EEOC Announces ADA Guidance for Using AI in Screening and Assessment

In May 2022, the Department of Justice and the Equal Employment Opportunity Commission (EEOC) announced plans to monitor employers’ use of artificial intelligence in employment decisions.

Working together, the agencies are specifically looking at AI systems that may discriminate against persons with a disability. Guidance was issued by both that outlines their concerns, and may suggest avenues for litigation they plan for the future.

The use of screening and assessment technologies continues to grow. Companies are moving beyond simple resume/application screening software to more advanced assessment tools.

Some businesses require applicants to pass basic or complex tests before their application is advanced. Others use chatbots or other interactive technology for pre-screening or ‘first interview’ assessments.

Existing employees may be subject to testing and screening technologies to become eligible for promotion. The use of tech can make it easier to assess employee skills and capabilities and help inform whether or not they’re ready to move up.

The government is concerned these may tend to exclude if structures are not built into the systems to accommodate the needs of a diverse applicant pool.

Where the agencies believe there is a potential for bias

There are several areas of technology the guidance specifies as potentially discriminatory. These include systems that generally rely on standardization to assess. Some areas of concern are:

• Application and employment tools may not be accessible to persons who use screen-enhancing software. Online processes should be available in a variety of user-interface options to accommodate the widest range of candidates and employees.
• Voice analysis technologies may evaluate an applicant’s verbal communication skills. For a person with a speech impairment, the system may tend to exclude. More advanced voice analysis systems may assess confidence or truthfulness of an applicant or employee’s statements. Some algorithms may be unable to adjust for neurodiverse candidates.
• Facial recognition technology may work in the same way as voice, but with more data points to analyze. These systems assess speech patterns and body language; whether or not a person is looking away when forming a response; or other physical characteristics. Candidates may not fit the technology’s ‘standard’ and be excluded.
• Computer-based testing, either pre- or post-employment, may rely on the ability to clearly see a computer screen, which could exclude those with a visual impairment. Many applicants’ home screens have technology to assist with visuals, but a company’s system may not be compatible.
• Timed testing may also tend to exclude. For persons with arthritis, for example, even though their answers were correct, the speed in which they were able to respond may be a determining factor not to interview or advance them.

Other areas where agencies believe there is a potential for bias

• Some systems pre-screen either with written questions or through the use of chatbots that may not allow applicants to reply completely. A feature that asks if candidates are able to stand for several hours per shift may not recognize a response that the task can be performed with an accommodation.
• Screening software may measure keystrokes to assess a job seeker or employee’s qualifications. These may not factor in voice recognition software that people could use to perform the necessary functions of the position.
• Other screening software may eliminate candidates with employment gaps, even though otherwise qualified. These may eliminate workers whose physical or mental health issues required time away from work to attend to their needs.
• Advanced AI algorithms form decisions not on an applicant or employee’s abilities and qualifications, but data about personality traits of those currently successful in the role. They may look for characteristics the system has determined align with success performing specific tasks or roles, but may not factor in the ability to execute the work with an accommodation.

What is screen-out and why does it occur?

The agencies refer to these types of exclusions as ‘screen-out.’ This occurs when a applicant or employee’s disability prevents them from meeting the minimum selection criteria set by the algorithm.

Whether eliminated entirely, or scored lower than the acceptable level set, screen-out occurs because of the disability — not because of the candidate or employee’s level of skill or proficiency.

The result is a missed opportunity for hire or promotion, often without an explanation or an opportunity to ask for an accommodation.

What are steps employers can take to combat AI bias?

As businesses work to create a diverse, welcoming culture, the technology that helps make recruitment and assessment easier may be working against them. A diverse workplace is more productive and innovative. To meet diversity goals, it will be necessary to assess any technology in use to assure screen-out isn’t occurring.

There are steps that hiring authorities can take to verify their internal and external systems are screening in — not out.

• Advise candidates and existing employees that you use technology to make screening or hiring decisions, but these do not preclude a request for a reasonable accommodation. Depending on the technology used, you may offer specific or general options. Welcome job seekers’ and employees’ requests for an accommodation.
• Let candidates and employees know they have the option to speak directly with a person, if they (or you) are concerned voice or facial recognition software may tend to exclude them.
• Inform candidates and employees who are required to take tests that accommodations can be made. This could include verbal/audio versions instead of written tests; additional time to complete the test; or other options that remove barriers to completion based on physical or mental health.
• Ensure your online processes, including screening and testing, are compatible with available screen reading software and other interfaces that create accessibility for as many individuals as possible.

Outsourcing screening does not outsource the responsibility

If you use a 3rd-party provider for screening and/or testing, inquire about the steps they take to accommodate applicants and employees who use the software.

Make sure vendors are working to include, rather than exclude, candidates and employees.

Outsourcing the work does not outsource the responsibility. Make sure vendors are working to include, rather than exclude, candidates and employees.

• Test the software yourself. Make sure it prominently notifies applicants or employees of their right to request an accommodation during any step of the process. Look for easily accessible means to make the request for an accommodation.
• Ask for verification that systems accommodate persons with a disability in all user-interfaces. Require assurances that a request for an accommodation does not weigh against any candidate or employee.
• Look for or request any alternate formats the vendor uses to accommodate requests, such as verbal or virtual options.
• If using software that screens for ‘characteristics’ that may predict success, assess whether these may tend to screen-out or lower the overall results of a person with a disability.
• Ask the vendor if they’ve tested the algorithm to ensure it’s not screening-out or disadvantaging persons with a disability. You’ll want to inquire what tools they used to evaluate their systems.

Make reasonable accommodations that work for both parties

While accommodations should be readily and easily available for every applicant and employee, they must be reasonable. The Americans with Disabilities Act specifies employers must make reasonable accommodations for qualified individuals throughout the employment process.

A request for expensive user-interface software to apply for a job may be a reasonable request to a multinational corporation: for a local small business it could pose an undue hardship.

Look for the easiest solution for your business and the applicant. If a job seeker is visually impaired and cannot use your online application, prominently include you’re happy to accommodate them filling out the form by phone or virtually.

If employees or candidates are required to take tests, offer an opportunity to add time to the test. You could also issue the test in another format, if necessary. You may find that a few minutes asking questions by phone nets an outstanding new hire.

Employers are within their rights to verify an accommodation request. An employee may ask for a longer time period to take a test, for example, based on their physical or mental health condition.

An employer who was previously unaware of the condition can request verification from a doctor to support the accommodation. While it may not be reasonable to ask every applicant for a doctor’s note, businesses may consider opening options to everyone, to be more inclusive.

Verify that hiring and employment processes work for all

Technology that allows HR professionals to outsource rote tasks, like screening resumes, offers more time to focus on high-value duties, like employee development. These conveniences come with risk, however.

The ultimate responsibility for non-discrimination in hiring and employment belongs to business.

To ensure that a workplace is diverse and welcoming, businesses will have to verify their hiring and employment processes work for every applicant and employee, regardless of their status.

Audit internal and external processes to ensure you’re in compliance with the spirit as well as the letter of all ADA legislation. Don’t rely on 3rd-party providers or technology. The ultimate responsibility for non-discrimination in hiring and employment belongs to business.