Artificial Intelligence in HR Could Make Hiring More Biased
Artificial intelligence in HR offers many advantages, from faster application processing to automated screening interviews. Perhaps the best advantage, AI offers is to completely eliminate bias from the hiring process. After all, a machine wouldn’t discriminate based on someone’s gender or ethnicity.
Unfortunately, machines can be biased – sometimes, even more blatantly so than humans. In 2018, Amazon was forced to scrap a sophisticated recruitment platform when they discovered that the AI was ruthlessly discriminating against women. Facebook’s advertising algorithm has a well-documented issue where ads for low-paying jobs, such as taxi drivers or janitors, are disproportionately targeted at minority users.
The biggest tech companies in the world struggle with bias in AI. That should give the rest of us pause for thought when automating HR processes.
How bias affects artificial intelligence in HR
To understand how a machine can discriminate, we have to look at how AI works.
Contemporary AI uses a technique called Machine Learning. ML is a software process that studies huge volumes of information (known as “training data”) and identifies deep-lying patterns. The system learns from these patterns, and then it uses this knowledge to make decisions.
So, for example, imagine you’re training an AI recruitment tool. You might start by giving it access to all of your hiring data and your file of resumes. The AI will study them and learn things about your previous hiring patterns like:
- Which resumes lead to an interview?
- Which resumes get rejected?
- Which candidates get hired?
- How long did each candidate stay in their position?
- Who went on to a leadership position later in their career?
Once the AI understands your historical hiring activity, it can make decisions about the future. The AI will flag up promising candidates and filter out anyone unsuitable.
If you can already see the problem here, you’re one step ahead of Amazon’s engineers.
AI learns from historical data, which contains all of our current biases. For example, if you asked an AI to look at Fortune 500 companies’ current leadership, it would notice a clear pattern: 92.6% of CEOs are men. Based on this data, an AI would determine that male candidates are better suited to leadership roles.
This seems to be what happened in systems such as Amazon’s failed human resources AI project. Historical data said that most previous successful hires were men; therefore, the AI prioritized applications from men. And even if the resume doesn’t explicitly mention gender, the system might discriminate in other ways. For example, it could ignore candidates who went to an all-women college, played on a women’s sports team, or who worked in a field traditionally associated with female employees.
How to tackle AI bias
Should we use artificial intelligence in HR at all?
Of course we should. AI is already a major part of human capital management, with 70% of large companies reporting significant gains in productivity when they adopt automation.
And AI can help to reduce discrimination from the recruitment process, resulting in a more diverse workforce, when used correctly. It’s all a matter of context, oversight, and data quality. A McKinsey report on AI puts the issue in a nutshell: “AI can help reduce bias, but it can also bake in and scale bias.”
If you’re working with AI tools such as automated applicant tracking systems, here are a few things you need to keep in mind:
Always investigate the training data
Every AI or machine learning tool was developed with set of training data. Talk to your software vendor about where this data was sourced, how it was processed, and what measures were taken to remove bias from the data. Many software vendors will be proud to tell you about their anti-discrimination strategy.
Broaden your search for candidates
AI can only work with the data you give it. If you want your system to produce diverse candidates, you’ll have to supply it with diverse applicants. This means reviewing your process and addressing difficult questions like:
- Are we advertising vacancies in the right places?
- Do we use inclusive language in our advertising?
- Does our employer branding demonstrate promote diversity?
- Are we using employee referral programs to reach out to minority communities?
- Do we work with recruitment partners who are equally committed to inclusiveness?
An AI can’t help you answer any of those questions. But if you start attracting diverse candidates, an AI can help you find the brightest talent.
Diversify your HR team
Bias is usually unconscious. We don’t all have the same lived experience, so we can’t perceive the challenges others face. That’s why it’s essential to have a wide range of perspectives within the HR team, so that our colleagues can flag up the problems we might have missed. Remember – representation alone isn’t enough. Everyone must have an equal voice so they can start a conversation about potential inclusiveness issues.
Keep a close eye on outputs
One great thing about AI tools is that you can analyze every stages of the process. Reports will give you detailed data about which resumes were rejected, which ones progressed, and which were prioritized. Pay close attention to the entire automated process and look for signs that bias may be creeping in. It’s also good to start diversity goals, such as 50% female candidates, and to measure outputs against your targets.
Don’t forget the human touch
Amazon has recovered from previous setbacks and leaned into automated hiring again. Now, some candidates are reporting that they’ve gone from application to offer without speaking to a single person. This strategy works for a tech giant, but most companies need face-to-face human contact with potential hires. It’s a chance to assess things an AI can’t perceive, like soft skills, culture fit, or leadership potential.
AI can take a lot of work out of recruitment, and it can go some way towards reducing unconscious bias. But to find and retain the best talent, you’ll need a hiring process that’s made by humans, for humans