Seminaria w Instytucie Podstaw Informatyki PAN

Streszczenie (autorskie):

Multi-agent systems are increasingly used in applications where agents must make strategic decisions while considering properties that are important in human societies. In this talk, I will present our recent work on strategic reasoning about such properties in multi-agent systems. In particular, I will focus on privacy and opacity, fairness, and responsibility, including counterfactual responsibility attribution. We develop logical and quantitative frameworks that allow us to specify and verify these properties while accounting for agents’ strategic interactions, uncertainty, and system behaviours. I will discuss how formal verification can help answer questions such as: What information can an agent infer from observing others? Is an outcome fair across different agents? Who is responsible for an outcome, and how would the outcome change if an agent had acted differently? Through these examples, I will highlight how logic, strategic reasoning and formal verification can provide principled tools for building more trustworthy, transparent, fair, and responsible multi-agent AI systems.