AI Ethics · Commentary
UNESCO Recommendation on the Ethics of Artificial Intelligence
UNESCO’s 2021 recommendation treats gender as a concern across impact assessment, research funding, design and deployment.
Article analysis
UNESCO’s General Conference adopted its Recommendation on the Ethics of Artificial Intelligence on 23 November 2021. Paragraphs 87–93 address gender and call for a gender perspective throughout ethical impact assessment. Harms can originate in data collection or procurement and become visible only after a system enters use.
Bias and authority over design
The document addresses both stereotypes in systems and who can participate in building them. Training material can encode an existing division of labour. Hiring and promotion also shape who decides which objectives a system pursues. The recommendation addresses research funding and representation in teams, opening questions about how budgets and objectives relate to who benefits.
Access depends on whether people can afford and use technology. Infrastructure, education, care time and exposure to harassment can affect participation. Passing one bias test leaves questions about who can raise a complaint and which institution is responsible for responding to it.
Follow a decision from procurement to appeal
Take a hypothetical recruitment system and read the recommendation’s impact-assessment provisions alongside its gender section. A model that uses prior promotions as a signal of ability can carry the effects of past barriers into a new decision. Omitting a gender field leaves behind records of who led a team and who was assigned work that could produce visible results. An assessment needs to ask how those indicators came to exist before it draws conclusions from a uniform scoring rule.
Then ask who can correct the outcome. The buyer decides how to use the system, the supplier may hold the records needed to explain the score, and an applicant needs to know that an automated assessment took place. Pointing to a pattern of disadvantage has limited practical effect if nobody can review a hiring decision. Treating gender as relevant throughout a system’s life opens questions about who can inspect the data, review an individual result or suspend use. Answers depend on how a particular system is governed.
Read the Feminist Principles of the Internet for a movement document concerned with power over technology. The UNESCO recommendation sets policy directions. Claims about a specific system need evidence from its assessment, oversight and responses to harm.
Explore another medium
- Papers
Data Feminism for AI
Lauren F. Klein and Catherine D'Ignazio
Klein and D’Ignazio bring data feminism into AI research, using tenant screening, feminist data activism, and hidden annotation labor to examine power, consent, and environmental costs.
AI Ethics · Digital Feminism - Books
Weapons of Math Destruction
Cathy O'Neil
Cathy O’Neil explains how scoring, prediction, and risk models amplify existing inequalities across work, education, finance, and policing.
AI Ethics · Algorithmic Bias
Continue through the issues
Reader responses
Letters from readers
Respond to this article or add a source that belongs in the discussion.
Join the Discussion
Loading comments...
Support Us
If you find our content valuable, please consider supporting FemRes.