Computing has a problem — there are too few women in the field.
Only 26 percent of computer professionals were women in 2013, according to a recent review by the American Association of University Women. That figure has dropped 9 percent since 1990.
Explanations abound. Some say the industry is masculine by design. Others claim computer culture is unwelcoming — even hostile — to women. So, while STEM fields like biology, chemistry, and engineering see an increase in diversity, computing does not. Regardless, it’s a serious problem.
Artificial intelligence is still in its infancy, but it’s poised to become the most disruptive technology since the Internet. AI will be everywhere — in your phone, in your fridge, in your Ford. Intelligent algorithms already track your online activity, find your face in Facebook photos, and help you with your finances. Within the next few decades they’ll completely control your car and monitor your heart health. An AI may one day even be your favorite artist.
The programs written today will inform the systems built tomorrow. And if designers all have one worldview, we can expect equally narrow-minded machines.
AI are already biased
Last year, a Carnegie Mellon University study found that far fewer women than men were shown Google ads for high paying jobs. The researchers developed a tool called AdFisher that creates simulated profiles and runs browser experiments by surfing the web and collecting data on how slight changes in profiles and preference affect the content shown.
“The male users were shown the high-paying job ads about 1,800 times, compared to female users who saw those ads about 300 times,” Amit Datta, a Ph.D. student in electrical and computer engineering, said in a press release.
The systems aren’t just gender biased. In May, an investigative report by ProPublica uncovered that a popular software used to predict future criminals had racist tendencies. The system would falsely flag black defendants as high risk while often incorrectly labeling white defendants as low risk.
Microsoft researcher Margaret Mitchell called AI a “sea of dudes.” Machine learning researcher Kate Crawford claimed the industry has a “white guy problem” in an article for the New York Times. “We need to be vigilant about how we design and train these machine-learning systems, or we will see ingrained forms of bias built into the artificial intelligence of the future,” she wrote. “Like all technologies before it, artificial intelligence will reflect the values of its creators. So inclusivity matters… Otherwise, we risk constructing machine intelligence that mirrors a narrow and privileged vision of society, with its old, familiar biases and stereotypes.”
Fixing the system
The sea of dudes was overflowing at Rework Deep Learning Summit in London this September, where women were few and far between the roughly 500 attendees.
Rework founder Nikita Johnson recognizes the gender disparity in AI and wants to do something about it.
“If AI systems are built primarily by men only, then they are more likely to create biased results and the representation of the builders will dominate,” she told Digital Trends. “By limited diversity in teams, we limit the breadth of experience that can be brought into a project. For instance, data sets need to be assembled by both men and women to ensure that the results from the data includes a broad look at gender issues.”
Through “Women in Machine Intelligence” events, Johnson and her mostly-female team highlight female talent and encourage attendees to find peers, partners, and mentors in other women. She sees this networking as a necessary step towards growing female representation in the field.
“One of the reasons [for the lack of diversity within the AI community] is the cycle between a lack of role models for young women and girls to look up to,” Johnson said. “Therefore the lack of motivation for women and girls to chose AI and computer science as a potential career route.”
But as Johnson pointed out, there are plenty of inspiring women in computing, a handful of whom spoke at the conference.
Irina Higgins from Google DeepMind demonstrated her team’s concept for a system that can learn visual concepts without supervision. Raia Hadsell – also of DeepMind — discussed her team’s research into simulated deep reinforcement learning, a process by which a physical robot trains skills through a simulated version of itself. She likened the technique to learning like humans do.
A couple hours later, Miriam Redi of Bell Labs Cambridge shined a light on the invisible side of visual data by showing how researchers can uncover the subjective biases built into systems. In a study on aesthetics, Redi and her team developed a deep learning system that automatically scores images in terms of their compositional beauty. The study helped clarify what makes a good portrait different from a bad one from an algorithmic perspective, but also revealed a group of features about race and gender, which the system incorrectly deemed relevant.
“By doing these kinds of studies and explorations of what’s going on inside of our subjective machine visions system, we can avoid these kind of racist behaviors of the machines,” Redi said.
Monitoring algorithms to avoid gender-bias helps develop more inclusive systems and events like those organized by Rework offer encouragement through connections, but interest and motivation may come much earlier for young women. Support from family is key. And there are dozens of apps and educational toys to teach kids the basics of coding.
Indeed, University of Cambridge AI researcher Shaona Ghosh insists the initial motivation should come from closer to home. “First and foremost the support, belief and encouragement should come from the parents,” she said. “It’s a huge and important step that no amount of education and opportunity can compensate for.”
There’s no simple way to turn the tide of the sea of dudes and promote women in AI, but it’s in our collective best interest to do so. It takes effort from organizers like Johnson, writers like Crawford, and researchers in industry and academia, but as Ghosh advised, perhaps the most effective effort begins at home with the parents who raise the next generation of computer scientists.