A recap of “Artists on AI and Data”

by
Translate this post

Presented by Art+Feminism in partnership with Wikimedia Deutschland, featuring artists Nora Al-Badri and Michael Mandiberg (Art+Feminism co-founder). 

On October 8 in Berlin, we gathered for a hybrid program at the Wikimedia Deutschland office to talk about AI and data. This event launched Year Two of Art+Feminism’s campaign, What Would a Truly Feminist Internet Look Like? While Year One was about visibility—bringing more people and sources into Wikipedia, Commons, and Wikiquote. Year Two is about structure, especially Wikidata, data models, and governance. We chose this conversation because artists are already working with datasets and AI in concrete ways, and those choices shape the stories our movement can tell.

Kira Wisniewski, Executive Director of Art+Feminism

Solenne Lazare (she/her), Head of Product Strategy at Wikimedia Deutschland, opened the program with a quick primer. She explained Wikidata as a collaborative knowledge graph that links facts so people can verify them and machines can reuse them. She pointed to Wikibase, which lets GLAMs and communities run their own graphs while staying connected, and she also shared work on “embedding,” a way to make Wikidata searchable by meaning while still pointing back to the specific Wikidata items for verification. 

Solenne Lazare, Head of Product Strategy at Wikimedia Deutschland

Nora Al-Badri (she/her) came next. An Iraqi-German conceptual media artist, she works with what she calls big cultural data—the digitized collections that shape what the world thinks it knows about West Asia. When institutions declined to share internal datasets, she and her collaborators scraped the public record and trained a model on thousands of images of artifacts. Their results make a simple case: whoever controls the dataset controls the story. If museums hold the objects and the data, they can script entire cultures, and that power seeps into the training sets that feed today’s systems.

Her project, Post-Truth Museum, turns that logic inside out. Using deepfake techniques, she has institutional voices speak about care and restitution, then lets the objects speak for themselves. The piece shows how a dataset often treated as neutral can become an instrument. Sometimes, the ethical path is not wider access but clearer governance in the hands of the communities whose knowledge is at stake.

Nora Al-Badri (left) and Michael Mandiberg (right)

Michael Mandiberg (they/them), an American artist and computer programmer, followed with a different archive and a related problem. They’ve collected and analyzed 130 million stock photographs. The analysis shows consistent biases: office scenes dominate, beauty is narrowly defined, subjects skew white, and the same gestures repeat. In practice, those labels become lessons for models. When topics like “fashion” or “wedding” carry gendered assumptions in their tags, models learn those assumptions as facts. Tell a machine, again and again, that “professional” looks one way and “care” looks another, and it will mirror that back at scale.

For Wikimedians, the lesson is this: the choices we make on Wikidata and Commons become patterns that readers and models absorb. From Berlin to your local project, the message is the same, make structure fairer. Add what’s missing, document why, respect community protocols, and link it all with citations. If AI is learning from us, let it learn better.

Ways to plug in now

We want to thank Wikimedia Deutschland for partnering with us for this important conversation.

Can you help us translate this article?

In order for this article to reach as many people as possible we would like your help. Can you translate this article to get the message out?