# Exploring How Machine Learning Practitioners (Try To) Use Fairness Toolkits

**Authors:** Wesley Hanwen Deng, Manish Nagireddy, Michelle Seng Ah Lee, Jatinder Singh, Zhiwei Steven Wu, Kenneth Holstein, Haiyi Zhu  
**arXiv:** [2205.06922](https://arxiv.org/abs/2205.06922)  
**Date:** 2022-05-13  
**Summary page:** https://www.summarizepaper.com/en/paper/2205.06922v2/

## Key points

- Open-source machine learning (ML) fairness toolkits have been developed to help practitioners address unfairness in their systems.
- Little research has been conducted on how these toolkits are used in practice.
- A study was conducted with 11 industry practitioners who were tasked with building a model to determine which students needed additional tutoring resources using the Student Performance dataset.
- The study identified opportunities for fairness toolkits to better address practitioner needs and scaffold them in using toolkits effectively and responsibly.
- External pressure from stakeholders was identified as an important factor that drives practitioners' engagement with fairness issues.
- Future toolkits should consider incorporating features that enable practitioners to communicate effectively with stakeholders about fairness concerns.
- The study provides valuable insights into how industry practitioners use existing ML fairness toolkits in practice and highlights areas where improvements can be made.
- The Colab notebook used in the study is available for others conducting relevant evaluations.

## Summary

In recent years, there has been a surge in the development of open-source machine learning (ML) fairness toolkits aimed at helping practitioners assess and address unfairness in their systems. However, little research has been conducted to investigate how these toolkits are used in practice. To bridge this gap, a study was conducted to explore how industry practitioners work with existing fairness toolkits. The study involved 11 participants who completed all phases of the research, including a pre-interview task and a 60-minute think-aloud semi-structured interview. The participants were tasked with building a model to determine which students were in need of additional tutoring resources using the Student Performance dataset, which includes student grades as well as demographic, social, and school-related features. The aim was to observe participants' thought processes during the exploratory data analysis (EDA) and problem formulation stages. The study identified several opportunities for fairness toolkits to better address practitioner needs and scaffold them in using toolkits effectively and responsibly. The findings highlight implications for future open-source fairness toolkits that can support practitioners in better contextualizing, communicating, and collaborating around ML fairness efforts. For instance, external pressure from stakeholders was identified as an important factor that drives practitioners' engagement with fairness issues. Therefore, future toolkits should consider incorporating features that enable practitioners to communicate effectively with stakeholders about fairness concerns. Overall, this study provides valuable insights into how industry practitioners use existing ML fairness toolkits in practice and highlights areas where improvements can be made to better support them in addressing unfairness issues. The Colab notebook used in the study is also available for others conducting relevant evaluations.