Small businesses that use LinkedIn, Indeed or any online platform to recruit for positions must be wary of the ways in which they can easily pull themselves into the requirements of New York City’s new rules related to AI in hiring. This is possible because:
- online services offer multiple free supports to identify the best candidates;
- the city’s rules define AI in hiring processes broadly enough to include those free, helpful supports; and
- the city’s rules don’t just apply to worksites in New York City.
My own recent job posting experience was an eye-opening opportunity to see and reflect on the traps and pitfalls, and how I might help my clients solve for them going forward.
Just a Little Help Easily Ensnares the Unwary
LinkedIn wanted to help me. I had posted an open position and it was eager to help me identify candidates who would be the best match. As I composed my posting, it offered for me to include three screening questions – anyone who answered incorrectly or whose credentials did not align would be screened out. “No thank you,” I said, and I bypassed that section. But LinkedIn was not done. As the first stream of applicants filled my inbox, LinkedIn offered another automated tool – if I told it which candidates had potential and which did not, it would screen for those preferences going forward. “No thank you,” I said again.
How NYC Defines AI in Hiring
With each step in my online job posting process, the definition of an AEDT – an automated, electronic decision tool under New York City’s new regulatory framework for AI hiring processes – was resonating in my head. The city defines an AEDT as a tool used “to substantially assist or replace discretionary decision making,” and its new regulations interpret that to include:
- when an employer relies solely on a score, tag, ranking, or classification generated by the tool (a “simplified output”) to select or eliminate candidates, with no other factors considered;
- when an employer looks at factors besides the simplified output, but gives that simplified output more weight than any other criterion considered; or
- when the employer uses the simplified output to overrule conclusions made based on other factors.
Would accepting LinkedIn’s help automatically screen out job applicants without any human consideration? If the tool worked as LinkedIn seemed to suggest, then my conclusion was “yes,” that the algorithms I could deploy to control my job posting inbox would most certainly exclude some subset of applicants. Accepting the free help, I feared, would land me squarely within the realm of the AEDT regulations.
Implications for Employers Found to Use AI
To be clear, New York City is not banning employers from using AI or any form of an AEDT. But the City has clearly declared it is wary of how those tools are deployed, and how rapidly and effectively they can generate biased hiring decisions. Therefore, beginning July 5, 2023, employers who use an AEDT in their hiring process will need to ensure it has undergone a bias audit conducted by an independent third-party, and post the results of that audit and related information on their websites.
That sounds like a bit of a hassle but innocuous enough – some new certification that service providers will need to add to their AI product, and yet one more blurb I will need to fit onto my website in an accessible place. Except for two problems. In the broader sense, AI tool providers are not currently are set up to offer this type of certification, nor can they do it themselves, as the law clearly requires it to be conducted by an independent third-party – not the tool creator, and not the tool user (meaning, not the employer). There is an entire cottage industry that will need to sprout up to meet this new need, and the likelihood that it will be in place just two short months from now is not so great.
The second problem comes back to my experience with the LinkedIn job posting. Had the screening tool I was using been tested for bias? Not that I could discern, but more importantly, I do not see how it really could be, as the tool would have been responding to my unique and subjective data inputs, which it could not have predicted and tested for in advance of me posting this position. Reflecting on the purpose of the AEDT regulations, my subjective screening questions could most certainly have automated a biased review process, if I had a proclivity to take things in that direction.
For the small businesses like mine that leverage LinkedIn and other online platforms to recruit candidates, it seems the only option at present to comply with the New York City law is to decline any tools that can screen out candidates. Even if free, those tools can create a large headache for the unwary employer.
One week after placing my job posting on LinkedIn, the applications have just filed in to my online inbox, where reviewing them will be my weekend project. LinkedIn has another tool available for me to use with those that have arrived – allowing me to sort and filter by relevance, rating, location and years of experience. “Yes, please,” was my reaction. Filtering allows me continued access to all the data on all the applicants, but it enables me to manage my analysis of the applications in smaller, more relevant batches. That type of help I will gladly accept, as it falls outside the city’s regulations.
Employers Outside NYC Are Also Covered
I am based in Westchester County, outside of New York City. Did I need to consider complying with New York City law? Unfortunately, yes, as I had posted a hybrid position, one that could be performed remotely at least some portion of the week. Given the commuting distance, it is entirely feasible that the remote portion would be performed by someone residing in one of New York City’s five boroughs.
The AEDT in hiring rules apply whenever an employer screens candidates for employment or employees for promotion “within the city” of New York, and other portions of the law expressly extend application of the rules to include those who reside in the city. The saving grace for me was that this search was being conducted in May 2023, and not in July when New York City begins to enforce its new hiring rules. But it was a stark lesson, from a small business owner trying to manage her workforce, as to how pervasive AI already is in our hiring processes and how broad New York City’s new regulations may reach.
By Tracey I. Levy