Where are the Humans? A Scoping Review of Fairness in Multi-agent AI Systems

Fairness in multi-agent AI systems is understudied and fragmented: While fairness has been extensively researched in traditional AI applications, the paper finds that examining fairness in Multi-Agent AI (MAAI) systems—where multiple AI agents interact autonomously—remains "nascent and fragmented," suggesting this is an emerging and disorganized area of research.

Current approaches to fairness in MAAI are superficial and lack clear foundations: The analysis of 23 studies reveals that fairness in MAAI systems is "often addressed superficially" and "lacks robust normative foundations," meaning developers frequently don't have clear, well-grounded principles guiding their fairness decisions.

Agent autonomy and interactions create fairness challenges that are often overlooked: The paper identifies that the complex dynamics introduced by autonomous agents interacting with each other frequently go unaddressed, suggesting that fairness approaches designed for simpler AI systems don't adequately account for the complications that arise when multiple agents make independent decisions.

Fairness must be built in from the start, not added later: The authors argue that "fairness must be embedded structurally throughout the development lifecycle of MAAI, rather than appended as a post-hoc consideration," emphasizing that treating fairness as an afterthought is insufficient for these complex systems.

Meaningful fairness evaluation requires human involvement and clear objectives: The paper stresses that assessing fairness in MAAI systems requires "explicit human oversight, normative clarity, and a precise articulation of fairness objectives and beneficiaries"—meaning developers must clearly define who benefits and how, with humans actively involved in oversight.