The Women Breaking Barriers series celebrates female leaders who have risen to prominence in traditionally male-dominated roles. Engineering and computer science have acute shortages of women, especially in upper management positions. OnlineEducation.com interviewed three top-level professionals with experience in artificial intelligence and deep learning. These accomplished women share how they got into AI and their advice for aspiring female leaders in this field.
Heather BowermanHeather Bowerman is the CEO of DotLab, a medical company which developed the world’s first non-invasive test for endometriosis. Prior to cofounding DotLab with Dr. Hugh Taylor (Chair of Obstetrics, Gynecology and Reproductive Sciences at the Yale School of Medicine), Ms. Bowerman led business operations at Enlitic, a Silicon Valley company which uses deep learning to assist doctors in interpreting diagnostic images. Notably, she was recognized among the world’s top 35 innovators by the MIT Technology Review and one of the 100 most intriguing entrepreneurs by Goldman Sachs. She earned her degree in bioengineering from UC Berkeley and served as a teaching and research fellow at Harvard University.
Dr. Ahna GirshickDr. Ahna Girshick is a senior computational research scientist at AncestryDNA. She is an expert in perceptual neuroscience, data visualization, and machine learning. Her research has been published in Nature Neuroscience and SIGGRAPH, and her work has been featured in The New York Times, WIRED, Rolling Stone, and NPR. Also, she used her data visualization expertise to create an interactive music app with musicians including Philip Glass and Björk—work which was featured at the New York MOMA and the San Francisco’s Contemporary Jewish Art Museum. She earned a master’s degree in computer science with a minor in cognitive science from the University of Minnesota and a PhD in vision science from the University of California, Berkeley.
Dr. Rachel ThomasDr. Rachel Thomas is an expert in machine learning and the founder of fast.ai, a non-profit organization which makes artificial intelligence more accessible to people all over the world. In addition to a large resource library, fast.ai provides free online classes, including “Practical Deep Learning for Coders,”which has enrolled over 100,000 students. Her writing has been translated into multiple languages and she served as an early engineer at Uber. She also is a researcher and assistant professor at the University of San Francisco, where she developed the first open access deep learning certificate program. She earned her PhD in mathematics from Duke University.
“Twenty years ago, a girl could be a secretary, a school teacher…Now have come the big dazzling computers—and a whole new kind of work for women: programming.” (Lois Mandel, Cosmopolitan 1967)
Without women’s inventions, would we have modern computers?
In spite of deeply rooted prejudice and discrimination, women have played an indelible role in the development of these machines. Mathematician Ada Lovelace was the first person to develop a computer algorithm and imagine an “Analytical Engine,” a device with capabilities beyond simple calculations. Rear Admiral and computer scientist Grace Hopper is responsible for the development of COBOL—a pioneering, high-level programming language still in use today—due to her commitment to using more natural English syntax.
Hopper famously stated, “Programming requires patience and the ability to handle detail. Women are naturals at programming.” In fact, many of the world’s first computer scientists were women, and IBM even had a female VP in 1943. These days, however, computer science is overwhelmingly male.
One subfield of the discipline—artificial intelligence—has become instrumental in shaping the modern world. Researchers within AI often refer to ‘machine learning’ and ‘deep learning,’ two related concepts.
Machine learning is when an algorithm learns from a dataset and uses that knowledge to make decisions and inferences. Deep learning takes this one step further: the machine can use its own experience to evaluate the quality of its findings and predictions.
AI now touches every industry and has a wide range of uses. Dr. Thomas marveled at how students in her free fast.ai courses are applying what they have learned all over the globe. Here is just a handful of ways AI is being applied in various arenas:
Of course, with each advancement building on what came before, the ethics and experience of AI’s creators become evermore important. If a large majority of AI’s creators are white men, what biases get baked into the algorithms? The homogeneity of AI’s architects has already created problems, but as Heather Bowerman, Dr. Ahna Girshick, and Dr. Rachel Thomas demonstrate, there is nothing natural about the male-dominance of this field.
The question is: are we going to build technologies that calcify the status quo, or are we going to design tools for the world we want to create?
“I think [the gender disparity] is really pernicious because people working in tech often pride themselves on being super logical and rational. That can make it even harder to recognize our biases.” (Dr. Rachel Thomas)
The shortage and exodus of women working in tech is well-documented, and there is evidence that the male-dominance of artificial intelligence is even more pronounced.
While there is little data about the percentage of women working specifically in AI, most estimates fall between 10 and 15 percent:
Figures for women earning CS degrees and working in computing are only marginally more promising. The National Center for Women and Information Technology (2016) reports that only 18 percent of software developers and 21 percent of computer programmers are women. And according to the National Science Foundation Science & Engineering Indicators (NSF 2018), the share of women earning degrees in computer science and engineering has decreased:
The NSF (2018) found a similar trend among women employed in computing, as well as an alarming salary gap between men and women working in science and engineering (S&E) occupations:
Median Annual Salaries, Highest Degree-Holders Working Full-Time in S&E (NSF 2018):
Median Salary, 1995
Median Salary, 2015
It is important to note that these salaries differ by degree achieved, industry, and experience. Low-paying degree areas within S&E tend to have higher concentrations of women and minority groups. For example, the 2015 median salary among the highest degree-holders in computer and mathematical sciences—a relatively male-dominated field—is $97,000. In life sciences, it’s only $62,000.
