Bias Audits of Automated Employment Decision Tools and AI

Growing Regulatory Requirements

While automated employment decision tools (AEDT) and artificial intelligence (AI) in selection systems are already subject to existing anti-discrimination laws and regulatory frameworks like the Uniform Guidelines on Employment Selection Procedures (UGESP), an increasing number of cities and states are implementing requirements specifically regarding these types of tools. New York City, California, Colorado, and Texas have passed laws affecting employers' use of AEDTs and AI-powered technology, with new regulations and requirements on the horizon. Many of these laws require or will require employers to conduct statistically-based bias audits to ensure fairness and non-discrimination.

How DCI Can Help

Employers must comply with a patchwork of laws regulating the use of AI systems and DCI can help your organization determine how these laws apply to the tools you are using, comply with analytical requirements of these laws, and design custom analyses when needed. Our experts have in-depth knowledge of UGESP, relevant state and local laws, the statistical nuances of conducting adverse impact analyses, and the ins-and-outs of developing, implementing, and validating selection systems and assessments.

Why DCI?

DCI brings 25 years of experience conducting high-stakes adverse impact analyses of assessment results across various proactive and legal scenarios. Our expertise with both traditional assessments and tools powered by artificial intelligence enables DCI to guide your organization towards compliance with bias audit laws while providing the same white glove service for which our consultants are known. We take the time to properly structure your data, explain your results, and prioritize your next steps. Lawsuits, fines, and negative press all affect the bottom line. Partnering with us can save your organization money and protect its reputation - all while ensuring your selection practices are equitable and legally compliant.

Expert I/O (Qualitative) Review

DCI performs a deep-dive review of the algorithm development and scoring processes, fairness and bias analysis evidence, job-relevance (validation) information, and process-oriented documentation
DCI evaluates the selection procedure and associated algorithm(s) against DCI’s Customizable AI Audit Framework to identify areas of risk/concern and to provide recommendations and strategies to mitigate areas of higher risk/concern

Algorithm Feature Review: Scoring Relevance and Logic

DCI evaluates specific features of an algorithm one-by-one to determine if they are: clearly related to job performance, scored in a logically explainable manner, scored in a manner that may produce inadvertent demographic group differences

Operational Implementation Review

DCI evaluates guardrails and standard operating procedures that protect against improper tool use and identifies opportunities for improvement

Data Inputs Review

DCI evaluates the quality and job-relatedness of algorithm inputs (e.g., job postings) to identify concerns, opportunities for improvement, and alignment with recommended practices for describing job requirements

Quantitative Analyses: Contemporary Adverse Impact

DCI evaluates the existence and magnitude of differences in scores by demographic group (e.g., gender, race) for each job

NYC Local Law 144 Bias Audit Analyses

DCI produces a bias audit report that follows the specific requirements of NYC Local Law 144

Validation Research: Expert Algorithm Alignment Study

DCI evaluates the extent to which the inputs to an algorithm are producing outputs that are consistent with independent, expert expectations
For resume-job match algorithms, DCI uses expert judgment approaches to evaluate job requirements, candidate qualifications, and degree of match for comparing against algorithm outputs

Validation Research: Criterion Study

DCI designs and conducts a criterion validation study that evaluates the relationship between scores on the selection procedure and job-related outcomes (e.g., turnover, job performance) for individuals at the organization