What is it about?
Machine learning systems are used in more and more areas, and many people are worried about bias and unfair decisions. Fairness toolkits help developers find and reduce bias in their models, but we know little about why practitioners decide to use them. In this study, we looked at the factors that lead software practitioners to adopt fairness toolkits. We used a well-known model of technology acceptance (UTAUT2), collected survey data from practitioners in industry, and analyzed it with a statistical technique (PLS-SEM).
Featured Image
Photo by Luke Hales on Unsplash
Why is it important?
We found that two factors matter most: performance expectancy—the belief that the toolkit helps to reduce bias effectively—and habit. For organizations, this means showing clearly how well these tools work and making their use a routine. Practical steps include improving the usability of the toolkits, adding bias checks to everyday development workflows, and giving ongoing support so that professionals see clear benefits from regular use.
Perspectives
Fairness toolkits exist, but having a tool is not the same as using it. In this study, we used the same model I used for LLM adoption (UTAUT2) to understand what makes practitioners actually use these toolkits. The answer—performance expectancy and habit—suggests that fairness checks work best when they become part of the daily routine, not an extra step.
Prof. Stefano Lambiase
Aalborg Universitet
Read the Original
This page is a summary of: From Expectation to Habit: Why do Software Practitioners Adopt Fairness Toolkits?, April 2025, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/icse-seis66351.2025.00015.
You can read the full text:
Contributors
The following have contributed to this page
