On the Fairness of Machine-Assisted Human Decisions: Formal Model and Lab Experiment

A study by Bryce McLaughlin of Columbia University investigates fairness in machine-assisted human decision systems. The work includes a formal statistical model and a lab experiment examining how properties of machine predictions affect human decisions. The findings suggest that excluding protected group information from predictions may fail to reduce, or even increase, ultimate disparities.

Use Cases

• Analyzing the impact of biased human decision-makers on algorithmic fairness based on the formal model described.
• Studying how gender-specific information in predictions affects gender disparities based on the lab experiment.
• Investigating the interaction between algorithm structure and human decision quality as discussed in the theoretical results.

Strengths

• Combines a formal theoretical model with an empirical lab experiment.
• Focuses on a high-stakes, contemporary problem in machine learning deployment.
• Authored by a researcher from Columbia University, a reputable institution.

Limitations

• Theoretical results rely on specific assumptions about data, algorithm, and decision-maker.
• The lab experiment focuses on a single, particular prediction task.
• Row count, column details, and sample data are unavailable for assessment.

Provenance

Source

paperswithcode

Collection Method

Likely contains data from a formal model and a controlled lab experiment.