High salaries in computing are actually a relatively new phenomenon. In fact, The New York Times and The Atlantic have both summarized research showing that the salaries and prestige in programming only began to rise when men entered the profession. The opposite phenomena has been observed in teaching, which was male-dominated in the early 1800s; the salary and prestige actually fell as more women entered the field.
The bottom line: women’s work is devalued, both monetarily and culturally. Of the 30 top-paying jobs in the U.S., 26 are male-dominated. And the cultural biases are so deeply rooted that one study showed that children assigned higher status to jobs when images were depicted with male workers as opposed to female workers.
“Most people are making products for problems they’ve experienced and if that’s just a narrow slice of the population, you won’t create solutions for different types of groups.” (Dr. Rachel Thomas)
Artificial intelligence is an expression of its creator. Imagine a world where this technology is designed for and by one group of people: This is a world in which:
Actually, all of this has already happened.
Without the thoughtful intervention of a more diverse group of stakeholders, the biases in AI will not only reflect the existing misogyny and racism of our world; it will intensify them.
Kathleen Siminyu, the head of data science at Africa’s Talking, puts the issue perfectly: “While it may seem that limiting the role of humans in such decisions would limit subjective biases, these machine learning models learn from data that are, in many cases, representative of existing societal biases.”
Let’s take the example of employment: when high-paying jobs in leadership are more likely to be advertised to men, those men are more likely to apply and get those jobs. As a result, more men than women will hold these positions, feeding new data back into the system. This amplifies the problem and codifies the existing inequalities. It’s alarming to extend this analogy to other areas such as criminal justice, which already is very racially biased.
Artificial intelligence should reflect the variety of people who use it. Not only does this make for a better system, but there’s abundant evidence that diverse teams are smarter; make better decisions; are more likely to innovate products; are more productive; and even make businesses more profitable.
A 2015 McKinsey study on 366 public companies found that those in the top quartile for racial and ethnic diversity were 35 percent more likely to outearn their industry median, and those in the top quartile for gender diversity were 15 percent more likely.
“In my education, it was largely male. I was a computer science major and I was the only female in class, or one of two. That was pretty common and I got used to it.” (Dr. Ahna Girshick)
Why aren’t there more women working in computing and artificial intelligence? As Dr. Thomas points out, the lack of women is more than a simple pipeline issue: the foundations of the culture in tech academia and companies are often not welcoming places for women.
Ignoring the fact that many of the first programmers were women, a University of Washington lecturer recently published an article titled “Why Women Don’t Code.” Defending engineering’s enfant terrible, the ex-Google employee James Damore, Stuart Reges argues that women’s concerns about working in tech are a “false narrative” and claims that overt discrimination has been eliminated.
On the contrary, 74 percent of women who work in computing (e.g., software developers, programmers, computer scientists) report gender-related discrimination. Many women leave the field due to a lack of opportunities for advancement, and only 15 percent of tenure-track CS professors in North America are female. Given that both men and women tend to wrongly view men as more competent in mathematics and computing, it’s no surprise that the share of females working in these fields has declined recently.
There’s evidence that these biases can extend to hiring decisions. In a randomized, double-blind study at Yale University, researchers submitted identical resumes with male or female names to science faculty. The researchers found that the faculty members were not only more likely to select the male candidates, but they also offered them higher salaries than the females with the exact same qualifications.
These are only a few of the forces creating headwinds for women who want to succeed in artificial intelligence. A full examination of the historical injustices, stereotypes, sexual harassment, and socioeconomic factors, among other problems, is beyond the scope of this article.
“Tech is not going to stay in the lead in the United States unless gender diversity gets materially better. It’s just not.” (Tim Cook, CEO of Apple)
Attracting more women to work in artificial intelligence takes more than creating a campus group or offering mentorship opportunities. Recognizing the limitations of homogeneity in AI, there are a few colleges and employers which have been successful in increasing their diversity.
In 2018, Harvey Mudd College (HMC) awarded bachelor’s degrees in computer science to a class that was 56 percent women—the highest-ever proportion at the school. Maria Klawe, previously the dean of engineering at Princeton, has served as president of HMC since 2006. She helped introduce several changes which have dramatically increased both gender and ethnic diversity:
In 2017, Worcester Polytechnic University, a tech-heavy school, enrolled a freshman class that was 43 percent women—a 9 percent increase over the year before. To strengthen its share of female enrollees, the school had reallocated $1 million of its aid to supporting top female applicants.
