Fairness and Accountability Design Needs for Algorithmic Support in High-Stakes Public Sector Decision-Making

Explore a conference talk examining the challenges and design needs for fairness and accountability in algorithmic decision-making within high-stakes public sector contexts. Delve into insights from interviews with 27 machine learning practitioners across five OECD countries, uncovering the disconnect between organizational realities and current research in usable, transparent, and discrimination-aware machine learning. Discover potential design opportunities, including tools for tracking concept drift in secondary data sources and building transparency mechanisms for both managers and frontline public service workers. Gain valuable perspectives on ethical challenges and future directions for collaboration in critical applications such as taxation, justice, and child protection.

Introduction
Automating Decisions
Anticipation vs Detection
General Literature
Canonical Problems
Machine Learning Pipeline
Irregular Data
Motivations
Broken Focus
Themes
Political Challenges
Other Issues
Feedback loops
Secondary uses of data
External interactions
Augmentation of outputs
Organisational routines
Application Dependent
Output Dependent
Around Moving Practices
Challenges
Conclusions