I keep thinking about orchestras, where simply auditioning performers behind a curtain completely fixes the bias problem.
Of course the trick is that you don't need to see the candidates or talk to them, just listen to their playing. In software, we would need to find some similarly effective way to measure anonymized performance.
In fact, completely aside from fixing gender and racial biases, that's something we could really use just to make good hiring decisions! I don't believe anyone really knows how to make consistently great hires in software.
For a start, the hiring decision could be based on gender-anonymized feedback from the interviewer(s), although that obviously wouldn't fix any underlying biases in the feedback itself.
There appear to be problems with this approach that wouldn't be acceptable to the people driving the diversity effort [0]. To summarize the linked example: ElectronConf tried a gender-blind selection process for speakers, and when they lifted the veil on their selections, discovered they had only selected men to speak, so they canceled the conference.
I believe interviewing.io did something with voice-pitch-adjusting (to make girls sound like guys, or vice-versa) in phone interviews to try to study exactly this effect.
I think their results were confusing and uncertain, but the methodology seemed brilliant and I think should be the gold-standard for tech interviews. If we keep it phone it would remove other subliminal biases (attractiveness, physical disabilities) as an additional benefit.
Of course the trick is that you don't need to see the candidates or talk to them, just listen to their playing. In software, we would need to find some similarly effective way to measure anonymized performance.
In fact, completely aside from fixing gender and racial biases, that's something we could really use just to make good hiring decisions! I don't believe anyone really knows how to make consistently great hires in software.
For a start, the hiring decision could be based on gender-anonymized feedback from the interviewer(s), although that obviously wouldn't fix any underlying biases in the feedback itself.