弥合数字鸿沟：利用人工智能促进亚太地区的性别平等（英）

KEY POINTS Bridging the Digital Divide:Harnessing Artificial Intelligence forGender Equality in Asia and the Pacific •Artificial intelligence (AI) israpidly reshaping economiesand societies, with globalrevenues projected at$407 billion by 2027. ForAsia and the Pacific, thisunderscores the urgency ofinclusive strategies to ensurewomen’s participation in theAI ecosystem. Byeongjo KongDigital Technology Specialist(AI and Data Analytics)Digital Sector OfficeSectors Department 2ADB Jinha Kim Gender Specialist (Climate Change)Gender Equality DivisionClimate Change and SustainableDevelopment DepartmentAsian Development Bank (ADB) •Women hold only 30%of professional roles and12% of research positionsglobally, while also facingunequal digital access. BiasedAI systems often restrictwomen’s opportunitiesand reinforce stereotypes.Structural barriers and“gender-blind” nationalAI strategies further limitwomen’s digital participationand leadership. CONTEXT Artificial intelligence (AI) is rapidly transforming economies and societies across Asiaand the Pacific, with global AI market revenues projected to reach $407 billion by 2027.1Yet, most of these investments remain concentrated in the United States—the largestand fastest-growing AI market—followed by the People’s Republic of China and theUnited Kingdom.2This concentration underscores how the benefits of AI areunevenly distributed. Similar imbalances are visible within the AI workforce itself, wherewomen remain significantly underrepresented. However, without deliberate effortsto ensure inclusion, this digital transformation risks reinforcing and widening existinggender inequalities. •Case studies show thatinclusive AI design,deployment, and governancecould reduce inequalities;expand women’s economicopportunities; improve accessto services; and enhancesafety in digital spaces. Women remain significantly underrepresented across the AI ecosystem—from systemdesign to leadership. This lack of diversity affects not only fairness, but also the quality,safety, and inclusiveness of AI systems. •Priorities to harness AI forwomen’s empowermentinclude integrating genderequality in AI policies andoversight; codesigninginitiatives with gender-inclusive tools; expandingfunding for women-led AIsolutions; and strengtheningwomen’s digital literacy,STEM participation, andleadership pipelines. To ensure that AI serves as a tool for empowerment and not exclusion, it is criticalto understand how gender disparities manifest throughout the AI life cycle. The nextsection explores the AI–gender nexus by identifying key challenges and risks across thedesign, deployment, use, and governance of AI. 1MarketsandMarkets. Artificial Intelligence.2D. Liberto. 2025. Which countries are investing most in AI?Investopedia. ISBN 978-92-9277-473-8 (print)ISBN 978-92-9277-474-5 (PDF)ISSN 2071-7202 (print)ISSN 2218-2675 (PDF)Publication Stock No. BRF250419DOI: http://dx.doi.org/10.22617/BRF250419 THE AI–GENDER NEXUS: CHALLENGESAND RISKS technology, engineering, and mathematics (STEM). Globally, only30% of AI professionals are women, roughly 4% higher than it wasin 2016, and only 12% of AI researchers are women.4In Asia andthe Pacific, women represent 23.9% of STEM researchers, belowthe global average of 29.3%.5At the industry level, women holdonly 23% of senior positions—and only 8% of senior technicalroles—at the 50 largest Southeast Asian technology firms.6Women’s participation declines even further in senior roles,conference authorship, and technical leadership.7 Gender inequality in AI systems is embedded throughout thelife cycle—from who builds AI, to who accesses it, and who regulatesit. These disparities emerge across four key stages—design,deployment, use, and governance—that reinforce existinginequalities. For instance, a study analyzed 133 AI systems acrossdifferent industries and found that about 44.2% (59) of them showedgender bias, and 25.7% (34) exhibited both gender and racial bias.3 The underrepresentation of women in AI development influenceswhich challenges are prioritized, what data are used, and howsystems are evaluated. This can result in AI tools that reflect,reproduce, and reinforce gendered assumptions. For example,large-language models (LLMs) such as GPT-2 and Llama 2have been shown to associate women with domestic roles andmen with professional roles. A key observation was that whenmodels were prompted to complete sentences beginning witha mention of a person’s gender along with sexual identity, LLMsnot fine-tuned with human feedback—in technical terms Design and Development: Biased FoundationsAI systems are shaped by the data, design choices, and assumptions of their creators. Without inclusive approaches fromthe outset, these systems risk perpetuating bias and creating harm. A key driver of such bias is the systemic underrepresentationof women in AI research and development. Figure 1 shows thepersistent underrepresentation of women in AI and in science, Bridging