Artificial Intelligence is the ability of a machine to mimic human cognitive functions that requires intelligence such as thinking and learning. As one of the new and notable technologies, AI is developing rapidly and broadly adopted in many fields, replacing people due to the many benefits it has, such as accuracy, speed, objectivity, and fairness. However, it turns out that Artificial Intelligence also discriminates against people.

One of the examples of discrimination in AI can be found in facial recognition. Facial recognition, first utilized in smartphones in 2017 as Apple introduced iPhone Xl, started to be adopted in many areas such as businesses, police, and the legal system. It is now a common technology for many people who use smartphones. But Joy Buolamwini, a researcher at the M.I.T. Media Lab, revealed that the facial recognition system discriminates against people.

In her experiment, she calculated how race and sex influence the error rate of facial recognition programs from IBM, Microsoft, and Face++. Surprisingly, the facial recognition programs of all three companies showed the same result: they were more accurate when a person in a picture was male and had a lighter skin color. To be specific, AI misclassified darker-skinned females the most, with the error rate of about 35%. On the contrary, the error rate of lighter-skinned males was less than 1%.

This difference in accuracy appears to be due to the underrepresentation of pictures of people of color and female in the data the programs used to learn. Since it is people that provide data to AI, the data can be biased, resulting in a biased AI.

Another case that Artificial Intelligence discriminated against happened when Amazon tried to implement AI for recruiting. In 2014 Amazon built an AI program that reviewed and rated the resumes of job applicants to mechanize the process of hiring people. But in 2015, the company found out that the AI gave low scores to female applicants. It penalized resume when it included the word “women” and devalued candidates from women’s university.  

Amazon’s AI program was trained to rate candidates by observing the patterns from resumes accepted to the company for over ten years. However, in the male-dominated tech field, the resumes submitted were mostly men’s, which caused the AI to be trained to prefer male applicants. Again, AI learned and mimicked the already existing discrimination.

AI is neither good nor bad. However, since AI is made by people with biases, their biases and beliefs can easily be embedded in the AI, whether it is intentional or not. AI’s fairness depends on how we make and use it. More awareness is required to solve the discrimination in Artificial Intelligence.

김주현 강남포스트 학생기자  webmaster@ignnews.kr

<저작권자 © 강남포스트, 무단 전재 및 재배포 금지>