Seyfarth Synopsis: The New York Department of Consumer and Occupational Safety (“DCWP”) today released the much-anticipated final regulations implementing New York Local Law 144 of 2021 (“Local Law 144”), which governs the use of automated employment decision aids ( “AEDT”). While Local Law 144 went into effect on January 1, 2023, in a submission email accompanying the final rules, DCWP stated that the rules will go into effect on July 5, 2023 and enforcement will begin.
As background, DCWP first published a set of proposed rules to implement local Law 144 on September 23, 2022 and held a public hearing on November 4, 2022. Employers, employment agencies, law firms, AEDT developers and advocacy groups published DCWP revised on December 23, 2022 proposed rules. Following a January 23, 2023 public hearing on the revised proposed rules, the final rules have now been issued, governing the use of AEDTs in NYC.
According to DCWP, the changes to the final rules are:
Change the definition of “machine learning, statistical modelling, data analysis or artificial intelligence” to broaden its scope; Add a requirement that bias audits report the number of people who assessed an AEDT who were not included in the calculations because race/ethnicity and sex data are unknown, and require that number be included in the summary of results becomes; allow an independent auditor to exclude from the impact rate calculations a category comprising less than 2% of the data used for the bias audit; Clarify examples of a bias audit and when an employer or employment agency can rely on a bias audit conducted using the historical data of other employers or employment agencies; Give examples of when an employer or employment agency can rely on a bias audit conducted using historical data, test data, or historical data from other employers and employment agencies; and Clarify that the number of applicants in a category and the rating score of a category, if applicable, must be included in the results summary.
These specific changes are discussed in more detail below.
Machine learning, statistical modelling, data analysis or artificial intelligence
The Final Rules modified the definition of “machine learning, statistical modelling, data analysis, or artificial intelligence” to mean a group of mathematical, computational techniques that:
make a prediction, i.e. an expected outcome for an observation, e.g. an assessment of a candidate’s suitability or likelihood of success, or making a classification, ie assigning an observation to a group, e.g. B. categorizations based on ability or suitability; and for which a computer identifies, at least in part, the inputs, the relative importance attached to those inputs, and other parameters, if any, for the models to improve the accuracy of the prediction or classification.
The definition eliminates the previous requirement that, in order for a mathematical computational technique to be considered “machine learning, statistical modeling, data analysis, or artificial intelligence,” it must also allow for the refinement of inputs and parameters through cross-validation or through training and test data.
Bias audits, disclosure and data requirements
The final rules clarify that bias audits are required even when an AEDT is not used in conjunction with a final hiring decision. Rather, bias audits are also required when used for “screening” early in the hiring process.
Further clarifications on the information to be disclosed in relation to the required bias audit have also been provided. Specifically, NYC employers are required to disclose a summary of the results of their bias audit, which must include (1) “the source and explanation” of the data used to conduct the bias audit, (2) the date of the last audit , (3) the number of applicants or candidates, (4) the selection or evaluation rates, (5) the impact ratio by race/ethnicity, sex, and intersectional categories, and (6) the number of individuals with unknown race/ethnicity and sex that were evaluated by the tool. To conform to the “unknown” disclosure, the following example was provided: “The AEDT was also used to score 250 individuals with an unknown gender/ethnicity category. Data on these individuals has not been included in the above calculations.”
When conducting a bias audit, independent auditors may exclude a race/ethnicity or gender category that accounts for less than 2% of the data used from the required impact ratio calculations. If a category is excluded, the summary of results must state the “reason for the exclusion and the number of applicants and the evaluation rate or selection rate for the excluded category”.
It also further explains when historical data or test data can be used when conducting a bias audit. As a general rule, a bias audit must be based on historical data. However, a bias audit may only rely on historical data from other employers or employment agencies if (1) that employer or employment agency has provided historical data from its own use of the AEDT to the independent auditor conducting the bias audit , or (2) if such employer or employment agency has never used the AEDT in question.
In cases where insufficient historical data is available, an employer or employment agency may rely on a bias audit that relies on test data to conduct a statistically significant bias audit. If test data is used, the summary of the results of the bias audit shall explain why historical data was not used and describe how the test data used was generated and obtained.
enforcement period extended
In an attempt to give employers and employment agencies more time to comply with the final regulations of Local Law 144, DCWP announced it will delay enforcement of the final regulations until July 5, 2023. Because this three-month delay doesn’t give employers much time to comply, those affected by this law must carefully evaluate the tools it uses, which may qualify as AEDT under NYC law, and how the final ones will affect Rules impact their operations and what next steps need to be taken to mitigate non-compliance risks.
workplace solutions
We encourage you to contact the authors of this article or a member of Seyfarth’s People Analytics team as soon as possible if your organization needs assistance in complying with the final Local Law 144 regulations.
