Through a field scan, this paper identifies emerging best practices as well as methods and tools that are becoming commonplace, and enumerates common barriers to leveraging algorithmic audits as effective accountability mechanisms.
ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT)
An emerging concern in algorithmic fairness is the tension with privacy interests. Data minimization can restrict access to protected attributes, such as race and ethnicity, for bias assessment and mitigation. This paper examines how this “privacy-bias tradeoff” has become an important battleground for fairness assessments in the U.S. government and provides rich lessons for resolving these tradeoffs.
ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT)
Recent studies demonstrate that machine learning algorithms can discriminate based on classes like race and gender. This academic study presents an approach to evaluate bias present in automated facial analysis algorithms and datasets.
Algorithmic impact assessments (AIAs) are an emergent form of accountability for organizations that build and deploy automated decision-support systems. This academic paper explores how to co-construct impacts that closely reflects harms, and emphasizes the need for input of various types of expertise and affected communities.
ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT)
This academic article develops a framework for evaluating whether and how automated decision-making welfare systems introduce new harms and burdens for claimants, focusing on an example case from Germany.
This report explores technologies that have the potential to significantly affect employment and job quality in the public sector, the factors that drive choices about which technologies are adopted and how they are implemented, how technology will change the experience of public sector work, and what kinds of interventions can protect against potential downsides of technology use in the public sector. The report categories technologies into five overlapping categories including manual task automation, process automation, automated decision-making systems, integrated data systems, and electronic monitoring.
Little is known about how agencies are currently using AI systems, and little attention has been devoted to how agencies acquire such tools or oversee their use.
This article suggests that a lack of clear, shared definitions makes it harder for the public and policymakers to evaluate and regulate technical systems that may have significant impacts on communities and individuals by shaping access to benefits, opportunities, and liberty. It presents and evaluates a definition for automated decision systems, developed through workshops with interdisciplinary scholars and practitioners.
This academic paper examines predictive optimization, a category of decision-making algorithms that use machine learning (ML) to predict future outcomes of interest about individuals. Through this examination, the authors explore how predictive optimization can raise concerns that make its use illegitimate and challenge claims about predictive optimization's accuracy, efficiency, and fairness.
This quarterly research update aims to highlight key learnings related to improving unemployment insurance (UI) systems in the areas of equity, timeliness, and fraud, and monitor for model UI legislation and policy related specifically to technology. Subscribe to receive future editions.
