THE BEST HIRE

Machine learning can’t fix algorithmic bias. But humans can

Studies show direct links between diversity and workplace performance.
Studies show direct links between diversity and workplace performance.
Image: AP Photo/Shizuo Kambayashi

The fact that tech has a long way to go when it comes to its lack of diversity shouldn’t be news to anyone at this point. The technology sector is the third biggest contributor to the US economy. And the people behind it—from founders to hiring managers to investors—overwhelmingly look like me: white men with a visible degree of affluence.

Similarly, more than 90% of American venture capitalists are white men, and those white men tend to fund startups led by people who also look like them. And as a result, men receive 35 times more funding than women.

Obviously, this should not be the reality. While programs like Girls Who Code work to turn the tide and diversify STEM fields, women still earn just 18% of computer science bachelor’s degrees in the US. Meanwhile, white men are getting the education, the opportunities, and, eventually, the leadership roles.

Ultimately, tech is the arm of innovation in our country. But since it’s largely being programmed by people who look and think alike, the impact on everything in our world couldn’t be more immense.

Robots versus humans?

If there’s one thing we’ve learned, it’s that we can’t rely on technology as a solution to fix, well, technology. In other words, don’t blame the algorithm: AI and machine learning won’t fix our tech bias problem when they are inherently biased because of how it was designed, and by whom. Humans got us into this mess, and humans need to solve it.

A host of studies have identified direct links between diversity and workplace performance. In 2017, for example, global consultancy firm McKinsey & Company examined 1,000 companies and found that diverse teams yielded substantial improvements in profitability and long-term valuation. A study that looked at 4,277 companies in Spain illustrated that the companies with the most women had a better chance of bringing radical innovations to market.

When technology like machine learning is designed, coded, built, and scaled by a homogenous team, the results can be disastrous and even, quite literally, deadly. A recent story from the Georgia Institute of Technology indicated that autonomous vehicles might have a more difficult time detecting pedestrians with darker skin. The study found that because those driverless cars were programmed mostly by young white male engineers, systems were 5% less accurate when recognizing people who have darker skin than those with lighter skin, stemming from the fact that fewer images of people with darker skin tones were used during programming and testing.

The impact is heavily felt in the hiring and recruitment process as well. Some suggest letting AI sort and rank job candidates. But that solution runs into the same problem as the driverless cars, because in most cases homogeneous groups are building it. For instance, when Amazon relied on an internal recruiting tool  programmed based on past hiring decisions, because those decisions were made by people who had largely favored men over women, the algorithm learned to do the same. When engineers programmed the machine learning tool to ignore overtly gendered words in response, the technology instead started seeking implicitly gendered adjectives to accomplish what it had learned was the goal: selecting and promoting male applicants.

We often subconsciously try to mirror ourselves, spending time and working with people of similar backgrounds. Bias is real and more often than not, creeps into so much of the hiring process. One of the best solutions we’ve found is to intentionally put on blinders. Blinding certain candidate information and focusing on their achievements or experience over other elements like names, zip codes, and backgrounds eliminates the risk of subconscious bias creeping into decisions. Looking instead at where people have excelled or shown willingness to learn new skills and adapt lets our decision-making focus on what matters—professional potential and ability—rather than any preconceived, inherent notions we have about someone.

It should come as no surprise that tech visionaries turn to their engineering prowess for everything, but there comes a time when people problems need to be solved by people.