Where are the women in AI?

Tomorrow is the International Women’s day. If you ask your smartphone what that implies, it’s highly possible that a female voice will answer. Why? Some might say that it’s because of the lack of representation in AI – women make up 23...

Tomorrow is the International Women’s day. If you ask your smartphone what that implies, it’s highly possible that a female voice will answer. Why? Some might say that it’s because of the lack of representation in AI – women make up 23 percent of the AI professionals in the U.S. and 22 percent globally. Others refer to studies that present evidence that consumers prefer female voices when they need assistance. Either way, the five biggest American tech companies – Facebook, Apple, Amazon, Microsoft and Google/Alphabet – have the possibility of making a huge impact in making AI more equal.

Number of women in technological roles globally
Facebook: 23 percent (2019)
Apple: 23 percent (2018)
Amazon: does not report this
Microsoft: 21,4 percent (2019)
Google: 23 percent (2019)

Number of women in AI research staff globally
Facebook: 15 percent (2019)
Google: 10 percent (2019)

The lack of representation both in AI at large and in the abovementioned big five contributes to a bias in AI systems. As IBM puts it, the systems “are only as good as the data we put into them”, and biased data can easily be reproduced and used as non-biased. An example of when this happened is when Amazon tried an AI-program to do their recruiting. The program was trained using resumes submitted to Amazon over a 10-year period – but because of limited diversity in the company most of the resumes came from men. This taught the program to disadvantage resumes were the word “women’s” was being used, as in “all-women’s college”. The program is no longer in use.

“Amazon’s system taught itself that male candidates were preferable”

Dastin, 2018

Bias in AI can also be seen in for example speech-to-text technology or computer vision systems, where female voices and faces seem harder for the AI-systems to recognize than male ditto. However, female voices are more often used in AI voice assistants like Apple’s Siri and Amazon’s Alexa. UNESCO has examined this pattern of using female voices when creating AI assistants, and describes it as “a powerful illustration of gender biases coded into technology products”. They see that educating women and girls in technology is a way to mitigate these biases.

With education and awareness, hopefully we’ll soon see more women behind AI-systems than in them – and maybe ask Alex about the International Women’s day instead of Alexa.

Nora Myrne Widfors
Intern at the Office of Science and Innovation in Washington D.C.