Recording data about the AI workforce assuming that men and women are represented equally is bad practice
Woman in AI and data science, in general, might have suspected this for a long time. At this point, the anecdote that men apply to jobs when they only meet part of the job criteria is well circulated as a piece of advice to women seeking employment – If a man thinks he can do it, so can you, may have heard countless times while looking for data science job.
Whether you believe this anecdote or not, the fact is that women in AI and data science were found to have higher formal education levels than men across all industries that employ an AI workforce, from Tech, Finance and Corporate Services, to Healthcare and Non-Profits. While the qualification gap looks fairly inconspicuous on average, with 6% more women than men holding a graduate degree or above, this figure shoots to 13% for positions in the C-suite (company executives such as CEO, CTO, etc).
Women Self-Report Fewer Skills Than Men:
Using data from profiles of AI workers, the researchers showed that women are likely to self-report 30 or fewer skills on their profile. Less than 12% of women report over 45 skills, compared to over 16% of men. Because women are on average more qualified than men by a significant margin, researchers believe the cause might be lower confidence among women.
We already know that “impostor syndrome” affects women in AI and data science, particularly women of colour, disproportionately. In fact, impostor syndrome was first described in a cohort of 150 high achieving women. There is likely a confidence issue at play, even 40 years after the phrase “impostor syndrome” was coined.
Women are Over-Represented in Industries and Positions That Require Fewer Technical Skills:
If thought this couldn’t get any worse, hold onto your seat: women in AI and data science are actually over-represented in industries that require a lower de facto level of technical skill. Chiefly, the gender gap in AI for Healthcare, Education and Non-Profits ranges from 1.2 to 1.4 approximately, where 0 means no women and 1 is an equal women-to-men ratio. When looking at software engineering, hardware and networks, this ratio drops to 0.4 and below.
Furthermore, they showed that women are more likely to hold jobs in data cleaning, preparation and analytics. Meanwhile, men are more likely to hold higher-paying jobs, such as machine learners or software engineers. While the gender pay gap per se is largely resolved, the lack of appropriate representation in high paying roles still comes up as an issue.
Only 8% of Participants on Stack Overflow are Women:
Where would we be as programmers and data scientist if we did not have widespread community engagement on platforms such as Stack Overflow? As reported, a meagre 8% approx. of active users on Stack Overflow were women. Take a moment to sit with that information.
Why we can’t find women in male-dominated communities because of: gender bias. Make those three words: widespread gender bias, not only talking about online gaming communities, known for their hostility against players who they can identify as women.
Need to Ditch the Interventions That Don’t Work:
Data shows we can’t just wish gender bias away. We have to act. However, interventions such as mandatory diversity training and women-only conferences do not produce the results that we hope for. In fact, they make things worse. In 2016, data gathered showed that five years into diversity training for team managers, ranks of black women shrank by 9%. Ranks for Asian men and women shrank by 4% approx., while white women, black men and Hispanics saw no change in ranks at all.
Other data suggests that diversifying interventions works better. Instead of only monitoring diversity at recruitment, or only hosting workshops targeted at certain categories of employees, companies need to (quite literally) read the room and find a diet that suits their staff.