The Equal Employment Opportunity Commission (EEOC) recently released guidance to help private sector employers avoid disability discrimination when using algorithms to assess employees and applicants. The guidance explains the ways in which software that relies on algorithmic decision-making can run afoul of Americans with Disabilities Act (ADA) requirements. The guidance also includes “promising practices” to help employers avoid ADA violations when using algorithms and artificial intelligence (AI) tools in the employment selection process.
In October 2021, the EEOC announced an initiative focused on ensuring that employers using AI and algorithmic tools comply with federal civil rights law.
Title I of the ADA prohibits employers, employment agencies, labor organizations and joint labor-management committees with at least 15 employees from discriminating on the basis of disability. Title I also imposes an affirmative obligation on employers to provide reasonable accommodations to applicants and employees during the hiring and promotion process.
In the new guidance, the EEOC explains employers’ responsibilities when using computer-based and online assessments, including those administered by third parties, to ensure that the assessments are not resulting in discrimination against persons with disabilities. The EEOC also clarifies in the new guidance that algorithmic decision-making may still violate the ADA even though an assessment has been properly validated under the Uniform Guidelines on Employee Selection Procedures (UGESP), which applies only to compliance with of the Civil Rights Act (Title VII) and not the ADA.
What Kinds of Assessments are Covered by the Guidance?
The EEOC’s guidance covers software, algorithms and AI when used in connection with making employment decisions. “Software” broadly includes programs and applications that perform a given task or function. “Algorithms” means a type of software that processes data and evaluates, rates and makes employment decisions by applying a set of instructions. “AI,” which is also referred to as machine learning, develops algorithms based on the computer’s own analysis of data to make predictions about what will make an applicant a successful employee, and then decides the criteria that should be used to assess applicants and employees based upon those predictions. Examples of the kinds of assessments and tools covered by the EEOC’s guidance include:
- Screening applications with resume scanning software that prioritizes certain keywords.
- Online interviews with virtual assistants or “chatbots” that screen for pre-determined candidate responses.
- Computerized tests that measure applicants’ abilities, personalities, traits or characteristics, including through the use of games.
- Video interviewing that evaluates candidates based on their facial expressions or speech patterns.
How can Algorithmic Decision-Making Violate the ADA?
The guidance explains three common ways that algorithmic decision-making might violate the ADA: (i) failing to provide a reasonable accommodation in the assessment process that would have allowed a job applicant or employee to be evaluated accurately and fairly; (ii) loss of a job opportunity because an individual with disability, who is able to do the job with a reasonable accommodation, is screened out by the assessment; and (iii) including questions in an assessment that are considered disability-related inquiries or medical exams under circumstances prohibited by the ADA.
Reasonable Accommodation and Algorithmic Decision-Making Tools
As in other contexts, the EEOC’s guidance explains that employers must provide reasonable accommodations when a medical condition may impact an applicant’s or employee’s performance on an assessment. Reasonable accommodations in this context could include extended time to take the assessment, or an alternative version of the assessment that is compatible with accessible technology. The guidance gives as an example: an applicant with limited manual dexterity may have difficulty with a knowledge test that requires the use of a keyboard, trackpad or other manual input device that would not accurately measure that applicant’s knowledge. An employer might reasonably accommodate the applicant by allowing them to provide responses orally, rather than manually.
For non-obvious disabilities, employers may request supporting medical documentation, and if the documentation shows that a disability “might make the test more difficult to take” or “might reduce the accuracy of the assessment,” the guidance states that employers must provide an alternative means of assessing the candidate absent a showing of undue hardship (defined as involving significant difficulty or expense). The guidance also confirms that the reasonable accommodation obligation extends not only to employers, but also to third-party administrators acting on the employers’ behalf.
As a “promising practice,” the EEOC suggests that employers using algorithmic decision-making tools inform applicants up front how they will be evaluated and that reasonable accommodations are available for persons with disabilities, including clear and accessible instructions on how to request an accommodation. The guidance specifically recommends that employers tell candidates, in plain language, the traits that are being evaluated by an assessment, the method of assessment and any variables or factors that may affect an applicant’s score. While the EEOC’s guidance recommends this level of transparency to candidates, the ADA does not require it. New York City, however, recently enacted a law that will take effect in 2023 that will require such disclosures when employers use algorithms and AI in the selection process.
