TL;DR
This paper introduces "language-based games," a flexible framework that generalizes classical game theory to model richer player preferences, including those based on beliefs and reference points. By defining utility over "situations" described by a formal language, it unifies existing models like psychological games and reference-dependent preferences, while also naturally accommodating "coarse beliefs."
What Problem Does This Paper Address?
Classical game theory assumes players' preferences are solely based on the final outcomes of a game. However, many real-world scenarios involve preferences that depend on beliefs about others' actions, expectations, or even how outcomes compare to a reference point (like in behavioral economics). Existing models that try to capture these nuances, such as psychological game theory or reference-dependent preferences, often require complex, hierarchical structures for beliefs. This paper aims to provide a more unified and general framework that can elegantly handle these richer preferences, including those that are "coarse" or less precise, without necessarily resorting to infinite belief hierarchies.
Methods
The core innovation is the concept of a "language-based game." Instead of defining utility functions directly over game outcomes (strategy profiles), utility is defined over "situations." A situation is formally defined as a maximal consistent set of formulas from an underlying formal language. This language dictates what aspects of the game state a player can "describe" and thus base their preferences on.
- Generalization: This framework generalizes classical games by allowing the underlying language to be richer than just the set of outcomes.
- Unification: It unifies existing models:
- Classical games: Use a language that only describes outcomes.
- Psychological games: Use a language that can express beliefs.
- Reference-dependent preferences: Use a language that can express comparisons between outcomes and expectations.
- Coarse Beliefs: The framework naturally accommodates "coarse beliefs" (e.g., categorizing expectations into a few levels rather than precise probabilities) by using languages with limited expressive power.
- Equilibrium Concepts: The paper generalizes Nash equilibrium and rationalizability to this new setting.
Datasets
Not discussed in the paper. The paper focuses on theoretical framework development and uses illustrative examples rather than empirical datasets.
Key Insights and Ideas
- Language as the Key: The expressive power of the underlying language is the central determinant of what kinds of preferences can be modeled. A richer language allows for more complex preferences.
- Situations, Not Just Outcomes: Shifting the domain of utility from outcomes to "situations" (which can include beliefs, expectations, etc.) is the crucial step for generalization.
- Coarse Beliefs are Powerful: The paper highlights that even simple, "coarse" representations of beliefs can be sufficient to capture complex psychological phenomena, challenging the assumption that fine-grained probabilities are always necessary.
- Unified Framework: This approach provides a single, elegant way to represent diverse preference structures that previously required separate, specialized models.
- Tension in Surprise: Games where players aim to surprise their opponents often lead to a fundamental tension, making Nash equilibria difficult or impossible to find.
Key Results
- Generalization of Classical Games: The framework can recover classical normal-form games by using a language that only describes outcomes.
- Existence of Rationalizable Strategies: Under certain "compactness" assumptions on the underlying language, every language-based game is guaranteed to have rationalizable strategies.
- Non-existence of Nash Equilibrium: The paper demonstrates through examples (like "indignant altruism") that Nash equilibria may not exist in general for language-based games, particularly when surprise is a key element.
Limitations and Future Work
- Nash Equilibrium Existence: While rationalizability is shown to exist under mild conditions, Nash equilibrium is not guaranteed in all language-based games, especially those involving surprise. This is acknowledged as a limitation and a point of departure for further research.
- Formalization of Language: The paper uses propositional logic and modal logic as examples, but the general framework is designed to be adaptable to various formal languages. Future work could explore the implications of different language structures.
- Computational Complexity: While the framework is conceptually elegant, the computational complexity of finding equilibria or determining rationalizability in specific language-based games is not fully explored in this extended abstract.
- Empirical Validation: The paper relies on illustrative examples. Future work could involve applying the framework to real-world scenarios and testing its predictive power against empirical data.
Summary updated