For women, hurdles are everywhere. Despite the critical role women play in societies, unequal access to education, loans, jobs, healthcare, technology, and political discourse are commonplace — and worsened by COVID-19.
Technological innovations like artificial intelligence (AI) promise to identify and close these gaps through claims of a more data-driven, objective approach, but ironically pose another hurdle for women. Often, these digital systems inadvertently carry the same old analog gender biases.
AI Gender Inequity Example
Imagine that you run a small woman-owned business from home. Excited by its early success, you apply for a loan to hire staff and get more space. But you’re soon dejected — every bank approved you for a smaller loan than requested, while a friend got his loan request approved in full. You’re surprised — he has similar assets and savings as you. The only obvious difference is that you’re a woman.
You learn that your credit application was evaluated not by a person, but by a machine. Banks use AI technology to assign a creditworthiness score to applicants. Using a machine learning algorithm, the tool ‘learns’ patterns of behavior associated with higher creditworthiness based on previous applicants’ data and their associated repayment.
Historically more men received loans than women, so the algorithm determined that men were more creditworthy. Although banks turned to AI to be more objective and equitable in lending, the results were actually the opposite.
Gender Inequity in Artificial Intelligence
This is not a hypothetical example — there are now-famous instances of the unintentional consequences of AI, such as automated resume screeners rejecting women, facial recognition disproportionately failing for women, and algorithmic credit-scorers ranking women lower than men.
As AI tools are being tested and used in developing economies to derive insights and gain efficiencies across sectors — and as we rely more and more on them to give loans, diagnose diseases, triage medical care, and respond to humanitarian crises — we must work to prevent them from discriminating. There is an opportunity and urgency to optimize for innovative and equitable AI — especially in developing countries.
Development actors are taking steps to address disparities, for example, using AI to close gender-related data gaps in child marriage. Through the WomenConnect Challenge, USAID is beginning to tackle algorithmic gender bias in lending and is committed to taking action on the fair development and use of AI more broadly, working with partners to create a report and online course to better integrate AI fairness in development.
But we know there are many more ways that bias manifests in AI. Complex contributors to these harmful outcomes can include unrepresentative datasets, largely male data science teams, cultural norms around gender, and local policies and practices around data, among many others.
Gender equity is critical to achieving AI fairness, and as we work to build an agenda for action, we want to hear from you. In recognition of Women’s History Month, we want to shine a spotlight on inequitable gender outcomes related to AI in the developing world, and everyone’s help and collaboration is needed.
The development community is committed to gender equality and women’s economic empowerment. We must ensure that using AI does not reverse the substantial gains of the past decades. Together, we can raise awareness about gender inequity in AI and take action to change course.