The sixty-seventh session of the Commission on the Status of Women (CSW67) convened against a backdrop of intersecting global challenges—pandemic aftershocks, accelerating climate change, inflationary pressures, political authoritarianism, and armed conflicts. Within this complex environment, rapid advances in artificial intelligence, exemplified by systems such as ChatGPT, are reshaping industries and daily life in ways that remain difficult to forecast. Yet persistent inequities endure. The World Economic Forum has projected that achieving global gender equality will take another 132 years, with recent crises erasing prior gains.
The CSW67 priority theme of “innovation and technological change” underscores the urgency of closing the gender gap in technology. Over two decades of data reveal entrenched underrepresentation of women in STEM disciplines—particularly computing, IT, engineering, mathematics, and physics. Economists link this disparity to unequal human capital development, heavier domestic responsibilities, and workplace discrimination. Artificial intelligence reflects these patterns. UNESCO’s 2019 estimates indicated that women comprise only 12 percent of AI researchers, 6 percent of software developers, and are 13 times less likely than men to file ICT patents.
Bias in AI systems can emerge during algorithm design, dataset training, and automated decision-making. Algorithms transform input data into outputs; when input data carries bias, it propagates through the system, reinforcing inequities over time. Subjective decisions in dataset selection and preparation can embed such bias. In natural language processing, word embeddings have been shown to encode sexism, racism, and ableism. A notable case involved Amazon’s automated resume screening tool, which discriminated against women because it was trained on historical resumes dominated by male applicants. The model relied on linguistic signals tied to male candidates’ success, prompting Amazon to discard it once the bias was identified.
Gendered design choices in AI interfaces reflect and perpetuate societal stereotypes. Voice assistants like Amazon’s Alexa, Microsoft’s Cortana, and Apple’s Siri originally defaulted to feminine voices, with personalities described by UNESCO as “submissive” and “helpful.” By contrast, IBM’s Watson used a masculine voice in medical contexts, perceived as “authoritarian and assertive.” Even robot design follows occupational stereotypes: “male” robots are deployed for security roles, while “female” robots have staffed hospitality positions, as seen in Japan’s robot-run hotel in 2015.
These feminized AI systems often perform affective labor—tasks involving emotional management such as comforting, listening, and reassuring—traditionally expected of women. Voice interfaces handle scheduling, reminders, and information retrieval without the fatigue or emotional strain faced by human workers, creating an idealized “fantasy” of care. However, humanization of these systems can lead to objectification. The humanoid robot Sophia was designed with “exceptionally attractive” features, evoking “mechanico-eroticism.” Developers have noted that feminized AI devices face verbal harassment, leading Amazon to introduce a “disengagement mode” for Alexa.
The pace of AI adoption is accelerating, yet normative frameworks to address gender bias remain inadequate. At UNESCO’s 2020 Global Dialogue on Gender Equality and AI, participants observed that effective AI principles addressing gender equality as a standalone issue were largely absent. Identifying bias sources is essential: data collection practices, algorithm authorship, and developer assumptions all contribute. Since most female-characterized AIs are created by men, they often reflect male perspectives on women. Increasing women’s participation in STEM is critical, but workplace challenges persist—globally, half of women scientists report experiencing sexual harassment. Support structures, zero-tolerance policies, and enforcement mechanisms are necessary alongside recruitment efforts.
There are emerging tools and practices aimed at mitigating bias. AI-powered gender decoders can promote gender-sensitive hiring. Developers are becoming more aware of AI’s gendered impacts, particularly for younger users. A “human-centered AI” approach, focusing on user needs, can help. The European Union Agency for Fundamental Rights recommends “fundamental rights impact assessments” and real-world “discrimination testing” to detect and eliminate bias. In 2017, the European Parliament adopted a resolution establishing a framework to counter discrimination from algorithmic systems.
Ethical AI demands intersectional analysis, integrating considerations of gender, race, ethnicity, socioeconomic status, and other determinants, while adhering to human rights principles of transparency, accountability, and dignity. Collaboration among corporations, tech firms, academia, UN agencies, civil society, and media is essential to develop solutions. The UN secretary-general’s proposed Global Digital Compact, slated for agreement at the Summit of the Future in September 2024, aims to address these issues, including gender bias in AI, ensuring that AI serves the common good without perpetuating inequities.