Google AI Dumps Gender Labels to Dodge Bias From Cloud Vision API

“Woman” or “man” as gender labels will no longer be used within the Cloud Vision API created by Google Cloud AI. Until today, the company has been using labels to identify biological sexes in its images and this labeling is being used to classify images and train machine learning models. According to Google, this system is being branded as a gendered program and therefore is removing such labels because it violates their gender-bias-free code. “Given that a person’s gender cannot be inferred by appearance, we have decided to remove these labels in order to align with the artificial intelligence principles at Google, specifically Principle #2: Avoid creating or reinforcing unfair bias. After today, a non-gendered label such as ‘person’ will be returned by Cloud Vision API,” said Google in an interview.

Google is showing that they are willing to keep up with the times

Today, a majority of top AI companies like Microsoft’s Azure and Amazon’s AWS still continue to use gendered labels and one wonder if because of Google, these and others too will follow suit, however, only time will tell. In a move that is making waves, it was also bound to happen because in 2018 Google blocked the use of gender-based pronouns in one of their AI tools and this meant that the tech behemoth was, in fact, paying much attention to the times that they are living in and that their tech is something everyone is using today, regardless of their gender. Today, with regards to facial recognition programs, many of them do predict the gender of people, but these programs have shown their shortcomings in the sense that they are having trouble identifying people who are transgender and people of color. One study conducted by the University of Colorado, Boulder revealed that ‘AI from Amazon, Clarifai, Microsoft, and others maintained accuracy rates above 95% for cisgender men and women but misidentified trans men as women 38% of the time. People with no gender identity were misidentified 100% of the time.’ While there were no gender classifications, there were biological sex classifications like man and woman, while companies like Microsoft can identify certain professional people as male or female, like, waiter/waitress, airman, airwoman, for example. 

It can’t be said that these technologies are of high-performance because they are incredibly biased against certain groups

The truth is that all systems in the world today are in some way political and therefore it comes as great news that a company like Google is taking steps to take political notions and correctness seriously. The last few years have shown that when it comes to facial recognition people of color and those from the transgender community are the ones on whom such technology has failed the most and a lot of this is to do with the fact that these programs are being created by straight, white men. Therefore, it can’t be said that these technologies are of high-performance because they are incredibly biased against certain groups of people. However, not everyone agrees that this is the best move by Google since they are prioritizing politics over product quality, but it is undeniable that their decision will at least remove one area of bias. “Anytime you automatically classify people, whether that’s their gender, or their sexual orientation, you need to decide on which categories you use in the first place — and this comes with lots of assumptions,” said linguist and programmer Angus B. Grieve-Smith on Twitter. Hence, is something will help eliminate gender bias, it definitely is a win in the larger scheme of things. 

Image credit:

Related Articles