AI Glossary / Graphs and symbolic AI / Intermediate
Strategic form
Definition
Strategic form, also called normal form, represents a game by specifying its players, each player's available strategies, and the payoff each player receives for every combination of strategies. For a small two-player game, it is often shown as a payoff matrix. The representation makes the interdependence of choices explicit: one player's result depends on the others' choices too.
What it means
A strategy is a complete plan for the choices a player may face. In a simple one-move game, a strategy can be just one action. In a sequential game, it can specify different actions for different situations, including situations that do not occur in the eventual play.
The strategic form records the resulting payoff relationships. It does not display timing and information as directly as an extensive-form game tree. It is also a representation rather than a prediction: a solution concept, such as Nash equilibrium, adds assumptions about how players choose within that representation.
How it works
Identify the players and enumerate their feasible strategies. For each strategy profile, determine the resulting outcome and assign each player's payoff. If chance affects the outcome, expected payoffs can be used under an appropriate expected-utility formulation.
For two players with finitely many strategies, put one player's strategies in rows and the other's in columns. Each cell contains a payoff pair in a clearly stated player order. Compare a player's alternatives while holding the other player's strategy fixed. This supports best-response and equilibrium analysis. Mixed strategies extend the model by assigning probabilities to pure strategies rather than introducing new physical actions.
Three practical examples
- A coordination exercise can model two software agents choosing compatible operating conventions.
- A resource-sharing simulation can compare the incentives created by different allocation rules.
- A game-theory lesson can translate a short sequential interaction into complete contingent strategies and their associated payoffs.
Think of it this way
Think of a table listing what each participant receives for every combination of their plans. Looking down one column shows how the row player's choice matters when the other plan is fixed. The analogy is limited because a payoff table is a model of preferences and consequences, not direct evidence that real participants will behave exactly as its equilibrium analysis suggests.
Worked example
Consider an original coordination game between agents R and C. Each simultaneously chooses operating mode X or Y, without observing the other's choice. If both choose X, the payoff pair is (4,3). If both choose Y, it is (2,5). Either mismatch gives (0,0). The first number always belongs to R and the second to C.
Represent R's choices as rows and C's as columns. The four cells are therefore X,X: (4,3); X,Y: (0,0); Y,X: (0,0); and Y,Y: (2,5).
If C chooses X, R prefers X because four exceeds zero. If C chooses Y, R prefers Y because two exceeds zero. Similarly, C prefers X when R chooses X, and Y when R chooses Y. Both matching outcomes are pure-strategy Nash equilibria: neither player benefits by changing alone while the other choice remains fixed.
The agents disagree about which matching equilibrium they prefer. R receives more at X,X, while C receives more at Y,Y. The existence of an equilibrium does not choose between those outcomes or establish that their division of payoffs is fair.
There is also a mixed-strategy calculation. Let p be the probability R chooses X and q the probability C chooses X, with independent randomization. R's expected payoff from choosing X is 4q, and from Y it is 2(1 − q). Indifference requires q = 1/3. C's expected payoff from X is 3p, and from Y it is 5(1 − p). Indifference requires p = 5/8.
Thus these probabilities form a mixed equilibrium for the stated matrix. R's expected payoff is 4/3 and C's is 15/8. These are model calculations, not recommended operating choices.
Now imagine C observes R's choice before acting. C's complete plans include choose X after either observation; choose Y after either observation; match R; and oppose R. That gives four pure contingent strategies, not just two action labels. The changed timing and information create a different game representation.
Illustrative case study
In a fictional multi-agent design workshop, engineers use a payoff matrix to discuss compatibility choices. They first verify the player order in every payoff pair and ask whether the numbers represent preferences consistently.
When participants propose allowing one agent to observe the other's choice, the team revises the model rather than reusing the simultaneous-choice analysis unchanged. This fictional exercise illustrates the value of making information and incentives explicit, without claiming that the table predicts deployed behavior.
Common misconceptions and mistakes
Strategic form is not restricted to games where each player physically acts only once. Complete contingent plans can represent sequential play. Conversely, an action list can be incomplete when a player must respond differently at several decision points.
Do not assume every game is zero-sum or that every matrix has a pure-strategy equilibrium. Do not confuse the largest combined payoff with a stable outcome under unilateral deviations. Finally, a matrix's numerical payoffs need a defensible interpretation, particularly when calculating expected payoffs under randomization.
Business use
The representation can make incentives easier to inspect in multi-agent systems and organizational simulations. It highlights situations where individually attractive actions conflict with coordination or where multiple stable outcomes exist.
Review the available strategies, information, and payoff assumptions before interpreting a result. A missing option or an omitted consequence can change the analysis. Use small matrices to explain a mechanism, then evaluate whether the simplified game captures the operational setting. Treat equilibrium as a model-dependent analytical concept rather than a guarantee of actual behavior.
People also ask
Is strategic form the same as normal form?
Yes, the terms commonly refer to this representation.
Does every strategy equal a single action?
No. Sequential settings can require complete contingent plans.
Is a payoff matrix itself a solution?
No. It specifies the game; further analysis is needed to identify equilibria or other solution concepts.
Check your understanding
Why are both matching outcomes equilibria in the worked example?
Answer: At either matching outcome, changing one player's action alone causes a mismatch and lowers that player's payoff to zero.
Related terms
- Extensive form
- Pure strategy
- Mixed strategy
- Strategy profile
- Normal-form game
- Payoff matrix
- Best response
- Nash equilibrium
- Coordination game
Learn next
Next, verify the mixed-equilibrium payoffs by substituting q = 1/3 into both of R's action payoffs and p = 5/8 into both of C's. Then explain why indifference allows randomization but does not mean the two agents receive equal expected payoffs.
More in Graphs and symbolic AI
- Stoer-Wagner minimum cut - The Stoer - Wagner minimum-cut algorithm finds a global minimum edge cut in an undirected graph with nonnegative edge weights.
- Stopping rule - A stopping rule specifies when a sequential process should end or switch to another option, using information available at the time of the decision.
- Stopping time - A stopping time is a random time whose occurrence can be determined using only the information available so far.
- Strategy - A strategy is a player's complete plan for choosing actions in a game or other model of strategic interaction.
- Strategy-proof - A rule is strategy-proof when no individual participant can obtain a better outcome, according to their true preferences, by misreporting their private...
- Streaming Inference Pipeline - A streaming inference pipeline is an application that continuously receives events, applies a trained machine-learning model, and sends predictions to an...
Sources
Primary source: Muhamet Yıldız, MIT Game Theory lecture notes, Basic Framework and Normal-Form Representation: https://ocw.mit.edu/courses/14.126-game-theory-spring-2024/mit14_126_s24_yildiz-lecture-notes.pdf . The mode-coordination matrix, mixed-equilibrium arithmetic, and contingent-plan variation are original.