We deal with algorithms every day. On social media platforms, algorithms are sold as tools that help improve our user experience. They show us content we would most likely interact with and items we are interested in purchasing.
One downside is that algorithms can limit what type of content we see. For example, Instagram’s switch from a chronological to a more algorithmic-reliant feed limited users’ ability to see posts from all of their follower lists. Although these examples may sound trivial, one serious downside to algorithms is an algorithmic bias or AI bias.
AI bias occurs when an algorithm produces systemically prejudiced results during the machine learning phase. Usually, these results are based on erroneous assumptions currently embedded in our society. Issues with AI bias are introduced by individuals who design or train the machine systems because they can unintentionally train algorithms with the same preconceived notions or prejudices they possess.
Sample bias and prejudice bias are two types of AI bias. Sample bias occurs when the data lacks diversity and, therefore, is not representative of the population. Public datasets are often used to train machines but, they usually lack representation from minority groups. While prejudice bias is when well-known stereotypes and prejudice are embedded in the machine learning process.
So, you might be thinking, how does this seriously impact me?
In my previous Data Analytics course, my final project dealt with analyzing the fairness of the home mortgage lending process in the United States. Using the home lending dataset provided by The Home Mortgage Disclosure Act (HMDA), we were asked to uncover whether applicant bias existed. We found that white applicants were more likely to be granted loans in comparison to most minority applicants. Black applicants were least likely to secure a loan, even if they were in the same tax bracket as their white counterparts.
The first image depicts the home loan approval rates by race. As you can see, Asian and White applicants are the most likely to be accepted.
The second image compares approval rates between Black and White applicants. Black applicants experience a lower acceptance rate at every socioeconomic status in comparison to their white counterparts.
So, if you’re looking to purchase a home in the future with the help of a home lending business as a minority applicant, then there is a good chance you might face discrimination.
Forbes not only touches on this discrimination but, really nails the issue on the head. Many businesses that we interact with are becoming more reliant on AI. A majority of resumes, home lending applications, and applications alike are routinely screened by machines before ever reaching a real person. As more companies embrace this trend of AI dependency, bias will continue to impact many of us.