The world of AI has a new problem: Very few women

Julia Kroll’s career in computer science started almost by mistake. 

As a freshman at Carleton College in Northfield, Minn., Kroll, now 24, was set on enrolling in a cognitive science class, but it was full. As a backup, she took an introductory class in computer science. She was hooked. The Madison, Wisc. native shifted her plans from majoring in something like English, to comp-sci.

That decision led her to Amazon in 2017, where she is now a language data engineer working on machine learning for the well-known personal assistant, Alexa.

“I had always been good at science and math in school, but I hadn’t had that exposure to computer science,” she says. “My parents were happy that I wasn’t going to be an English major after all.”

Kroll is among a fast-growing group of professionals across industries who is specializing in artificial intelligence. While working in AI and machine learning is becoming increasingly common, Kroll stands out: Only 22% of AI professionals across the globe are female, compared to 78% who are male, according to LinkedIn data released in the World Economic Forum's Global Gender Gap Report.

An AI gender skill gap this expansive has critical implications for both women and society. The number of workers with AI skills grew 190% globally between 2015 and 2017, and for good reason. Six of the 15 fastest-growing jobs of the past year call for these skills and AI specialists are becoming some of the highest paid experts across tech. If the gender gap in this specialty persists, it may widen the existing pay disparities between men and women in this industry and beyond. Meanwhile, experts told LinkedIn that having mostly male teams build AI technology that soon will be run mostly by machines poses serious ethical challenges.

The fear? Tech that’s built solely from the perspective of one gender or ethnic group may ultimately disregard or discount the needs and values of another.

Work hard, or ‘miss this moment’

The gender gap among professionals with AI skills closely matches the gap we see among graduates of college and masters programs with related degrees, says Maria Klawe, the president of Harvey Mudd College. Typically, about 20% of U.S. undergraduate computer science degrees go to women. When you look at domestic PhD programs like statistics and math, roughly 30% of students are female.

The same can’t be said at Harvey Mudd, where more than 50% of computer science graduates are women. Klawe attributes the school’s success to subtle tweaks to their program. A majority of incoming freshman who arrive on campus with advanced computer science skills are male, so Mudd created a section for students who lacked previous experience. The college has also made an effort to emphasize computer science’s potential to make a social impact. This has encouraged more women to apply, Klawe claims.

“We are in this transition time where we either really work hard to make sure that we have a diverse set of people thinking about these skills, or we miss this moment and it’s hard to catch up,” says Klawe.

Creating ‘an armor’

The challenges don’t simply end after graduation, though. Many women who complete their education in AI-related degrees do not pursue careers in the field. The way jobs in machine learning and artificial intelligence are marketed may have something to do with it.

In an analysis of more than 78,000 job listings for software engineers, those referencing “machine intelligence” are by far the most likely to use phrases proven to be masculine in tone such as "exhaustive," "enforcement" and "fearless." In turn, these listings also have the most male applicants, according to Textio, an augmented writing platform that uses AI to surface language patterns that attract or deter people to job descriptions and recruiting emails.

To highlight women who enter AI and machine learning careers and encourage them from leaving, several organizations have launched networking events. Nikita Johnson founded RE.WORK, a London-based events company that convenes academics and experts working in AI, about six years ago. When they first started organizing events, Johnson says about 30% of the 400 to 800 attendees were female. That figure is now closer to 40% for their larger summits and they convene nearly 100 women working in AI a few times a year for smaller dinners. 

For Johnson, it’s critical to shine a light on role models in AI that other women can look up to.

“There are a lot of different female role models working in artificial intelligence. A shift is only going to happen when women that are still in school can see them and think [artificial intelligence] is a valuable opportunity for them.”

Such was the case from an early age for Kieran Snyder, the founder of Textio. The daughter of an engineer who makes radar systems for spaceships and a writer, her parents sat at the intersection of science and language processing. She studied math and linguistics as an undergraduate at the University of Pennsylvania before earning a PhD in linguistics and cognitive science.

After nearly 10 years at LinkedIn-parent Microsoft in various project manager roles, 44-year-old Snyder set her sights on starting her own AI company.

“The fact that my dad and mom had the expectation that I could do anything I wanted was a pretty big deal…. I think it created an armor for me,” she says.

‘Need the full range of human perspective’

Many members of the AI community are concerned about the ethical implications of building an AI workforce that is predominately male. That’s, in part, why Snyder has been conscious from the beginning of building a diverse team. Half of Textio’s roughly 100 employees are female and 30% identify as Black or Latinx.

Snyder says she’s been successful at building a diverse team because she started early and looked in different places to find talent.

“If you have a company that is in our size and stage and it is all white guys, it is harder to convince that next person to break into that environment,” she says.

Roy Bahat and Karin Klein, both founding partners at venture capital firm Bloomberg Beta and investors in Textio, say they have seen a steady increase of women starting businesses in recent years. Ensuring that a startup has a diverse founding team is now critical to their decision to invest — particularly when it comes to companies working with AI.

“It is crystal clear that in every single AI company, the life experience of the founders and technical team drive the thinking of the technology they are building,” says Bahat. “The idea that the algorithms are objective and unbiased is complete nonsense. We are going to need the full range of human perspective building these technologies.”

Amazon’s Kroll explained these unintended biases through the lens of the software she helps program: Alexa. In order to ensure that Alexa can understand a wide range of voices that it interacts with, it’s essential that her team has a diverse set of training data. It is not just the words people say, she says, Alexa must be able to recognize the many different accents and pitches that people use when they speak. So, if her team only used male voices to train Alexa, for example, it would be more challenging for the device to assist women.

Ethical implications aside, Harvey Mudd’s Klawe pointed out that we need more women to enter AI simply to meet the ballooning demand for such talent: Only 22,000 people in the world are experts in machine learning or artificial intelligence, according to data from Montreal-based startup Element AI, a figure that won’t keep pace with increasing demand.

To meet that demand, more women like Kroll will need to enter the field; not just accidentally, but with intention.

“I would want to tell young women that just because you have other interests besides technology or computer science, that absolutely doesn’t mean you need to choose either field over those passions,” says Kroll. “Computer science is a way that you can apply your passions to many different problems.”

Photo Captions: 1) Experts in artificial intelligence attend a REWORK event in Houston, 2) Kieran Snyder speaks with Textio co-founder & CTO Jensen Harris

Kristin Keveloh and Rachel Bowley conducted the original analysis of LinkedIn’s data