In 2016, the undergraduate enrollment in computer science at Carnegie Mellon University was 48.5 percent women—roughly triple the national average. CMU cited several factors for this growth, including a coordinated effort between leaders at all levels; recruitment initiatives aimed at middle- and high-school students; and making the faculty more diverse. The provost and chief academic officer Farnam Jahanian noted, “We recognize that cultivating diverse perspectives and promoting inclusion will breed the intellectual vitality essential for the health and progress of our campus community.”
Despite these gains, there are still challenges at CMU. In mid-August, two popular CS professors, Lenore and Manuel Blum, submitted their resignations, which will take effect in 2019. Ms. Blum founded the Project Olympus business incubator and the campus group “Women@SCS.” Mr. Blum won the prestigious Turing Award. In an email blast to the CS department, the couple cited “sexist management” and “professional harassment.” The details are unclear, but their departure is a blow to CMU’s diversity initiatives.
As the Blums illustrate, having a supportive environment can make all the difference. The American Association of University Women summarizes some of factors that make women working in engineering, CS, and AI more likely to stay in their jobs. Employers should:
Notably, there have been recent gains in female-led professional advocacy and policy making. MIT Technology Review points out that three rising AI groups are all led by women:
“My interests are in using technology to make us more human. How can we use technology to help us connect to each other more? How can we use technology to spark our creativity, to educate ourselves, to create better societies? Technology for the greater good—it’s for humanity at all different levels. It’s for one person and for our whole society. That’s what inspires me.” (Dr. Ahna Girshick)
One of the most striking facts about Heather Bowerman, Dr. Ahna Girshick, and Dr. Rachel Thomas is that they all turned to artificial intelligence for its humanitarian applications. Machine learning and deep learning have a tremendous potential to improve societies.
Drawing from the interviews with these three successful professionals, here’s advice for women aspiring to work in artificial intelligence.
Let your interests—not your salary—guide your choice of industry. As with other lines of work, to avoid burnout, seek out personally meaningful opportunities. Fortunately for women in AI, there are applications in virtually every economic sector. Dr. Thomas pointed out, “A practical approach is to get in there right away and start working on problems you’re interested in!” Ms. Bowerman echoed this sentiment, stating, “There’s a shift toward support for mission-driven technology companies, including topics that might appeal to women uniquely.” Find a way to make your skills in deep learning address a problem you care about.
Hire a boss who already treats women well (or become your own). Rather than getting hired by a company, set your standards and hire your own boss. In one of many awesome posts she’s written about diversity in tech, Dr. Thomas explains how companies that value women conduct themselves. They do more than donate to organizations or put women on AI panels; they challenge women with opportunities to grow their skills, show them how to earn promotions, praise their accomplishments in meetings, and pay them fairly. Women and other underrepresented groups should seek out employers who recognize that having diverse voices shaping their technology is better for their business.
Show confidence in your skills and your coworkers will follow. As mentioned, a study found that women with eight years of experience in programming demonstrated the same confidence-level as men with one year or less. Do not discount your knowledge and never undersell yourself.
Support women and other underrepresented groups. A large part of breaking the cycle of women’s underrepresentation in tech and AI is to help others. Pay it forward and emulate your own mentors or become the inspiring, selfless person you wish you’d had.
Tell your mentors and sponsors exactly what you want. Do not be shy with your goals and ambition. Be clear with your managers and others about where you’re headed.
Make yourself irreplaceable by constantly upgrading your skills. Learning how to code in Python is a solid start, but especially in the dynamic field of AI, the importance of continuing education cannot be overstated. Reflecting on her college experience, Dr. Girshick shared, “We didn’t have Coursera and Udacity. They make it so easy to get started in machine learning without having a computer science degree.” For those with some programming experience, Dr. Thomas added that fast.ai provides a free course online called “Practical Deep Learning for Coders.”
Never forget the importance of diversity to AI’s success. Given the abundance of evidence that diverse teams are smarter, more productive, and make better products, one’s unique perspective is an invaluable asset. Ms. Bowerman conveyed this perfectly: “Know your worth in these conversations. We have a lot of white men in AI and we know that getting deep learning into the hands of a more diverse group—women and people of color—can help bring a new point of view and new solutions to problems.”
The future of artificial intelligence is too important to be left to a small, non-representative faction of the population. The field’s early blunders illustrate the dangers of homogeneity in decision-making.
Consider a harrowing example from a different type of technology: cars. Until 2011, all crash test dummies were based on average male bodies. The result: female drivers are 47 percent more likely to be seriously injured in car accidents. Now imagine how this myopic thinking could impact the future of self-driving vehicles or any other AI product.
Grace Hopper (a.k.a., Grandma COBOL, named after the early programming language she helped develop), famously said, “Humans are allergic to change. They love to say, ‘We’ve always done it this way.’ I try to fight that.”
In the spirit of Hopper, let’s change where we must, and let’s ensure that our machines don’t replicate our mistakes and biases.