Google has decided to remove gendered terminology from its Cloud Vision API in an effort to avoid bias on the AI platform. The Cloud Vision API is an object recognition service that automatically identifies and labels items in photos.
The API can identify anything from animals to brand logos, making it much easier to search through a large number of images for a specific object. In the past, any people that turn up in those images have been assigned gendered labels like ‘man’ and ‘woman.’ Moving forward, the service will replace those gendered terms with gender-neutral alternatives like ‘person.’
In its announcement, Google noted that it is impossible to determine someone’s gender based only on their appearance. As a result, any attempt to label an individual as a man or a woman would reflect the assumptions of the API and its creators. Google therefore decided to ditch the gendered terminology in accordance with the second of its Artificial Intelligence Principles, which demands that Google “Avoid creating or reinforcing unfair bias.”
The decision acknowledged that AI Algorithms can recreate bias, and indicates that Google is actively working to alleviate the problem.
“Classifying people as male or female assumes that gender is binary,” said Mozilla tech policy fellow Frederike Kaltheuner. “Anyone who doesn’t fit will automatically be misclassified and misgendered. So this is about more than just bias — a person’s gender cannot be inferred by appearance. Any AI system that tried to do that will inevitably misgender people.”
Of course, issues of bias in AI go well beyond gender. Some algorithms have been shown to have a pronounced racial bias, which has particularly troubling implications when applied to law enforcement. Companies like IBM and Applause are currently hoping that larger datasets will lead to more equitable outcomes for civilians.
Google’s attempt to eliminate gender bias in its Cloud Vision API will not solve all of the problems with AI. However, it is nevertheless a welcome step in the right direction.
Source: Business Insider
February 21, 2020 – by Eric Weiss