Seyfarth Synopsis: On January 23, 2023, New York City’s Department of Consumer and Worker Protection (“DCWP”) held a second public hearing regarding proposed rules implementing Local Law 144 of 2021 (“Local Law 144”) which regulates employers’ use of “automated employment decision tools,” including artificial intelligence, in employment selection decisions. The hearing focused on DCWP’s December 2022 revisions to the set of proposed rules DCWP initially released in September 2022. DCWP has announced a delay in the enforcement of Local Law 144 until April 15, 2023, however, we do not yet have information regarding when the final rules will be released.
On December 23, 2022, the DCWP published highly-anticipated Proposed Rule Amendments to Local Law 144 (the “Amendment”). The Amendment incorporates several improvements and changes based on the comments and testimony submitted by numerous stakeholders, including Seyfarth, during its first public hearing held on November 4, 2022. Seyfarth’s comments on the original proposed rules can be found here.
The following summary addresses several of the key changes proposed in the Amendment, as well as remaining areas that employers will need to consider when implementing Local Law 144.
Amendment to the Applicability of the Rule
As Seyfarth noted in its public comments, the rules needed clarity to confirm that AEDTs applied to individuals who are neither (1) employees nor (2) persons who have “applied for a specific position” are not covered by Local Law 144. Seyfarth recommended striking the undefined terms “individuals” and “applicants” in the initial draft of the rule and replacing them with the term “candidates for employment.” That suggestion was accepted and has been implemented into the Amendment.
Scope of Employment Decisions Impacted by AEDTs Narrowed
While there were improvements to the definition of an AEDT in the Amendment, the scope of coverage remains fairly broad. The Amendment maintains that an AEDT is “any computational process, derived from machine learning, statistical modeling, data analytics, or artificial intelligence, that issues simplified output, including a score, classification, or recommendation, that is used to substantially assist or replace discretionary decision making for making employment decisions that impact natural persons.”
To provide additional clarity and guidance to the employer community, Seyfarth recommended in its public comments that DCWP “identify, with specificity, the ‘computational process[es]’ that are intended to be covered by the law.” In the Amendment, DCWP revised the meaning of the phrase “to substantially assist or replace discretionary decision making,” clarifying that the law only applies to AEDTs that result in employment decisions (1) relying exclusively on a simplified output, (2) where a simplified output is one of several criterion used but outweighs any other factors being considered, and (3) using a simplified output to overrule conclusions derived from other factors, including human decision making. The revisions help clarify that an AEDT that “modifies” conclusions derived from other sources (including human-decision making) would not be covered by Local Law 144’s requirements.
While our view, and that of many employers, is that the scope of Local Law 144 remains broad, several commentators at the January 23 hearing took an opposite view, arguing that the definition of “AEDT” is now too narrow. In particular, several commentators criticized DCWP’s attempt to clarify the phrase “to substantially assist or replace discretionary decision making” as overly restrictive contending that employers seldom or never “rely solely” on AEDTs or use AEDTs to “overrule conclusions derived from other factors including human decision-making.” They suggested that these clarifications be eliminated and the definition of AEDT be expanded to cast a wider net and capture a greater number of employers’ electronic tools, subjecting them to Local Law 144’s audit and notice requirements.
Independent Auditor and Bias Audit Scope, Redefined
The Amendment seeks to clarify that an “independent auditor” may not be employed or have a financial interest in an employer or employment agency that seeks to use or continue to use an AEDT or in a vendor that developed or distributed the AEDT. By updating this definition, it appears employers and vendors alike will be unsuccessful in taking the position that bias audits should be handled internally by either (1) the employer who actually uses, and is thus familiar with the application of the AEDT in their workplace, or (2) the developers of such tools that are best positioned to conduct bias audits. Instead it appears that users of an AEDT in New York City will be required to use entities unaffiliated with the use or development of such technology.
With regard to concerns and the lack of guidance on the scope of the required bias audit, the original draft left it unclear whether such audits could be based on a sample set of data, or whether such data needed to be limited to NYC candidates. The Amendment clarifies that the bias audit can rely upon “test data” if historical data is not available. Unfortunately however, the Amendment makes no attempt to define “test data” (other than stating it is “any data that is not historical data”). The Amendment also fails to provide any insight into whether that test data could be developed using data from outside of NYC. The Amendment does provide that if test data is used, the summary of the bias audit must explain how the test data was generated and obtained. Additionally, the Amendment clarifies that multiple employers using the same AEDT may rely upon the same bias audit so long as they provide historical data, if available, for the independent auditor to consider in a bias audit.
Amendments to the Impact Ratio Methodology
The initial draft of Local Law 144 provided if an AEDT was used to score applicants, a “scoring rate” analysis was required. Several stakeholders including, including Seyfarth, raised concerns with this methodology because it appears to be untested and not readily used in employment circles. Seyfarth encouraged further review and input from labor economists and statisticians familiar with employment issues, who could weigh in with regard to more appropriate and peer reviewed methodologies.
The Amendment includes a modified approach with regard to the average score methodology and replaced the term “average” with “median.” The Amendment also provides further guidance and examples of how to calculate impact bias, and which categories must be analyzed.
While employers may still continue to have challenges with regard to the implementation of Local Law 144, the Amendment provides clearer guidance for employers, vendors and employment agencies on how to comply with the proposed rule. However, some questions remain unanswered, and it is not clear whether they will be resolved by the time enforcement of this law begins in April 2023. In the meantime, employers should assess the tools that may qualify as an AEDT under NYC law and evaluate how the law and the proposed rules, as amended impact their operations.