Artificial intelligence is amazing. For HR, AI promises to eliminate the tsunami of transactions that plague us, find us the best of the best candidates, identify internal threats to our organization, and simply make life easier. Things we all want!
Artificial intelligence is also creepy and super scary at times. While it’s creepy Netflix’s AI knows what TV shows to recommend and Amazon’s AI predicts what I’ll need when I need it, they are helpful. Yet, the scary are truly scary. Microsoft’s artificial intelligent Twitter bot turned into a Hitler-loving racist in less than 24 hours. Google’s artificial intelligent photo recognition tool identified African Americans as gorillas. The reasons Microsoft turned off its bot and Google changed its tool was because of the discrimination risk regardless of whether it was a legal or public relations risk.
When HR looks to integrate AI into its operations, the potential for the same risks exist. Using AI or machine learning to improve our operations and identifying talent means the potential for discrimination exists. Even when we remove data related to protected class status, the potential still exists. It is almost inescapable, but it can be manageable.
One of the first things we need to examine is our relationship with vendors and how to appropriately use their expertise to reduce discrimination risk. We look to vendors to deploy AI. (HR rarely has the expertise in-house to create and release AI.) Vendors already help us do many, many things, like manage recruitment through ATSs, create payroll through T&A systems, and manage steps in our performance management process. We rely on vendors. Yet, our relationships with vendors does not shift the compliance risk from us to them. Employers are always on the hook for the decisions they make whether a vendor or technology helped.
Here are a few things to consider in the AI vendor relationship:
Vendor, Do You Know the Law?
I spend time with data and computer scientists who are on the cutting edge of AI. Every time I meet with them, I bring up the issue of discrimination, and they give me a dumbfounded look, like “Why would that apply to me?” Then, they say, “Well, I’ve never seen that.” (See the image above.) But as the mountains of evidence that AI and machine learning can discriminate, the more vendors are going to have their feet in the fire on this topic. Talk with vendors about the risk of discrimination, the EEOC’s requirements, and if they can intelligently respond to your questions and have plans in place to respond to your needs, SUPER! If they don’t, be skeptical.
Vendor, Explain Thyself
Remember our geometry teachers demanding that we “show our work?” The EEOC and other state and federal agencies already require employers do the same. With recordkeeping requirements and investigations, employers must show how they don’t discriminate. When employers rely on AI to assist with making decisions, the EEOC and other agencies are going to ask how the AI worked and what it did. When the AI is a vendor’s trade secret, will the vendor share? While vendors may not need to explain how it works in detail today, you may need them to in the future so you can respond to agency questions. Put this in your vendor contract. Also, require vendors to give you input and the authority to change how the AI works.
Vendor Test Strips
Finally, when AI helps you make employment decisions, are those decisions in-line with the Uniform Guidelines of Employee Selection Procedures? Are the criteria or decision-making of the AI really job-related and a business necessity? Are you able to explain that? Require vendors to validate, test, and revalidate their tools to your business and your positions. The more testing to show discrimination doesn’t exist the more helpful and compliant the decisions will be.
I want to reiterate my HR technology pledge. I don’t want to hold employers back from using tools that help them improve. But with every new piece of technology – just like every new employee – risk abounds. It’s the job of an employment attorney to help make the technology just a bit safer.