill, not an innate talent, and that even partial understanding of game theory can improve decision-making in conflict situations

In the important conflicts in your life, are you playing a pure strategy that a smart opponent could exploit?
When you face a decision under uncertainty, are you optimizing for the best case or protecting against the worst case, and which is more appropriate?
Where in your professional or personal life would deliberate unpredictability be an advantage rather than a liability?
Are you treating situations as one-person games against nature when they are actually two-person games against a thinking opponent?