acm-header
Sign In

Communications of the ACM

ACM TechNews

How Native Americans Try to Debug AI's Biases


The image recognition app botched its task, said Tracy Monteith, a senior Microsoft engineer and member of the Eastern Band of Cherokee Indians, because it didn’t have proper training data.

Credit: Juan Carlos Pagan

TThe annual conference for the American Indian Science and Engineering Society hosted a workshop where students created metadata to train a photo recognition algorithm to understand an image's cultural significance.

The students tagged images of ceremonial sage in a seashell and a 19th-century picture of Native American children outside a boarding school, with words carrying indigenous connotations.

The researchers then compared the algorithm's responses to those generated by a major image recognition application.

Microsoft engineer Tracy Montieth said the app was unsuccessful because it lacked proper training data, demonstrating that such data dictates the performance of artificial intelligence (AI), and in this case was biased against marginalized cultures.

Florida International University's W. Victor H. Yarlott said more accurate data makes AI systems more representative of human intelligence.

From The New York Times
View Full Article - May Require Paid Subscription

 

Abstracts Copyright © 2022 SmithBucklin, Washington, DC, USA


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account