Reasoning about Bias in Multi-Agent Systems Verification

Vadim Malvone, Chunyan Mu

Keywords: Multi-Agent Systems, Strategic Logics, Model Checking, Bias

Abstract: This paper investigates formal analysis of bias properties in Multi-Agent Systems. We present a formal framework to reason about bias properties within these models. We propose a novel definition of bias based on attribute interference and formalise the notion of bias by trace equivalence regarding bias-sensitive lattice. Intuitively, a model is considered unbiased or fair if the input values of bias-sensitive attributes will not affect the system's output, ensuring that the system treats all bias-sensitive attributes equitably. To effectively capture and specify bias and fairness properties, we extend Alternating-time Temporal Logic (ATL) with new bias operators as bATL. Using model checking techniques, we propose algorithm to rigorously verify these properties within the game structure.

Area: Representation and Reasoning (RR)

Generative A I: I acknowledge that I have read and will follow this policy.

Submission Number: 168