Welfarist Control Design promises fairer resource allocation in multi-agent systems

A new paper from Sophie Hall, Kai Zhang, Ilia Shilov, Heinrich H. Nax, and Saverio Bolognani tackles a crucial blindspot in modern automation: how to design control systems that allocate shared resources in line with what society actually wants. The work, posted on arXiv in June 2026, looks at multi-agent control problems ranging from highway lane management to energy access and pollution rights. Currently, engineers often fall back on industry conventions when making allocation decisions—a practice the authors argue is ethically irresponsible as these systems become more autonomous.

The paper introduces a framework called 'Welfarist Control Design,' which uses feedback loops to aggregate individual agent preferences into a formal control objective. It covers three paradigms: online feedback optimization (real-time adaptation), control of Markov decision processes (sequential decision-making under uncertainty), and model predictive control (planning over horizons). By leveraging the feedback nature of control systems, the approach can certify that the allocation matches a pre-defined societal mandate—making it both principled and auditable. The authors claim this is a significant step beyond current ad-hoc methods, offering a path to fairer and more transparent resource distribution in complex socio-technical networks.