EEOC, DOJ: Using AI Technology in Employment-Related Decisions May Violate ADA

Employers increasingly rely on computer-based tools to assist them in hiring workers, monitoring worker performance, determining pay or promotions, and establishing terms and conditions of employment. Automatic resume-screening software, hiring software, chatbot software, video interviewing software, analytics software, and employee monitoring and worker management software allow employers to find efficiencies in day-to-day employee management. Software may scan resumes and prioritize the use of certain keywords, rate employees based on their keystrokes, facial expressions or speech patterns, and obtain information about qualifications and cognitive abilities before a hiring manager ever takes a second look.

On May 12, 2022, the U.S. Equal Employment Opportunity Commission (EEOC) and the U.S. Department of Justice (DOJ) issued separate guidance addressing employers’ use of algorithms and artificial intelligence (AI) in employment-related decision-making. Both technical assistance documents focus specifically on how employers’ use of these technologies may adversely impact individuals with disabilities and violate the Americans with Disabilities Act (ADA).

The EEOC’s and the DOJ’s Technical Assistance Documents

The EEOC’s technical assistance document, “The Americans with Disabilities Act and the Use of Software, Algorithms, and Artificial Intelligence to Assess Job Applicants and Employees,” largely cautions that employers’ use of algorithmic decision-making tools without regard for reasonable accommodations may quickly run afoul of the ADA.

Similarly, the DOJ’s technical assistance document, “Algorithms, Artificial Intelligence, and Disability Discrimination in Hiring,” identifies ways that state and local government employers might use hiring technologies that could violate the ADA.

A. Providing Accommodations During the Assessment Process

As a foundational matter, algorithmic decision-making tools are designed to predict how a typical applicant or employee would perform in typical working conditions. But applicants and employees with disabilities may not present as “typical” — and may not perform in the workplace under “typical” conditions. Employers, therefore, must (1) provide reasonable accommodations during any automated assessment process; and (2) be cautious of AI that screens out applicants with disabilities, even if inadvertently. And in all cases, employers must avoid assessment tools that pose disability-related inquiries or constitute medical examinations under the ADA before a conditional offer of employment.

To illustrate its concerns, the EEOC and the DOJ offered multiple examples of potential assessment accessibility issues, including:

• A job applicant with limited manual dexterity because of a disability may have difficulty taking a knowledge test that requires use of a keyboard, trackpad or other manual input device, particularly if such assessment is timed. The applicant may score poorly on the knowledge test, despite having very good knowledge.
• Some employers rely on “gamified tests” — which use video games to measure abilities, personality traits and other qualities — to assess applicants and employees. If an employer required a 90% score on a gamified assessment of memory, an applicant who is blind and cannot play the game would be rejected, despite potentially having a very good memory.
• A city government that uses an online interview program that does not work with an applicant’s computer screen-reader program may prevent blind applicants from access to the interview process despite being otherwise qualified for the position and able to perform the essential functions of the job.

In these cases, the employer would be required to provide an accessible version of the assessment (i.e., one where answers are oral or a screen reader is used) as a reasonable accommodation, unless doing so would cause undue hardship. If it is not possible to make existing assessments accessible, alternative assessments must be offered, barring undue hardship.

B. Avoiding Impermissible Screen Outs

“Screen out” occurs when a disability prevents an applicant or employee from meeting — or lowers their performance on — selection criteria, resulting in the loss of a job opportunity. This is unlawful if the individual who is screened out is able to perform the essential functions of the job with or without a reasonable accommodation. The EEOC and the DOJ provide examples of how algorithmic decision-making tools may unlawfully screen out individuals on the basis of a disability, though the individual may be able to perform the essential functions of the job with or without a reasonable accommodation:

• Chatbot software designed to engage applicants online through texts and emails may be programmed with an algorithm that rejects all applicants who indicate that they have a significant gap in their employment history. If an applicant had a gap caused by a disability (e.g., a time in which the applicant was undergoing treatment), this may be an impermissible screen out.
• Video interviewing software may analyze applicants’ speech patterns to reach conclusions about their abilities to solve problems. Applicants with speech impediments may achieve a low or unacceptable rating, resulting in an unlawful screen out.
• Similarly, employers that use facial analysis technologies to evaluate an applicant’s skills and abilities may screen out people with disabilities, like autism, even if they are qualified for the job.
• Some applicants or employees with disabilities may be entitled to on-the-job accommodations to assist with avoiding distractions, including a quiet workstation or noise-canceling headphones. Nonetheless, that candidate may be screened out by personality assessments or gamified tests designed to look for employees with a strong ability to ignore distractions.
• Some personality assessments may inquire whether applicants are described as possessing certain traits, including “general optimism” that may not be compatible with disabilities like major depressive disorder (MDD). An applicant with MDD may score poorly on this assessment and be screened out, despite having an ability to perform the essential functions of the job with or without a reasonable accommodation.

C. Avoiding Impermissible Inquiries

Both the EEOC and the DOJ also caution employers against the use of algorithmic decision-making tools that pose “disability-related inquiries.” These are questions likely to elicit information about a disability, including simply asking whether an applicant or employee is an individual with a disability. They also include “medical examination” questions that seek information about an individual’s physical or mental impairments or health prior to a conditional offer of employment.

Best Practices Going Forward

Finally, the EEOC and the DOJ highlight best practices for employers to comply with the ADA when using algorithmic decision-making tools. Specifically, the EEOC and DOJ encourage employers to:

• Train staff giving assessments on accessibility and reasonable accommodations.
• Use tools that have been designed to be accessible to individuals with as many disabilities as possible. Keep in mind that even though one technology may not discriminate against a blind applicant, for example, it may still discriminate against an applicant with a different disability, like epilepsy.
• Inform all applicants and employees that reasonable accommodations are available and provide clear instructions for requesting accommodations.
• Describe, in plain language, what the assessment is designed to assess, the method being used, and the variables or factors that may affect the rating.
• Ensure that tools only measure abilities or qualifications truly necessary for the job. For example, make sure that your assessment is not really measuring predominantly an applicant’s ability to be able to see a screen if the job in question can be performed with an accommodation for a visual impairment.
• Ensure that necessary abilities or qualifications are measured directly, rather than by way of characteristics or scores that are correlated with those abilities or qualifications.

What’s Next?

The EEOC’s and DOJ’s technical assistance documents are among the first published guidance documents of any federal agency related to employers’ use of AI. The documents are not surprising, though. The EEOC has been interested in the topic since at least October 2016, when it held a meeting on “Big Data in the Workplace.” More recently, in December 2020, several senators wrote an open letter to the EEOC to request information about its oversight of hiring technologies. Among other questions, they asked if the EEOC planned to release any publications or provide any guidance on the use of data and technology in hiring. And in October 2021, the EEOC announced that it had launched an initiative on artificial intelligence and algorithmic fairness. So, while the EEOC’s and the DOJ’s technical assistance documents are among the first, they won’t be the last. Employers should expect that the EEOC and the DOJ will continue to expand their research and regulation of employers’ use of AI and how it implicates both the ADA and other federal anti-discrimination laws.