Back to Blogs

Bias in AI: How Recruiting With AI Reduces Costs... And Diversity

Published on Nov 24, 2021

In 2017, CareerBuilder conducted an online survey involving 231 hiring managers belonging to various industries and companies of different sizes. The survey found that, on average, managers that did not use process automation lost 14 hours a week compared to managers who fully automated manual, repetitive tasks. That is the average; slightly over 10% of the participants lost 30 hours or more.

The objective of the survey was to understand how technology is transforming a fundamentally human aspect of business. What did it find? Nothing that should surprise. Even in its weakest form of process automation, artificial intelligence had already disrupted most aspects of modern business. Sales, marketing, customer support, warehouse operations, security—you name it. It was transformed.

Hiring was next.

And managers agreed. Over half of the survey’s respondents claimed that they already use AI in some form to screen candidates and that ‘hiring tech’ will become mainstream within the next five years. It is almost 2022. Were they right?

Indeed. Today, virtually every Fortune 500 company uses AI to find, screen, and monitor candidates and their performance. Given the decreasing price of computing, even mid-sized and smaller, tech-savvy companies are now embracing the trend. And while many have restricted the use of AI to screen and capture candidate data, many have dared to use it end-to-end. That is, through screening, capturing data, interviewing, and onboarding.

That seems to be progress. But is it?

By adopting AI-based hiring tech, what companies seem to gain in efficiency, they lose in talent. After five years, AI has indeed come to disrupt hiring. But its use is deemed unethical. It runs on decades-old, biased data, reinforcing decades-old social, cultural, and economic prejudices. Instead of making hiring more diverse, equal, and inclusive, the technology is further widening the already wide skills, and consequently, wage gap.

‘The holy grail of hiring’

If businesses could, they would not hire. Instead, they would retain. Because hiring is not just challenging. It is also shockingly expensive.

Anyone who has hired or has been hired understands that the process is long and often involves several people. To hire is to really find the proverbial needle hidden in the haystack. And it is fair to say that no one finds it in their first attempt.

In the long run, hiring can eat up hours of productivity for hiring managers and team leaders alike, which is especially hurtful when candidates are ultimately rejected. Of course, that is just hiring—finding, screening, interviewing. Then comes training. A new employee cannot contribute what the former employee did from the get-go. Training, inevitably, costs companies—several million, and perhaps even billions, depending on the company’s size.

Technology is widening the already wide skills and wage gap

Savings. Millions and perhaps billions of dollars. That is why companies are desperately seeking AI-based recruitment tools and tech. IBM has claimed that since 2011, it has saved over $1 billion by introducing AI to hiring. And while companies use process automation to streamline, they really seek the holy grail of hiring. An algorithm that determines the best candidate without any human intervention—that is, with minimum expenditure.

Actually, such algorithms already exist. The algorithms do not automate hiring end-to-end. However, they offer tremendous leaps in efficiency that transcend simple operations like data extraction.

Amazon, for example, deployed such an algorithm. The world’s largest retailer is renowned for its aggressive approach to automation. It has deployed automation bots in warehouses. It automates pricing decisions. And in 2014, it decided to automate what is notoriously resistant to automation: hiring.

According to a report by Reuters, Amazon developed an AI tool that would find the five top talents in a pool of 100 candidates. Much like its products, the tool would rate the candidates on a scale of five stars. In 2019, HireVue developed something similar. The hiring tech company built an AI tool to analyze video interviews for facial expressions, tone of voice, language usage, and body language. HireVue would use the data to assess a candidate’s emotional intelligence, communication skills, and cognitive ability.

How technology is transforming a fundamentally human aspect of business

Artificial intelligence in hiring is the antithesis of what the best-selling author, Malcolm Gladwell, has called hiring nihilism. Hiring is so challenging because human beings are laughably bad at judging personalities and predicting performance. Gladwell, therefore, argues that pouring hours of time and money into hiring has as much value as hiring arbitrarily. AI, however, can determine the right candidate because it can crunch infinitely more data—different data—and is, therefore, more reliable in its judgment.

But it is not.

AI is no magic bullet

Amazon developed the holy grail of hiring. Except that the holy grail preferred men in technical roles over women. Amazon tried to reform the algorithm, but ultimately abandoned it in 2018 after executives lost all hope.

In 2019, the non-profit Electronic Privacy Information Center (EPIC) filed a complaint against HireVue. EPIC alleged that the use of HireVue’s tool constituted “unfair and deceptive trading practices.” HireVue discontinued its use last year.

Is it fair to reject a candidate because they tweeted something offensive

Simpler tools exist that automate background checks, accessing data available on public domains such as social media. Even simpler tools exist that screen CVs based on specific keywords. Both have their own problems. The former raises privacy concerns. It also threatens our autonomy. Is it fair to reject a candidate because they tweeted something offensive?

The second kind of tool prefers keywords primarily used by a majority. A report by Accenture also found that keyword-based tools value credentials over skills. According to the report, keyword-based screening tools unfairly filtered out over 27 million workers last year. A large chunk of the workers were immigrants, veterans, and caregivers.

Bias will inevitably creep in unless AI-based hiring tech is developed strategically and rigorously. Most people blame the data, and they are right. But how the algorithms are designed also affects who the algorithms select.

Amazon automation warehouse

For example, a quick glance at Big Tech’s employee breakdown shows that technical roles are dominated by men—white men, in particular. In fact, white-male dominance in tech can be traced back to the 1980s. Back then, a combination of cultural and socioeconomic factors prevented women and minorities from getting an education and breaking into tech. Historical data does not ask why, but reflects the resulting disparity. And an algorithm based on historical data will conclude that men are, say, more capable programmers than women, thereby sustaining the disparity.

But even if datasets do not introduce bias, designers could. A designer, for example, could create an algorithm that values hard skills over soft. Such an algorithm could screen out candidates who lack skills but are great team players. The inverse could screen out candidates who are talented, but introverted.

Yes, AI will revolutionize hiring. But today, the technology is not ready. Certainly not on its own. Data is not representative. It is not even transparent. The same is true for algorithms. We do not know what data companies collect and how algorithms filter candidates. The space is not even regulated. There exist no strict policies or penalties for using or misusing the technology. Nor does there exist an independent committee that reviews data or designs.

One wonders: Who gets to decide who gets hired, and why? Who is accountable? What qualities, according to Amazon, make for an excellent technical lead? Why? What qualifies, according to HireVue, as suitable? Which facial expression, tone of voice, and accent? And why?

We associate technology with progress. But instead of making the world equal and fair, AI in hiring today reinforces inequality and injustice, further widening the education-skills-wage gap. Of course, we strive to eliminate hiring nihilism. But we cannot rely absolutely on a technology that is biased and unregulated—a technology that threatens the progress made after decades of public reforms and activism.

AI-based hiring tech

Perhaps we are misguided and human hiring managers cannot be replaced in the first place. As CareerBuilder’s chief human resources officer, Rosemary Haefner, explained, robots and AI cannot replace the “human element of HR that shapes the company culture, provides an environment for employees built on IQ and EQ, works hand in hand with company leaders to meet business goals and ensures employees have the training and support to thrive. You need living, dynamic people who can navigate the ‘gray’ to do that, not robots that can quickly work through black and white.”

With offices in New York, Austin, Seattle, London, Zurich, Pune, and Hyderabad, SG Analytics is a leading research and analytics company that provides tailor-made insights to enterprises worldwide. If you are looking to make critical data-driven decisions, decisions that enable accelerated growth and breakthrough performance, contact us today.

Our Services