Alexa, Siri, Sophia: Deconstructing AI's Subliminal Gender Bias

For women in artificial intelligence research, gender bias is a major barrier to success. Silicon Valley's gender problem isn't just a social justice issue however. Should the top positions in tech continue to go to men only, the tech world could be stifling its own capacity for innovation and threatening the future of AI research. The world's top female researchers are redefining the field and enacting a sea-change in the way the AI industry think about gender.

Gender bias in the tech industry dates back to its inception. The historically male-dominated industry has long possessed an almost cult-like meritocracy, where employees are often encouraged to devote their entire lives to the success of the product. This creates an environment where discriminatory practices remain pervasive under the guise of a reward system; employees outside the standard masculine mold are often denied the same pay or promotion pathways as male employees, even when they meet or exceed job expectations.

In 2015, tech investor Trae Vassallo and several colleagues co-authored a survey titled “The Elephant in the Valley." The survey investigated the experiences of female leaders and innovators in the tech industry, and the results were bleak; 84 percent of interviewees were told they were “too aggressive," 66 percent experienced exclusionary practices, and a shocking 60 percent experienced sexual harassment. Just 18 percent of undergraduates in computer science in 2011 were women, down from 37 percent in 1985.

Like the rest of the tech industry, AI's gender bias is similarly pervasive. The artificial intelligence sector is expected to grow from $21 billion to $190 billion between 2018 and 2025, and the employment demographic is overwhelmingly male. The field has had a difficult time developing its female workforce, potentially due to the nature of AI research itself.

“Research has become very narrowly focused on solving technical problems and not on the big questions," says Marie desJardins, Professor of Computer Science at the University of Maryland. Also, desJardins notes the distance between the work being done in AI and the betterment of society in general.

That gap could be turning women away from the field, since women tend to value their work's contribution to their community higher than men.

AI's diversity issues affect women as well as other gender minorities like transgender and non-binary individuals, and these diversity issues also continue beyond gender. “Cultural diversity is big too," says Heather Knight, founder of Marilyn Monrobot Labs in New York City. Racial underrepresentation in the tech world compounds issues for women from minority ethnic groups. Gender and racial bias in AI are significant enough to have an effect on the way the algorithms themselves are developed, which could have lasting consequences for society if the problem isn't met head-on.

If researchers use biased datasets to train AIs, gender bias may become embedded in the technology itself. A study conducted on image-recognition software in 2016 found patterns that reinforced gender stereotypes. When asked to associate images with either men or women, the algorithm consistently linked women to images of kitchens, reflecting or exaggerating the gender biases it perceived.

Since the 100,000 images used were collected broadly from the web, biases in media were reflected in the AI's analysis. In a similar case, Microsoft's conversational AI “Tay" took in data from Twitter conversations and began repeating racist and misogynist phrases in less than twenty-four hours.

AIs will need to be closely managed to avoid mirroring the gender biases present in today's society.

The link between gender discrimination and artificial intelligence doesn't end in employment statistics. Gender bias is implicit in AI itself. “There's a clear bias in the way women are depicted in science fiction," says Alex Haslam, media relations specialist for HowtoWatch.com. “AIs are overwhelmingly female, and are often depicted as dangerous."

Many critics have also found it problematic that almost every digital assistant uses a female name and voice. Siri, Google Assistant, Cortana, and Alexa all reinforce the stereotype of the female administrator. “It's much easier to find a female voice that everyone likes than a male voice that everyone likes," Stanford communications professor Clifford Nass tells CNN. Whether psychological or cultural, the presence of female AIs helps these stereotypes persist.

New efforts to close the gender gap in the sciences are charting a new course for those who have often been marginalized in the AI industry. Female professors, researchers, investors, and scientists are tackling gender bias in AI using innovative applications of technology, education, and more than a little common sense.

“The field of AI has traditionally been focused on computational intelligence, not on social or emotional intelligence," explains Rana el Kaliouby, co-founder of the AI research firm Affectiva. Kaliouby and other AI experts are looking to develop a social conscience for the AI algorithms of tomorrow, embedding moral and ethical principles into the technology.

Other female leaders in the AI field are addressing enrollment issues by designing education programs specifically for young girls. Millions of individuals have enrolled in AI and machine learning courses through programs like Coursera, with disadvantaged or underrepresented groups reporting the most benefit.

The incredible capacity of artificial intelligence is also addressing gender bias in society directly. A new AI algorithm developed by Google and The Geena Davis Institute on Gender in Media uses AI to detect male and female faces in popular films. The AI algorithm logs screen-time and speaking time for characters of different genders.

In the top films of the past three years, the algorithm found discouraging gaps; female characters received roughly half of the screen and speaking time of male characters. In the future, this data could assist filmmakers in avoiding techniques or casting selections that reinforce biases, encouraging stronger gender diversity in film.

The embedded nature of gender bias in today's society makes progress towards equality difficult, but burgeoning fields like artificial intelligence have a higher potential for social progress. Top computer scientists and AI experts have turned their attention to addressing gender bias in AI. If artificial intelligence lives up to its expectations as a game-changing technology, a more socially responsible foundation today could have a big influence on our future.