With AI bias now a critical topic of conversation in hiring, companies leveraging the technology need the right data mining procedures to ensure underlying data sets aren’t perpetuating existing issues.
We see artificial intelligence (AI) at work every day – from determining what appears first in our Google search for “nearby restaurants,” our social media feeds and online shopping recommendations, to deciding our credit limit or sharing personalized job recommendations. These instances of AI have become so engrained in our daily lives that we often don’t even realize the tech behind quick tasks such as being able to order an Uber or purchase regular household items by simply clicking “order again.”
On a much bigger scale, AI’s ability to sift through, understand, and make predictions based on data has transformed nearly all industries from suggesting possible patient conditions and treatment options in healthcare to anticipating stock market changes in finance. While the tech is still new to the hiring industry, we’re seeing AI reshape talent acquisition by making the process much more efficient, proactive and predictive, leading companies to the right talent, at the right time.
But with great power comes great responsibility. AI offers many benefits for organizations, but we’re still in the early stages of the maturity curve, which means there are inevitably adoption growing pains to work through. The prevalence of AI bias is one of those pain points that has attracted media attention because prejudice is one of the exact things these algorithms were supposed to correct due to the technology’s superior decision-making capabilities.
Is AI Doing More Harm Than Good?
With the realization that AI bias exists, many in the business community have started to question whether the technology is truly helping in the ways believed. For example, one algorithm within a search engine was found to be more likely to show male job seekers a pair of ads for high-paying executive jobs than the equivalent female job seekers. Another algorithm used in the U.S. to guide sentencing by predicting the likelihood of a criminal reoffending, seemingly predicts that black defendants pose a higher risk of recidivism than they do. And some of the latest gender-recognition AI technologies can correctly identify a person’s gender from a photograph 99 percent of the time, but only for white men. Massachusetts Institute of Technology found that number dropped to just 35 percent for dark-skinned women.
The workforce management industry is certainly not immune to AI bias. Some companies have gone as far as removing AI tools from their hiring processes all together after discovering the technology was being discriminatory. It’s important to understand that the issue is not because of the AI technology itself, but rather the underlying patterns in the data that the tools are built on.
If historic data houses any form of prejudice, such as that men have consistently – even if unintentionally – been elevated to executive roles over women, the decisions made by these algorithms are inherently biased. AI doesn’t create bias, but it can perpetuate existing prejudice and foundational issues within an organization’s data sets.
But that doesn’t mean AI is a lost cause. It just means the global business community needs to adapt.
Fighting AI Bias at the Core
AI has the power to address hiring bias at the core – the more that employers apply the right algorithms and data mining processes that help hiring managers through the decision-making process, the more enabled these teams are to take existing bias out, not introduce it. Organizations need to adopt holistic data mining approaches and be more aware of the data sets used in AI and machine learning models.
The key to removing AI bias is knowing where biased data exists and having a global data set that doesn’t include qualifiers – like gender, race or other factors – that aren’t pertinent to the hiring decision. Organizations should also be careful of pulling in external demographic data sources – for example, filtering out zip codes that have high rates of failing criminal background checks, but could still have promising candidates, and making sure algorithms account for the fact that different countries use different screening criteria.
The most important factor contributing to organizations’ ability to fight AI bias is knowing data sets inside out and where bias potentially exists. Then, fine-tune algorithms and test potential outcomes to remove existing bias. AI can be an amazing tool for weeding out biased hiring practices, it just needs the right data to base its decisions off of.
AI at Work
Once organizations take the proper steps to eliminate bias in their hiring practices, AI enables them to find and match the best candidates for the job. For example, AI tools can automatically convert every CV into a searchable candidate record that can then leverage semantic search technology to more effectively match the meaning of the candidate’s content with the intent of the job requisition. This means candidate matching can include both fact-based requirements, such as years of experience and education, as well as softer characteristics, such as cultural fit.
A human-based review can then focus on the most likely candidates to meet job expectations, cutting down on the review time and ultimately, speed the path to fill an open position. AI capabilities not only allow teams to make hiring decisions faster, but also enables HR and procurement managers to focus on more creative, strategic planning that fosters efficient operational processes, employee retention, and a stronger bottom line.
Together, humans and AI can be an unstoppable combination for finding the best candidates. Success starts with the right data sets and leveraging that data to create algorithms that reduce bias in hiring, not perpetuate it.