Who Wins in the Age of AI?

Welcome to the 4th Industrial Revolution, where AI makes and augments life-changing decisions at scale. Perhaps most eerie about this impending singularity is the absence of women who will influence it: we have a 28-point gender gap in the AI talent pool.

Unlike previous industrial revolutions where women were still absent, the 4th Industrial Revolution magnifies the ramifications of gender inequity and we need to act now to close the 28-point gender gap in AI.

Why We Need Gender Equity in AI (Now)

The consequences of gender inequity in the 4th Industrial Revolution, specifically in AI, hinge on the premise that automation lies at the center of 4th Industrial Revolution technologies.

Since algorithms learn based on past data, they will learn that women are not CEOs, nor are they elected officials, nor are they AI professionals.

When engineers, largely unaware of their biased data, feed algorithms learning material, they not only program these biases into the system, they perpetuate current inequities because AI hardens the bias into future decision-making. (Remember Amazon’s recruiting software that favored male candidates?)

A secondary premise, one that relates primarily to the US, is that women are the breadwinners in 40% of US households with children, and an additional 31% rely on moms’ earnings for their economic well-being. Maintaining the gender gap in AI risks leaving behind 47% of the US labor base, women, as well as their children — our future labor force.

To this extent, the gender gap threatens the vitality and size of businesses’ labor pools, which continue to diminish alongside “diversity dividends,” or the financial gains associated with increased workforce diversity.

A study across 4,161 companies in 29 countries found that every 10% increase in gender equity raises revenue by 1–2%. That’s real money being lost to AI’s gender gap.

We don’t need more facial recognition systems that fail at accurately identifying half of the world’s population. What we need is gender equity in AI so that we can minimize the economic and societal threat of algorithmic bias from the start.

The Gender Gap in AI: Why Is This Happening?

Women make up 22% of the world’s AI practitioners, yet they are 50% of the population. The US, India, and Germany hold the highest concentrations of AI talent. And while the US’s AI talent pool is 23% female, just above the global average, Germany’s is a paltry 16%. Italy, Singapore, and South Africa boast the smallest gender gaps in AI. Collectively, women make up 28% of these three countries’ AI talent pools.

This data underscores the prevalence of AI’s gender gap around the globe.

To explain why the gap exists, we need to approach the issue from the perspective of skills acquisition.

Between 2015 and 2018, women and men gained primary AI skills at nearly equal rates. The raw number of women and men with AI skills, however, doesn’t follow the same trend. Between the same period (2015–2018), women’s representation among AI talent fluctuated from 21 to 23% of the total talent base.

So even as the rate of women and men acquiring AI skills increased in tandem, the hardened gap between male and female talent persists. It’s a problem that, according to the World Economic Forum, “will require focused intervention.”

The first place to focus our attention is the skills gap within AI talent.

The AI Skills Gap Is a Reality, Too

A range of proficiencies exist under the domain of AI, and it’s here where we find additional gender gaps.

Female AI professionals are more likely to have information retrieval skills as well as skills in natural language processing and data structures. Male AI professionals more commonly have machine learning skills as well as knowledge about emerging technologies (neural networks, computer vision, deep learning). For example, 47% of male AI professionals on LinkedIn list machine learning as a skill as opposed to 40% of female AI professionals on LinkedIn.

A concerning implication of the AI skills gap shows up where people are employed and how they are compensated.

Compared to their female counterparts, male AI professionals hold more lucrative, senior-level positions such as the head of IT, head of engineering, and CEO.

Women, on the other hand, are more likely to work as researchers, information managers, teachers, and data analysts.

At tech giant Google, women occupy only 25.7% of tech roles; at Apple and Facebook, it’s 23%; and at Microsoft, 19.9% of tech roles go to women. More striking is the gender break-down of Google employees listed as working on machine intelligence, where a mere 10% are women.

These gender gaps, whether they are among AI professionals, their skills, or their compensation, point to an underlying systemic issue that begins with STEM education.

To solve for the 28-point gender gap in AI, we must intervene at an early stage, before people enter the workforce.

Strategic intervention at this stage would fill the pipeline of AI talent from the start and bolster the representation of women in STEM-related fields.

It would, in effect, normalize women’s presence in the technology sector so that no longer would each woman working in AI have, on average, 3.5 male colleagues.

What the Gender Gap in AI Means for You, for Everyone

By 2022, 85% of AI projects will generate inaccurate reports as a result of algorithmic bias — whether that bias comes from the data or the engineers themselves. By failing to close the gender gap in AI, we are choosing to amplify our biases and cement them into the future. This comes at a steep moral and economic cost.

At the business level, the gender gap in AI prevents organizations from tapping into rich talent pools and reaping the financial gains from employing diverse talent. At the societal level, the gender gap jeopardizes the personal freedoms of 50% of the world’s population.

Analysts already predict strong use cases of AI in applications such as autonomous vehicles, healthcare diagnosis and treatments, education, journalism, and customer service. As the technology becomes increasingly more mainstream and commercialized, the importance of closing the gender gap becomes all the more urgent.

According to Anima Anandkumar, professor at the California Institute of Technology and former AI professional at Amazon, “Diverse teams are more likely to flag problems that could have negative social consequences before a product has been launched.”

We Decide Who Wins in the Age of AI

It is within our power to decide who will win in the age of AI. The decisions we make today about how we manage the talent pipeline will decide the impact AI has on our future.

• Who will we hire?
• Who will we promote?
• How will we evaluate them?
• How much will we pay them?

Here are three recommendations to bring us closer to gender equity in AI.

1. Standardize digital literacy in all K-12 education. We need to address systemic gender gaps early. Research shows that children as young as six begin to form gender biases, such as which gender is smarter in STEM subjects. These biases influence the educational and career trajectory of our boys and girls. Mandatory digital literacy, including coding, in K-12 education, helps mitigate the gender gap before it starts.
2. Remove barriers that keep women from thriving in computer science programs. Harvey Mudd, one of the world’s most elite STEM colleges, increased the percentage of women graduating from their computer science program by 380% in seven years by taking concrete action on this front. Among the measures they took include placing women in leadership positions around the school and requiring all first-year students to take a redesigned intro course that emphasizes practical uses for programming along with teamwork.
3. Keep women in STEM by ensuring equitable opportunities in the workplace. That means making sure skills training targets women and men equitably; it means valuing (i.e. paying and promoting) women and men equitably; and — since half of women in STEM positions report sexism at work — it means having resources in place to address and prevent workplace harassment

Ultimately, the choice is ours: will we program our biases into the future world or will we use AI to engineer them out?