A fall in Geneva, Strasbourg, The Hague and New York, and the question every room started asking
It has been a while. Here is why.
This fall ran from a Human Rights Council resolution in Geneva to a treaty committee in Strasbourg, and from the city of international justice to the annual stampede that is UNGA. In every room the same question surfaced: who is in the data, and who decides when it is enough? We have argued now for years that representativeness is a quality standard, not a political preference. A system built on half the world’s data fails on its own technical terms. This season that argument began turning up in other people’s documents, some of which will shape how AI is governed for decades.
Start in Geneva. With Armenia and Estonia, we worked on Human Rights Council resolution 63/13 on the rights of women and girls in the context of digital technologies. It passed on 6 October without a vote, which in Geneva is its own small miracle, and now has sixty sponsors. The text names unrepresentative data as a source of gender discrimination and calls for gender-responsive data governance across the entire data life cycle. Another of our proposals made it in too. The Council recognises the role of the UN Independent International Scientific Panel on AI in assessing how AI affects women and girls differently, and invites the Panel to the table at its sixty-eighth session. A scientific panel that never asks who is missing from its evidence base will only produce a science of the people already counted.
Then the CSTD. As co-chair of its Gender Advisory Board, I opened the issues paper for next week’s Intersessional, Data Governance for Development: From Asymmetry to Equitable Stewardship, and found our sentences coming back in UN prose, with the Board credited six times. A dataset can meet the FAIR principles and still be unrepresentative: that is now in a UN document.
So is the mechanism we call recursion half-life. As models train on their own outputs, gaps in representation compound from data to model to output, and the data that was thin to begin with goes first. This is not a future risk. It is the training pipeline doing what it was built to do.
Which is why we keep hammering on indicators. WSIS, the World Summit on the Information Society, has spent twenty years measuring connectivity. That tells you who came through the door. It tells you nothing about who leads the research team, files the patent, sets the standard or pockets the value. The General Assembly has mandated gender equality across all eleven WSIS Action Lines. Yet not one carries a gender indicator. A mandate without a measure is only a wish.
With the Gender Advisory Board and our Gender in Digital Coalition, we are making the case that indicators are transformative, not administrative, and we are taking it to Nairobi. “Counting What Counts: Gender Indicators in WSIS Action Lines” is a Day Zero event at the Internet Governance Forum (IGF), brought forward by Open Data Collaboratives.
In between, Strasbourg, more than once. At the Council of Europe’s Committee on Digital and New Emerging Technologies (CDNET), where we hold Observer status, the work on stakeholder engagement and draft procurement guidelines is where the Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law, (the first legally binding international treaty on AI), starts to become something a government can buy against. Procurement is where ethics stops being rhetoric and becomes a binding contract clause. Our Women at the Table standalone paper on AI procurement for municipalities, with a methodology and an interactive toolbox, lands in the coming weeks.
The rest of the autumn was a moving seminar. Down the rue de Lausanne to the WTO Public Forum for “AI and digital services: The trade case for considering women in standards development,” sponsored by Canada, Australia, Mexico and the Republic of Korea, with El Salvador’s Minister of Economy. There, on the same panel, colleagues from the women’s health company Organon called, unprompted, for representative data, which is not a sentence you expect to hear at a trade forum. Back to Strasbourg to speak for the Council of Europe at the WEOG Business and Human Rights Forum on “AI & Human Rights: Managing Risks and Seizing Opportunities.” Then to The Hague, where I moderated “Big Tech, AI and Feminist Digital Accountability” at the Gender Justice in International Criminal Law convening.
And New York, which already feels long ago and was, intellectually and physically, chaos. ITU and UNDP’s Digital Day and Digital@UNGA. ODET and UNDP on DPI Safeguards. An IEEE roundtable. Receptions at Google, Microsoft, UN Women and the Federal Reserve. The launch of Humane Intelligence’s work toward an international independent AI evaluation system. AI Safety Connect’s Governing at the Speed of AI dinner. And the most dynamic room of the week, the UN Foundation’s People, Power & AI at the Ford Foundation.
Somewhere between Geneva and New York I noticed that everyone had started to sound like feminists. Not in vocabulary. In the question. Feminist analysis has always started with who decides: by whom a system is built, for whom, in whose interest and at whose cost. Now a UN trade and development paper asks who sets the rules of interoperability and who walks away with the value. A Human Rights Council resolution names unrepresentative data as a driver of discrimination, and nobody calls a vote. A women’s health company asks a trade forum for better data. Trade negotiators and international criminal lawyers, starting from opposite ends of the legal order, land in the same place: you cannot hold a system accountable for harm to people it was never built to see. None of them said feminist. They didn’t need to.
Next week, the Palais. The UN Working Group on Data Governance. Our Gender Advisory Board side event, “From Data Asymmetry to Equitable Stewardship: The Gender Imperative for Development.” The CSTD Intersessional. WIPO’s first Global Forum on Intellectual Property and Artificial Intelligence. And the launch of our AI procurement work.
Who gets counted is settled long before anyone writes a line of code. By the time harm shows up, the people it falls on have already been written out of the record. This fall was about getting into the rooms where that record is written, and changing it while the ink is still wet.