Loss of Job Opportunity due to Algorithmic Decision-Making Tools
The ADA may be violated when a disabled candidate who is capable of doing the job with or without an accommodation is “screened out” from consideration because their disability prevents them from meeting a selection criteria or performing well on the assessment. As an example, the EEOC explains that a person’s disability may prevent the algorithmic decision-making tool from measuring what it is intended to measure, such as a candidate with a speech impediment being assessed by a speech pattern tool. If the candidate lost a job opportunity due to a poor score on the assessment, then the candidate may have effectively been screened out because of the speech impediment and not the candidate’s ability to do the job.
The new guidance also distinguishes ADA obligations when using assessments from Title VII’s obligations. Under Title VII’s disparate impact provisions, when an assessment has a disproportionate negative impact on a particular gender, race or ethnicity, an employer must prove the tests is job-related and consistent with business necessity. Employers can establish job-relatedness and business necessity under Title VII by “validating” a test in accordance with the UGESP. However, the EEOC explains that the UGESP do not apply under the ADA, and even a tool that has been properly validated may still be inaccurate when applied to a particular individual with a disability. And although the ADA also requires employers to establish job relatedness and business necessity to justify using an assessment that tends to screen out individuals with disabilities, the EEOC notes that each disability is unique and different steps may be required to make this showing, beyond those taken to address other forms of discrimination.
The new guidance specifically addresses personality tests, which have become increasingly popular among employers because such assessments have been found to correlate with successful job performance while resulting in far less adverse impact under Title VII than traditional cognitive tests. Under the ADA, however, individuals with disabilities such as post-traumatic stress disorder (PTSD) may perform poorly on personality assessments despite being able to perform the job successfully, even sometimes without any accommodation. The EEOC suggests that employers determine whether the traits or characteristics measured by a personality test correlate with certain disabilities, and take affirmative steps to ensure that individuals with autism or cognitive, or mental-health related disabilities are not being inaccurately assessed and unlawfully screened out. One way to do so, according to the guidance, is to provide as much information about the tool as possible in advance, and inform candidates that reasonable accommodations, including alternative means of assessment, are available to persons with disabilities. Alternatively, as a “promising practice,” the EEOC recommends using only tools that measure abilities or qualifications that are truly necessary for the job.
Algorithmic Decision-Making and Disability-Related Inquiries and Medical Exams
The ADA prohibits employers from making disability-related inquiries or conducting medical exams before making a conditional offer of employment. The new guidance warns employers that certain questions on personality tests, including those that use algorithmic decision-making, may violate the ADA if the questions are likely to elicit information about a disability or medical diagnosis. As an example, the EEOC explains that asking candidates about whether friends would describe them as optimistic is permissible because it is not a question likely to elicit information about a disability. However, if a candidate with major depressive disorder is screened out from employment based on that question, the ADA could be violated if the candidate can perform essential job functions with or without an accommodation.
Actions Employers Can Take to Comply
To ensure their AI and algorithmic tools are ADA compliant, employers can take the following actions:
- Prevent unlawful screening: Screening of a candidate with disabilities is unlawful if the screened individual would have been able to perform the essential function of the job with reasonable accommodation if an accommodation is legally required. Companies should examine their hiring tools to ensure algorithms do not unlawfully screen out individuals with disabilities.
- Use accessible design: Companies should ensure that their AI and algorithmic tools are accessible to individuals with as many different kinds of disabilities as possible. This will minimize the odds of unfairly disadvantage to people with disabilities.
- Make reasonable accommodations: A reasonable accommodation is a “change in the way things are done that helps a job applicant or employee with a disability apply for a job, do a job, or enjoy equal benefits and privileges of employment.” Companies should make it clear and easy for job applicants to request reasonable accommodations. Requesting an accommodation should not decrease an applicant’s chances of being hired.
- Screen vendors properly: Companies should make sure that their technology vendors are in compliance with the ADA. Employers may be held responsible for the actions of software vendors contracted to perform hiring tests, even if the employer was unaware that a job applicant had reported a problem to the vendor.
- Keep requests confidential: Under the ADA, employers must keep all medical information related to accommodation requests confidential, and must store such information separately from the requestor’s personnel file. Companies processing accommodations requests must remember to keep the associated dialogues confidential and store related information in a confidential “medical” file.
Emerging technologies have great potential to help employers streamline the often-laborious hiring process. Yet even the most highly advanced AI and well-built algorithmic tools can raise thorny discrimination issues. Companies should preserve the value of their new hiring technology by diligently monitoring those technologies’ applications for ADA compliance.