Recursive Joint Simulation in Games: Equivalence to Repeated Games and Practical Implications
Core Concepts
Recursive joint simulation games are strategically equivalent to infinitely-repeated games, inheriting properties like the folk theorem, with practical implications for cooperative outcomes.
Abstract
AI agents can leverage recursive joint simulation for strategic interactions. The paper explores equivalence to repeated games, showing how simulations can lead to cooperative equilibria. Different scenarios and variants are considered, including voluntary simulation and limited simulation budgets. Theoretical discussions on self-locating beliefs and objections to the simulation framework are also addressed. Practical challenges of creating indistinguishable simulations from reality are highlighted, suggesting future research directions.
Recursive Joint Simulation in Games
Stats
"The resulting interaction is strategically equivalent to an infinitely repeated version of the original game."
"There exists p0 ∈ [0, 1) s.t. for every p ∈ [p0, 1), there is some π ∈ NE(RJS(G, p)) s.t. ∀i : ui(π) = vi."
Quotes
"The scenario that we study in this paper falls primarily in the latter category."
"Recursive joint simulation allows the agents to reach mutual cooperation in equilibrium."
"We hope that by exploring these threads further, we might eventually arrive at frameworks that allow us to think about these topics in a more unified manner."