Algorithm Instance Games

TL;DR

This paper introduces algorithm instance games (AIGs) to analyze how the specific algorithms used to determine game outcomes influence strategic behavior, focusing on set-cover based games with deterministic and non-deterministic algorithms.

Contribution

It characterizes Nash equilibria in set-cover based AIGs, revealing how algorithm determinism affects agents' strategic choices and equilibrium structures.

Findings

Deterministic algorithm leads to equilibria with single-element strategies.

Non-deterministic algorithm allows multiple strategies, including zero or all elements.

Equilibrium strategies differ significantly between the two algorithm types.

Abstract

This paper introduces algorithm instance games (AIGs) as a conceptual classification applying to games in which outcomes are resolved from joint strategies algorithmically. For such games, a fundamental question asks: How do the details of the algorithm's description influence agents' strategic behavior? We analyze two versions of an AIG based on the set-cover optimization problem. In these games, joint strategies correspond to instances of the set-cover problem, with each subset (of a given universe of elements) representing the strategy of a single agent. Outcomes are covers computed from the joint strategies by a set-cover algorithm. In one variant of this game, outcomes are computed by a deterministic greedy algorithm, and the other variant utilizes a non-deterministic form of the greedy algorithm. We characterize Nash equilibrium strategies for both versions of the game, finding that agents' strategies can vary considerably between the two settings. In particular, we find that the version of the game based on the deterministic algorithm only admits Nash equilibrium in which agents choose strategies (i.e., subsets) containing at most one element, with no two agents picking the same element. On the other hand, in the version of the game based on the non-deterministic algorithm, Nash equilibrium strategies can include agents with zero, one, or every element, and the same element can appear in the strategies of multiple agents.