In NYC, companies will have to prove their AI hiring software isn't sexist or racist

It's like 10-15 years ago suddenly all the companies were claiming they used big data. Unfortunately it's just buzz words to entice investors or lazy reporting.

I think part of this is basically asking companies to keep a record of what information is used.

Hey, I am a machine learning engineer that works with people data. Generally you measure bias in the training data, the validation sets, and the outcomes ( in an ongoing fashion - AIF 360 is a common library and approach ). There are lots of ways to measure bias and or fairness. Just checking if a feature was used isn’t considered “enough” by any standards or practitioner. There are also ways to detect and mitigate some of the proxy relationships you’re pointing to. That being said, I am 100% skeptical that any hiring algorithm isn’t going to be extremely bias. A lot of big companies have tried and quit because despite using all the right steps the models were still bias https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G. Also many of the metrics used to report fairness have some deep flaws ( disprate impact ).

All that being said the current state is that there are no requirements for reporting so vendors don’t do the minimum 90% of the time because if they did it would cost a lot more and get in the way of the “AI will solve all your problems with no effort” narrative they want to put forward so I am happy to see any regulation coming into place even if it won’t be perfect.

AI resume screeners are very much at risk of bias. There have been stories about exactly this in years past. The ML models need to be trained, so they get fed resumes of candidates that were hired and not hired so the model can learn to differentiate the two and make decisions on new resumes in the future. That training, though, takes any bias that went into previous decisions and brings it forward.

From the Amazon I linked above, the model was prioritizing white men over women and people of color. When you think back to how these models were trained, though, that's exactly what you'd expect to happen. No one was intentionally introducing bias to the AI process, but software teams have historically been very male and white, and when referrals and references come into play, those demographics were further emphasized. And then let's not pretend that none of those recruiters or hiring managers were bringing their own bias to the table.

If you feed that into your model as it's training data, of course the model is going to continue to favor white men, not because it's actually looking for men, but because resumes that men typically submit are the kinds that get hired. Then they found that resumes that mention a professional women's organization or historically black or women only colleges were typically not hired. The model isn't "thinking" about why that is - it just knows that when certain traits exist, the resume is ranked lower, so it replicates that.

Building a truly unbiased AI system is actually incredibly difficult, not the least due to the fact that the demographics of the data scientists working on these systems are themselves predominantly male and white themselves. We've also seen this issue in the past with other AI systems, including facial recognition systems, where these systems built by teams of white men can't seem to make reliable determinations when looking at a picture of a black woman (with accuracy rates 20-30% lower for black woman compared to white men).

It depends how “bias” has been defined. The Ibram Kendi definition is unequal outcomes. Since no two groups are identical, such definitions require bias to be “unbiased.” Australia tried to employ blind recruitment and hired fewer women and minorities. That’s true unbiased recruitment, but I suspect it wouldn’t be praised today.

To your last point.

Use every single buzzword in any job listing on your resume. DO NOT CONJUGATE. Their "AI" is looking for "Rapid Analysis" not "Analyzes rapidly."

Yes, but if you train an AI on racist/sexist data, it will naturally do the same.