Understanding Fairness in Congestion Games with Learning Agents
Keywords: Braess Paradox, Reinforcement Learning, Multi-Agent Systems, Fairness
TL;DR: Transport networks shape social mobility, but interventions can create inefficiencies and inequalities; using game theory and reinforcement learning, we analyze how transport expansions impact fairness and efficiency in heterogeneous populations.
Abstract: To understand the societal impacts of adapting urban transport networks, we must consider their impacts on mobility patterns. Expanding transport networks can worsen congestion, as seen in the Braess Paradox, where adding a route increases travel times. Traditional game-theoretic models often assume rational agents, but real-world behavior is dynamic and influenced by exploration and learning. Moreover, socioeconomic factors such as income can affect exploration rates, leading to disparities in travel times and access. To investigate these issues, we model agents as reinforcement learners and study how disparities in exploration rates impact fairness and efficiency in a toy problem. Our findings reveal that unequal exploration rates can disproportionately harm less explorative groups. Network interventions targeting efficiency can worsen inequities, even when they do not affect the price of anarchy. We highlight the need to account for disparities emerging from individuals' adaptation, when designing transport systems.
Type Of Paper: Full paper (max page 8)
Anonymous Submission: Anonymized submission.
Submission Number: 13
Loading