Results 541 - 550 of 23763

I will talk about how the theories of local justice and of street-level bureaucracy can inform the potential application of prediction methods from machine learning in the allocation of scarce societal resources. In particular, I will discuss how separating the measurement aspects of ML methods from the allocation strategy can be helpful in ensuring that ML helps achieve societal goals. I will give some context on how these goals can be different for different domains through vignettes from our own research in allocation of services to households experiencing homelessness, K-12 educational supports, and caseworker time for providing support to those at high risk for eviction.

Many recent efforts use language models to assist with high-stakes decisions, for example via triaging medical symptoms, or by helping applicants navigate social services. Such applications require models to engage in decision making under uncertainty, weighing the costs and benefits of different courses of action under uncertainty about the user's true state (e.g., a patient's true diagnosis). How can system-builders evaluate, diagnose, and improve such capabilities? We propose a revealed-preference approach which elicits decisions alongside probabilistic beliefs from a model and infers the utility function which would rationalize those decisions. We then apply this framework to medical diagnosis tasks and trace how failures are jointly attributable to the beliefs and utilities that explain the model's actions.

Abstract not available

Every year, over 70,000 8th graders apply to over 800 high school programs in New York City. The process is grueling, requiring applicant families to learn about programs, form preferences, and navigate admissions likelihoods. Substantial prior work, including our own, has shown that there are disparities in how students apply to and match with high performing options, partially due to this complexity. I will talk about an ongoing collaboration with the NYC Public Schools, in which we designed and deployed an informational intervention to help students from underserved middle schools discover high-performing, nearby high schools where they have a strong individual admissions likelihood. However, recommending specific programs brings a methodological challenge: if many applicants are recommended the same program, then the recommendations are self-defeating, as many will be rejected and the admissions likelihood estimates will be proven incorrect. Thus, our individualized recommendations are "congestion-aware," such that the admissions likelihoods are correct in equilibrium.