This project studies algorithms for efficient and just allocation of scarce societal resources. The investigators will study fundamental questions related to algorithmic decision making in the context of disparate needs and resource availability. Key questions include how to define and quantify desirable outcomes in terms of both efficiency and fairness; how to predict outcomes of different types of interventions for different households; and how to optimize the allocation of scarce resources that achieve the best societal outcomes under constraints defined by notions of fairness, preferences of participants, and incentives created by service delivery systems as a whole. The algorithmic approach taken by the team will lead to more efficient and socially beneficial use of the limited resources available to communities for mitigating the central social problem of homelessness, while respecting the notions of need and fairness on which current allocation mechanisms are based. This project is among the first to explore how principles of local justice can be applied in the pursuit of ethical algorithmic intervention in society, contributing to the ongoing dialogue on fair machine learning and AI and Mechanism Design for Social Good. The project also contributes to the training of graduate students, research experiences for undergraduates, and broadening participation in computing through involvement of underrepresented populations in the research.
The team will develop new models to understand the effects of different algorithmic techniques for allocation of scarce societal resources, with a focus on defining objective functions that take into account both efficiency and considerations of fairness. They will develop analytical models of how agents evolve in their needs for services, and their vulnerability states, and solve these models in order to characterize optimal policies, given different assumptions about how intervention affects different types of agents. Using these models, the team will study the "price of justice" or the efficiency loss incurred in order to achieve different objectives. The theoretical modeling will be informed by real data on the use of homelessness services in a major metropolitan area. The team will also study new problems in optimal dynamic allocation of limited societal resources, considering carefully how predictions about the entire probability distributions of outcomes for agents (rather than just point estimates) can be used in the process of optimizing resource allocation over sets of agents that could need to use social services at different times.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
