“무지개 출력 안 되게 알고리즘 손봐야 했다”…소수자 외면하는 AI업계

In the AI industry of South Korea, a trans person named Dada experienced the need to prevent the output of 'rainbow' forms, as the colors symbolize marginalized identities. Customers expressed concerns about AI generating images associated with marginalized groups. Dada did not think the customers had malicious intent; rather, it reminded her of how certain political views can be offended by specific imagery.

However, she stressed that this exclusion, if repeated, ultimately amounts to discrimination. She recalled learning about algorithmic discrimination in college but faced pressure to prioritize business goals in her work. She noted that AI design often overlooks the issue of discrimination, as developers tend to focus on generating appealing images rather than ensuring unbiased representation.

The black-box nature of big tech companies, which do not disclose their internal AI learning standards, adds to the concern. While visually impaired individuals working for domestic tech giants faced issues with web accessibility due to AI-generated code, they lamented the lack of attention to inclusive data for marginalized groups.

The cycle of marginalized data training leading to inaccessible code perpetuates the exclusion of these communities. While AI might appear as a hope, it often serves as a barrier for marginalized individuals, who now face potential severe exclusion if the accessibility issue persists.