ABSTRACT

The Community Responsive Algorithms for Social Accountability (CRASA) project will establish a model for accountability that can be applied across a comprehensive range of algorithms being used in public policy in various contexts. The project?s goals will be achieved through a three-year community-based participatory research program focusing on Harris County, Texas, incorporating input from stakeholders in local government, the legal community, and industry. Relying on input from community stakeholders, this project will develop an algorithm-accountability benchmark (AAB) that will be applied to a variety of public policy algorithms used by governments, advocacy groups, and corporations for design and evaluation. In co-operation with community partners, CRASA will promote broad application of this benchmark approach in the public policy sphere. The development and explication of specific standards through the AAB will provide a clear and reproducible touchstone for development, evaluation, and implementation of algorithms in public policy. CRASA will also contribute to education and workforce development by producing a set of educational materials on the use of algorithms that can be easily accessed by legal professionals and the general public; by developing a multidisciplinary undergraduate/graduate course for students on the ethics of artificial intelligence; and, by training the next generation of scholars interested in responsive and transparent algorithms for use in public policy.

The use of algorithms in public policy has expanded dramatically in recent decades. They currently play an active part in informing policymakers in their decisions related to criminal justice, public education, the allocation of public resources, and even national defense strategy. However, standards of accountability reflecting current legal obligations and societal concerns have lagged far behind their extensive use and influence. CRASA?s community-based research strategy will answer questions about how to make the use of algorithms more accountable, and, specifically, how benchmarks of accountability can be established for these algorithms that will engender legitimacy and public trust. The project?s overall research strategy involves five objectives: Objective 1 ? collect needed information through interviews with stakeholders representing a wide variety of interests in the application of public policy algorithms and establish a community advisory board that meets regularly to guide and evaluate the research; Objective 2 ? conduct a comprehensive review of the quickly evolving legal precedents and academic proposals being set forth for algorithm regulation; Objective 3 ? design an algorithm-accountability benchmark (AAB) that can be applied across policy areas to evaluate and compare algorithms in terms of their accountability standards; Objective 4 ? conduct behavioral experiments, both within the legal community and the general public, to evaluate public trust and understanding of the AAB; and Objective 5 ? develop a software scoring toolkit that will provide the AAB score for any software and demonstrate its use in two application domains: criminal risk estimation and facial recognition.

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.

Please report errors in award information by writing to: awardsearch@nsf.gov.