Bias in AI Hiring: What the Evidence Shows

What the evidence says

• ✦Manual resume screening has been shown in a field experiment to favor resumes with names associated with white applicants over otherwise comparable resumes with names associated with Black applicants. [1]
• ✦Automated speech recognition systems from several major providers were found to make more errors for Black speakers than for white speakers. [2]
• ✦Commercial facial-analysis systems have shown higher error rates for some demographic groups, and NIST found demographic differentials in many face recognition algorithms. [3][4]
• ✦US federal guidelines use the "four-fifths rule" as a practical rule of thumb for identifying adverse impact in selection procedures. [5]
• ✦Many vendors of algorithmic hiring tools disclose little about how they test for and mitigate bias. [6]

AuraSync's interpretation

Because the components of multimodal assessment, such as transcription and face verification, have documented disparities, we do not describe AuraSync as bias-free. We aim for consistency and reviewability: one structured method for every candidate, evidence attached to every result, adjustments on request and a person making the decision.

What AuraSync claims

• ✦AuraSync does not describe itself as bias-free.
• ✦Results can be affected by audio and video quality, accent, connection, device, disability or assistive technology, and candidates can request adjustments and human review.
• ✦Customers remain responsible for monitoring their own hiring outcomes.

Limitations

The studies above measured specific systems at specific times; error patterns change as technology changes. None evaluated AuraSync. Consistent methods reduce some kinds of bias but do not guarantee fair outcomes.

Sources

1. [1]Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination. American Economic Review, 94(4), 991–1013. https://doi.org/10.1257/0002828042002561
2. [2]Koenecke, A., Nam, A., Lake, E., Nudell, J., Quartey, M., Mengesha, Z., Toups, C., Rickford, J. R., Jurafsky, D., & Goel, S. (2020). Racial disparities in automated speech recognition. Proceedings of the National Academy of Sciences, 117(14), 7684–7689. https://doi.org/10.1073/pnas.1915768117
3. [3]Buolamwini, J., & Gebru, T. (2018). Gender shades: Intersectional accuracy disparities in commercial gender classification. Proceedings of Machine Learning Research, 81, 77–91. https://proceedings.mlr.press/v81/buolamwini18a.html
4. [4]Grother, P., Ngan, M., & Hanaoka, K. (2019). Face Recognition Vendor Test (FRVT) Part 3: Demographic Effects (NISTIR 8280). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.IR.8280
5. [5]Uniform Guidelines on Employee Selection Procedures (1978), 29 C.F.R. Part 1607. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1607
6. [6]Raghavan, M., Barocas, S., Kleinberg, J., & Levy, K. (2020). Mitigating bias in algorithmic hiring: Evaluating claims and practices. Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 469–481. https://doi.org/10.1145/3351095.3372828

This page summarizes third-party research for orientation. It is not an evaluation of AuraSync, and any summary of law is not legal advice.