New Report: Disparate-Impact Reform in The Age of AI

 The legal doctrine should be limited but apply equally to algorithmic decision making 

NEW YORK, NY – For years disparate-impact liability helped drive hiring decisions. According to this legal doctrine, employers can face liability for disparities in hiring outcomes across sexes, races, or other characteristics, even if there is no intent to discriminate. But this year the Trump administration’s Office of Legal Counsel issued a memo sharply limiting federal agencies’ use of disparate-impact liability.  
 
In a new Manhattan Institute report, senior fellow Robert VerBruggen considers the state of disparate-impact liability by examining the doctrine’s development over the last 50 years and how AI is changing the landscape. Businesses are increasingly relying on algorithms and artificial intelligence to screen applicants, which presents an opportunity to eliminate discrimination in hiring. However, a look under the hood reveals complex models can be unfairly biased. For example, previous research has uncovered bias both in favor of and against women in the decision making of algorithms and artificial intelligence programs. 
 
VerBruggen suggests that algorithms used in employment decisions should be auditable and explainable, offering human-understandable assessments of specific job-related characteristics. He proposes that disparate-impact reforms should explicitly clarify that algorithms may not do what Americans have long been prohibited from doing under antidiscrimination law, while spelling out a narrow, fair test for determining violations. A person challenging decisions made by an algorithm should be able to show either (A) that it relies on factors that have no reasonable relationship to the job; or (B) that it warps specific job-related factors into proxies for protected characteristics. 

Click here to read the full report.