The Women in Data Science and AI project within the public policy programme at The Alan Turing Institute conducts
Explaining the science
Why do so few women enter data science and AI professions?
There is a troubling and persistent absence of women employed in the Artificial Intelligence (AI) and data science fields. According to the World Economic Forum, women make up only an estimated 26% of workers in data and AI roles globally. This issue begins when, from a young age, girls can feel discouraged from pursuing STEM subjects. In 2012, the OECD surveyed the UK’s 15-year old students and found that 41% of the girls agreed with the statement ‘I am just not good at mathematics’, while only 24% of the boys agreed. In 2015, the OECD surveyed the country’s 15-year olds again and found that 4.6% of the boys expected to work as ICT professionals at age 30, while only 0.5% of the girls had the same expectations for themselves. Data-driven research can help us understand how biases in media coverage and public perceptions and stereotypes influence girls’ lack of confidence in their own abilities, and explore how educational systems and other factors can discourage girls from engaging with mathematics and science. This ‘pipeline problem’, however, is only one part of the issue.
Why, once they enter these professions, do many women leave?
Women working in AI and data science in the tech sector have higher turnover and attrition rates than men. Like other studies, the Women in Data Science and AI project at The Alan Turing Institute has found that women spend more time than men in most industries apart from the Technology/IT sector, where they spend almost a year and a half less (Young, Wajcman and Sprejer, 2021). Similarly, the US National Centre for Women and Information Technology found that women leave technology jobs at twice the rate of men (Ashcraft, McLain and Eger, 2016).
The ‘chilly’, unwelcoming climate of the tech workplace, both physically and online, is a central contributor to the high attrition rates of women away from the data science and AI professions, as well as the differentiation of AI professional’s career trajectories by gender (e.g. Hicks, 2017). This climate can include gender pay gaps, slow career progression for women, sexual harassment, male-dominated office culture and gender bias in hiring, all of which discourage women from continuing their careers in the fields.
Which interventions work to increase the number of women in data science and AI?
Existing research highlights that many popular interventions meant to increase diversity, such as diversity training or women-only conferences, are of limited effectiveness. Rather, successful models have been provided by the US universities Carnegie Mellon and Harvey Mudd, which have dramatically increased the participation of women in their computer science departments through strong commitment from senior management. For example, Carnegie Mellon increased the number of women from 7% in 1995 to 42% in 2000.
A key aim of the research we undertake in the Women in Data Science and AI project is to identify policy interventions that encourage more women to pursue data and AI careers. Recommendations thus far include the need for proactive steps to ensure the inclusion of women and marginalised groups in the design and development of machine learning and AI technologies, which includes disclosure and scrutiny of the current available data and gender composition of companies’ technical, management and applied research teams.
How does the gender deficit shape both the research agenda and the applications of digital technologies?
The under-representation of women and marginalised groups in data science and AI, alongside algorithmic and data biases such as the gender data gap (Criado Perez, 2019; D’Ignazio and Klein, 2020), is not only an issue of social and economic justice, as well as value-in-diversity. It leads to the encoding and amplification of bias in technical products creating a dangerous feedback loop. A growing strand of research documents the fact that AI and machine learning systems can exhibit biases (e.g. Buolamwini and Gebru, 2018; Noble, 2018), and AI products are increasingly making headlines for their discriminatory outcomes. Growing women’s participation and tackling structural discrimination is essential to ensuring their perspectives and priorities will inform the insights that data scientists will generate, the AI systems that they will build, as well as the research agendas that they will define. It is crucial that we get ahead of this now, before flawed technologies become irreversibly integrated into the fabric of society.
Visit the dedicated Women in Data Science and AI Hub for relevant resources, news and our current project research.