Game Theory · GT-19
Chess is perfect information — every move is visible. Poker is imperfect information — hidden cards change everything about how to reason.
A game has perfect information if every player, at every point, knows the complete history of all moves made so far (chess, tic-tac-toe). It has imperfect information if some moves are hidden or simultaneous (poker, sealed-bid auctions, most negotiations). The distinction fundamentally changes which solution concepts and strategies apply.
The formal distinction was established in von Neumann and Morgenstern's 1944 foundational text, building on earlier work on 'games of perfect information' in Zermelo's 1913 chess analysis.
The Mechanism
What each player can see, at each decision point
The dashed oval is the whole idea — a player facing an 'information set' with more than one node genuinely cannot tell which specific situation they're in, and must choose one action that works across all of them. Perfect-information games never have this ambiguity.
01 · PERFECT INFORMATION GUARANTEES A PURE-STRATEGY EQUILIBRIUM
Zermelo's theorem, 1913
Ernst Zermelo proved that any finite, perfect-information, two-player zero-sum game (like chess) has a determined outcome under optimal play — in principle, one side either always wins, always loses, or the game is a forced draw with best play from both sides. This is why chess is 'solved' in theory even though it remains practically unsolved due to sheer computational scale.
02 · IMPERFECT INFORMATION IS WHERE BLUFFING AND SIGNALING LIVE
The hidden-node structure is what makes deception strategically meaningful
In a perfect-information game, there's nothing to hide — every move is visible, so bluffing has no strategic purchase. Imperfect information is a structural precondition for signaling, cheap talk, and strategic deception to matter at all, which is why poker and negotiation theory both live firmly in this branch of game theory.
03 · MOST REAL-WORLD STRATEGIC SITUATIONS ARE IMPERFECT-INFORMATION
Chess is the exception, not the rule
Business negotiations, auctions, hiring decisions, and most competitive interactions involve genuinely hidden information (the other party's true costs, true valuation, true alternatives) — treating them with perfect-information tools like simple backward induction, without accounting for what each side can't observe, is a common and costly modeling error.
Where It Fails / Inversion
Where it fails / inversion
Applying perfect-information reasoning (like naive backward induction) to a situation that's actually imperfect-information leads to systematically wrong predictions — assuming your counterpart in a negotiation can see everything you see (your true reservation price, your actual alternatives) when they genuinely cannot will misjudge both what they'll offer and what they'll believe about your offers.
How To Use It
Worked example · a salary negotiation
A salary negotiation is a clean imperfect-information game: neither party knows the other's true reservation point (the employer's real budget ceiling, the candidate's real walk-away number). Treating it as if it were perfect information — assuming your counteroffer will be read exactly as intended, with full context — ignores that the other side is making inferences from incomplete signals, which is why anchoring, framing, and credible signaling (a competing offer, a specific researched market rate) matter enormously more here than they would in a perfect-information setting like a posted, non-negotiable price.
How to use it
Before applying any game-theoretic reasoning to a real decision, explicitly identify what each side can and cannot observe. If genuine hidden information exists, don't reach for perfect-information tools (simple backward induction, 'obviously they'll do X') — reach for signaling, screening, and Bayesian reasoning instead, since those are the tools built for exactly this structure.
See Also