The EEOC and AI Based Assessments- The inside scoop!

Featuring: Dr. Romella El Kharzazi

While the EEOC’s Uniform Guidelines on Employee Selection Procedures (aka UGES) provide a solid foundation when it comes to the do’s and don’ts of employment testing- it is always nice to hear about the EEOC’s stance on key issues straight from the horse’s mouth.

In this episode special guest Dr. Romella El Kharzazi, of the Equal Employment Opportunity Commission teaches us about the EEOC’s mission and their stance on all things employment testing. This enlightening conversation also includes a good deal of talk about the agency’s stance on AI based assessment tools.

This episode is a must listen for anyone who is considering using AI based assessment tools but is concerned about the legal risk.

Here is a quick outline of the key messages that arose during the conversation.

Getting to know the EEOC
The EEOC sees themselves as “ A force for good”- focuses on the core mission of fighting discrimination and making sure that everyone has an equal shot when it comes to employment.
The EEOC’s priorities are:
to protect vulnerable populations,
to educate businesses on discrimination and how to avoid it, and
to “build opportunities” for everyone no matter who they are or where they come from

The EEOC is not just about enforcement- they also create policy and do research into all areas of employment discrimination with a special focus on monitoring the employment climate at all federal agencies.

The EEOC and pre-hire testing
When it comes to testing, the EEOC’s core focus is on working with employers to monitor, manage, and eliminate bias from the hiring process. This is done via a formal complaints and investigation process that is geared towards working with employers to eliminate bias more than it is towards dragging them into court.

The requirements for compliance in hiring and testing are laid down by the UGES. The good news is that the UGES has been around a long time and legal and I/O Psych fields know them well and have been building testing products and programs that are compliant.

Beyond the specifics of the UGES- the advice for staying on the EEOC’s good side includes:

Job analysis- Establishing job relatedness of selection procedures is critical in all situations and should not be ignored. Especially when it comes to AI based tools, the absence of a job analysis will create exposure and risk.

Record keeping- Companies are on the hook for keeping applicant records- with no exceptions. It is not acceptable to claim exemption from the rules because your firm does not keep any records. While it is not mandatory for applicants to supply demographic data, compliance requires that there is a place in the application process for applicants to provide it. Failure to attempt to collect applicant demographic data creates exposure and risk

Be proactive- It is always required that employers seek alternative measures to replace those that may have adverse impact. Claiming business necessity for tests that show adverse impact only goes so far. Instead – it is your obligation to search for alternatives that have less adverse impact. Employers should be proactive and when aware of a problem- seek to fix it instead of keeping status quo and hoping they are not challenged.

AI based tools
While the EEOC does not yet have a big data policy, usual rules apply to AI based tools. It is really important to understand that the EEOC does not feel that all AI based tools have no merit. As long as the development, calibration, and use of the tools meet the requirements of the UGES, all is good. While there has yet to be a case related to the use of AI based assessments- the EEOC is paying attention and is active in monitoring how these tools are used.

So, how do we leverage the benefit of AI based tools without increasing exposure and risk?

First and foremost – there is a need to be sure that companies building and using these tools create a seat at the table for persons who understand the ins and outs of employment testing. Companies who rely purely on data science and empirically driven relationships are at risk. Adding I/O psychologists to the mix makes a ton of sense both for ensuring compliance and helping ensure the human side of hiring is properly represented. A hiring assessment company with no I/O on staff immediately sends up a red flag.

I/Os can help ensure that employers don’t overfit models to data sets, creating unreliable prediction across locations and over time.

It is paramount that employers do not simply push the blame for bias associated with a specific tool. Humans program and train AI based tools and it is people who make the ultimate decision to hire an applicant or reject them.

Bias can easily enter the equation when unsupervised learning is used so there is a need for caution when using these tools. For instance features such as social media data can be chock full of landmines that are biased against protected classes such as zip code and consumer behaviors. It is really important not to confuse the use of AI for customer/consumer insights from AI uses for hiring insights. They are not the same and what is valuable for understanding consumer needs is not appropriate or equivalent to what has merit when it comes to making hiring decisions.

If bias is present even with a strong correlation between the predictor and job performance there is still a great deal of exposure. In such cases the employer will be on the hook to:
Prove the assessment is job related- via a job analysis
Show that there is a business necessity for the bias, And
Show that alternative predictors were considered

If you are considering using AI based assessment tools for hiring here are the things that should guide your efforts:

Don’t try to use advanced assessment tools based on a fear of missing out or just because it seems trendy. Make sure you have real business need.
Remember the adage – Garbage In, Garbage Out. Your models are only as good as the data you feed them.
Job Analysis is critical!!! It should be the starting point of all predictive hiring tools even advanced ones.
Include an I/O psychologist who is an expert in the UGES in the mix.

Follow Romella on Twitter at: