What do AI/ML practitioners think about AI/ML bias?

TL;DR

This paper investigates AI/ML practitioners' perceptions of bias, revealing a gap between their understanding and the definitions used by researchers and companies, highlighting the need for better alignment and support.

Contribution

It identifies a discrepancy in bias definitions between practitioners and researchers, emphasizing the importance of aligning understanding to improve bias mitigation efforts.

Findings

Practitioners' understanding of AI/ML bias differs from academic and industry definitions.

Addressing this gap can improve bias mitigation strategies.

Aligning perceptions may enhance development of unbiased AI/ML systems.

Abstract

AI leaders and companies have much to offer to AI/ML practitioners to support them in addressing and mitigating biases in the AI/ML systems they develop. AI/ML practitioners need to receive the necessary resources and support from experts to develop unbiased AI/ML systems. However, our studies have revealed a discrepancy between practitioners' understanding of 'AI/ML bias' and the definitions of tech companies and researchers. This indicates a misalignment that needs addressing. Efforts should be made to match practitioners' understanding of AI/ML bias with the definitions developed by tech companies and researchers. These efforts could yield a significant return on investment by aiding AI/ML practitioners in developing unbiased AI/ML systems.