What is AI Bias? How Machine Learning Systems Prevent Gender Equity

Artificial intelligence(AI) is used by humans as widely as humans worry about its’ use in 2024. Within the first five days of OpenAI launching ChatGPT, 1 million people were already using it. The AI market's net worth is projected to continue growing. Yet, 75% of people are worried about AI spreading misinformation - and those concerns aren’t unfounded. 

Automated robocall messages imitating Joe Biden. Taylor Swift’s deepfakes. The subtle biases woven into AI by the people who trained it – us. Let’s be honest - AI isn’t going anywhere. Frankly, I don’t want it to. I am not a writer who refuses to touch AI; I find it helpful when writing about a topic I’m unfamiliar with or need to tweak my tone. However, if we’re going to coexist with AI, it seems important that we truly understand its’ shortcomings and where they originate so we can address them. 

“What if Elon Musk personally calls you and tells you to vote for a certain candidate?” said Oren Etzioni, the founding CEO of the Allen Institute for AI, who stepped down last year to start the nonprofit AI2. “A lot of people would listen. But it’s not him.” 

Why Aren’t Computers Neutral?

There’s an episode of On Being that addresses AI as one of “the most human things there is.” I love this way of looking at AI, which is so often vilified because it’s true - AI was created by humans, for humans, with human input. So, of course, AI is biased. 

To put it more technically, systematic errors in machine learning systems create unfair outcomes, such as privileging one demographic over another. Originating from skewed data or flawed algorithms, these biases can have a ripple effect, influencing everything from your Netflix recommendations to more serious matters like job applications, legal decisions, and automated care recommendations. 

With AI infiltrating almost every job market, from automating factories to helping inform patient care and fixing the grammatical errors in your writing, the stakes are high – because large language models and machine learning systems aren’t programmed to treat everyone equally.

AI Bias in Healthcare Settings

Healthcare is anything but immune to AI bias. Picture this: a computer-automated diagnosis system, trained mostly on male data, trying to diagnose myocardial infarction(heart attack) in women, who are known to exhibit different symptoms. The result?

Misdiagnoses flying left and right—a true medical comedy of errors, except it’s not that funny. These biases can lead to a healthcare gap, widening the gender health divide rather than closing it. 

I wrote about gender equity in healthcare a few weeks ago, and I believe that AI bias in healthcare drives home a certain point:

The healthcare system tends to skew in favor of men. Men are what most models of well-being and illness are based on, and therefore, they are our biggest point of reference - specifically white men.

It’s not just that women exhibit similar symptoms to men but have been left out of studies; women and men often experience different healthcare struggles and different sets of symptoms for the same disease. 

One-size-fits-all medicine has never worked, and when we rely on a computer to assist in diagnosing patients or offering another opinion, that will become crystal clear.  

“Underrepresented data of women or minority groups can skew predictive AI algorithms. For example, computer-aided diagnosis (CAD) systems have been found to return lower accuracy results for black patients than white patients.” - IBM

How AI Bias Affects the Hiring Process

Moving on to the workplace, AI bias is present at work, too. We all know the frustration of sending in a job application only to get an email back an hour later that reads, “Thanks for your time. Unfortunately, you aren’t the right candidate for this position at this time.” or something along those lines, knowing that your application never touched human eyes. 

 Remember the infamous case of Amazon’s AI recruitment tool that turned sexist? The AI system was trained to recruit people from a pool of applicants over a 10-year period and essentially taught itself that men were preferable hires. 

CVs that included the word “women” were penalized, and in a less overt measure, the machine learning tool selected applications with words more likely to be used by men, like “executed.” 

This Amazon blunder could be a funny anecdote, except we live in a society where machine learning systems most often screen job applications. So, the women of this world can’t afford to live through another Amazon incident. As we hit “submit” on job applications or gig applications, how can we know we aren’t submitting to another sexist computer? 

The hiring process and simply navigating the workplace are minefields for women. Human job application screeners may be a hurdle—it’s not just computers that can penalize the word “woman.” AI should be a reliable tool, not another obstacle to overcome. 

AI Bias Matters to Everyone Now

AI has been around surprisingly long, with the first AI program invented in 1955. This time when AI is accepted and used as part of everyday life has been a long time coming. Marvin Minksy, in 1970, said, “From three to eight years, we will have a machine with the general intelligence of an average human being.” 

That was incredibly optimistic, and it is not true yet(in my opinion). Maybe it never will be. However, AI is capable of making our lives a little easier. That’s exactly why we should care about mitigating AI bias. 

Machine learning systems screen our job applications, decide who gets a loan, recommend TV shows, impact healthcare decisions, help the cops police, and decide what advertisements we see. As AI becomes an invisible but omnipresent decision-maker, its biases can significantly impact real lives and dreams, making the quest for equity in gender and across other identity groups an uphill battle. 

We need to address bias at the root and encourage the implementation of stricter AI regulation and governance. AI has promise, but it also has limitations, and it’s on us to address our own bias and all that it could affect.

Mitigating AI Bias

Countering AI bias is not just about tweaking codes and feeding better data; it’s about rethinking who gets to design these systems and how. IBM suggests these measures to address AI bias:

• Assess datasets for bias
• Address algorithmic bias, whether programming errors or a developer incorporating their own bias
• Address if certain datasets are favored over others
• Direct, monitor and manage the AI activities of an organization to ensure compliance, transparency, and fairness

I’ve previously said that we'll get left behind if we don’t embrace AI. So many people seem reluctant to use AI, and I understand why. Ignoring it will not make it go away. It very well may leave some of us without a job one day. 

So, my question is, will you embrace AI with your eyes open to its issues?

Moving Toward the Future with AI

From the doctor’s office to hiring committees and criminal justice systems, AI’s biases can keep us in an inequitable society. Or, we can address our biases and AI issues, solve them, and move forward with this powerful tool. 

Let’s not leave it to chance. Instead, let’s mold it into the enlightened ally we need, ensuring it champions, rather than challenges, gender equity. After all, a future led by a fair and unbiased AI is not just smarter; it's undoubtedly brighter for everyone.

Now You Can Confidently Incorporate AI Into Your Writing Strategy

When you work with Camille Prairie Creative Co., you’ll get tailored support for your copywriting needs from professionals who know how to work with AI to produce results that have a human touch.

Whether you need human-written or human-edited work, we’ll work with you to create bespoke, high-quality webpages and articles designed to attract your ideal customer. Learn more about optimizing your online presence in under 5 minutes today.

Subscribe for more Monday insights and check out my work! Have a great Monday :)