A Dynamic Analysis of Interactive Rationality

Eric Pacuit, Olivier RoyLORI 2011 · DOI

TL;DR

This paper introduces a formal framework for understanding how players in a game arrive at stable "informational contexts" through a process of iterated reasoning about each other's rational choices. They show that such a process stabilizes under specific conditions related to the decision rule and how information is updated, offering a dynamic perspective on game theory solutions.

What Problem Does This Paper Address?

Classical game theory often assumes fixed informational contexts (e.g., common knowledge of rationality). However, in reality, players' beliefs about each other's reasoning evolve as they deliberate. This paper tackles the question of how these informational contexts themselves arise and whether this dynamic process of reasoning can lead to a stable state. They are particularly interested in how players move from an "ex ante" stage (before decisions are made) to an "ex interim" stage (considering current beliefs and strategies) through a process of "rational deliberation." This is crucial because different informational contexts can lead to drastically different game outcomes, and understanding how these contexts are rationally formed is key to understanding strategic interaction.

Methods

The authors propose a framework based on plausibility models, which are a way to represent agents' beliefs. An informational context is modeled as a plausibility structure consisting of states and a plausibility ordering (representing what's considered more or less plausible). The core of their method is the concept of a deliberation sequence, which is an infinite sequence of plausibility models. Each step in this sequence involves:

  1. Categorization: Players use a specific choice rule (e.g., weak dominance) to categorize their available strategies into "good" (rationally permissible) and "bad" (irrational) options, based on their current beliefs about what strategies are in play.
  2. Belief Dynamics (Model Transformation): This categorization is then used to update the plausibility model. The authors consider various model transformations (like public announcement, conservative upgrade, and radical upgrade) that represent how agents incorporate new information about rational choices.

The central question is whether these deliberation sequences stabilize, meaning they eventually reach a point where the model no longer changes. They identify conditions on the choice rule and the model transformation that guarantee stabilization.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Informational contexts are not static: The paper emphasizes that the beliefs players hold about each other's reasoning are dynamic and evolve during deliberation.
  • Rational deliberation as a process: Instead of assuming a fixed solution concept, the paper models how players arrive at a solution through a step-by-step process of reasoning about their own and others' rationality.
  • Stabilization is key: The core contribution is identifying conditions under which this dynamic process of reasoning stabilizes. This stabilization is crucial for a coherent understanding of strategic interaction.
  • The role of choice rules and belief updates: The paper highlights that both how players categorize strategies (the choice rule) and how they update their beliefs based on this categorization (the model transformation) are critical for whether the deliberation process stabilizes.
  • Admissibility paradox: The paper uses the example of iterated admissibility to illustrate how certain choice rules and update mechanisms can lead to non-stabilizing, looping deliberation processes, mirroring known paradoxes in game theory.

Key Results

The paper presents a Stabilization Theorem (Theorem 3) which states that for a finite game, if the choice rule is monotonic and the belief transformation satisfies certain properties (related to how categorizations are incorporated and how beliefs are updated), then every deliberation sequence stabilizes. They also provide an observation that demonstrates how certain standard model transformations, when applied to the problem of iterated admissibility, do not lead to stabilization, thus offering a dynamic explanation for known paradoxes.

Limitations and Future Work

  • Probabilistic and Graded Beliefs: The current framework primarily uses qualitative beliefs (plausibility models). The authors acknowledge that extending this dynamic perspective to probabilistic or graded belief models is a significant but non-trivial task for future work.
  • Specific Choice Rules and Transformations: While the general theorem is presented, the paper focuses on illustrating the concepts with weak dominance and specific update rules. Further research could explore the interplay of other choice rules and more complex belief dynamics.
  • "Proper" Contexts: The paper aims to analyze how "proper" informational contexts arise but leaves the precise definition of what constitutes a "proper" context open to further investigation, particularly concerning substantive assumptions about beliefs.
  • Common Knowledge of Rationality: The paper touches upon the challenges of common knowledge of rationality (e.g., Samuelson's paradox) and suggests their dynamic framework offers a new perspective, but a full epistemic characterization is left for future work.

Summary updated

BibTeX
@inproceedings{pacuit2011dynamic,
  author = {Eric Pacuit and Olivier Roy},
  title = {{A Dynamic Analysis of Interactive Rationality}},
  booktitle = {LORI},
  year = {2011},
  doi = {10.1007/978-3-642-24130-7\_18},
}
Figure