Part 5
5 min read5 headingsSplit lesson page
Lesson overview | Previous part | Lesson overview
Nash Equilibria: Part 7: Common Mistakes to References
7. Common Mistakes
| # | Mistake | Why It Is Wrong | Fix |
|---|---|---|---|
| 1 | Treating equilibrium as social optimality | A Nash equilibrium can be inefficient or unfair. | Compare equilibrium outcomes with Pareto and welfare criteria. |
| 2 | Checking only one player's incentive | Equilibrium requires every player to lack profitable unilateral deviation. | Compute best responses for all players. |
| 3 | Ignoring mixed strategies | Some finite games have no pure equilibrium. | Use probability distributions over actions and the indifference principle. |
| 4 | Applying minimax to non-zero-sum games blindly | Minimax value is a zero-sum guarantee, not a general welfare solution. | State whether payoffs are strictly opposed before using minimax. |
| 5 | Confusing learning convergence with equilibrium | A learning process can cycle, diverge, or converge to a non-equilibrium behavior. | Track regret, exploitability, and stationarity separately. |
| 6 | Forgetting that other agents adapt | In multi-agent systems, each learner changes the data distribution of the others. | Model policies jointly and monitor nonstationarity. |
| 7 | Using average-case metrics against adaptive attackers | An adaptive opponent targets the worst exploitable gap. | Define threat sets and robust objectives. |
| 8 | Equating red teaming with complete security | Red-team examples are samples, not proofs against all attacks. | Use adaptive evaluation and explicit threat models. |
| 9 | Treating GAN instability as ordinary optimization only | GANs are games whose gradients can rotate instead of descend. | Analyze generator and discriminator objectives jointly. |
| 10 | Letting game abstractions erase values | Payoff design determines incentives and side effects. | Audit utility functions, constraints, and welfare implications. |
8. Exercises
-
(*) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(*) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(*) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(**) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(**) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(**) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(***) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(***) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(***) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
-
(***) Work through a game-theory task for nash equilibria.
- (a) State the players, actions, and payoffs.
- (b) Compute or characterize best responses.
- (c) Decide whether the proposed joint strategy is stable.
- (d) Interpret the result for an AI, LLM, or adversarial system.
9. Why This Matters for AI
| Concept | AI Impact |
|---|---|
| Best response | Explains how users, attackers, or agents adapt to a model policy |
| Nash equilibrium | Defines strategic stability for GANs, self-play, routing, and agent systems |
| Mixed strategy | Motivates randomized defenses, stochastic policies, and exploration |
| Minimax value | Formalizes robust worst-case guarantees |
| Exploitability | Measures how far a policy is from strategic stability |
| No-regret learning | Connects repeated play to approximate equilibrium |
| Security game | Models limited defensive resources against adaptive threats |
| Payoff design | Shows why objective misspecification creates strategic side effects |
10. Conceptual Bridge
Nash Equilibria follows causal inference because interventions often change incentives. Chapter 22 asks what changes when an action is taken. Chapter 23 asks what happens when other agents see that action, learn from it, and respond strategically.
The backward bridge is intervention. A policy change can have a causal effect, but if users or attackers adapt, the effect becomes part of a game. The forward bridge is measure theory: later probability foundations make the stochastic strategies, repeated games, and distributional assumptions more rigorous.
+--------------------------------------------------------------+
| Chapter 22: intervention and causal mechanisms |
| Chapter 23: strategic adaptation and adversarial objectives |
| Chapter 24: rigorous probability and measure foundations |
+--------------------------------------------------------------+
References
- Nash. Equilibrium Points in N-Person Games. https://pmc.ncbi.nlm.nih.gov/articles/PMC1063129/
- Osborne and Rubinstein. A Course in Game Theory. https://mitpress.mit.edu/9780262650403/a-course-in-game-theory/
- Shoham and Leyton-Brown. Multiagent Systems. https://www.masfoundations.org/toc.pdf
- Goodfellow et al.. Generative Adversarial Nets. https://arxiv.org/abs/1406.2661