Postdoc on the subject of Bias and Fairness in Knowledge Graphs

Are you inspired by the idea of an inclusive Semantic Web? We are looking for a talented postdoctoral researcher to study various types of bias in knowledge graphs, Linked Open Data and/or metadata.

The position is part of the HAICu project. HAICu (digital Humanities - Artificial Intelligence - Cultural heritage) is a large-scale Dutch research project in which AI researchers and Digital Humanities scholars collaborate with cultural-heritage institutions such as libraries, archives and museums. Linked Open Data is widely used in this sector for metadata about collection objects, for data enrichment, and for cross-collection links.

Inclusivity is a key value in the cultural heritage domain, with organizations employing a range of strategies to deal with unwanted bias in their (often historic) collections. Inspired by these efforts, for this position we focus on bias and fairness on the Semantic Web. Bias in this context may come in various forms. For example, groups of people may be over or under-represented among the entities in Linked Open Data. Or, the labels and descriptions used to represent people or their cultures may reinforce negative stereotypes, e.g., when outdated, colonial terminology is used. We will investigate one or more of the following topics:

• To what extent and in what way is social bias reflected in LOD?
• What are the strategies employed by the LOD community to reduce bias and promote inclusivity?
• What is the impact of bias in LOD on applications (e.g. generative AI) and users?

Within the HAICu team, the postdoc researcher will participate in Work Package 5, titled “Construction of polyvocal, multimodal narratives.” In this Work Package, CWI will collaborate with UvA and VU and with the National Museum of World Cultures.

The researcher will be based at CWI in a dynamic research group called Human-Centered Data Analytics (HCDA). HCDA investigates human-centered, responsible AI in the culture and media sectors. How can we ensure that digital systems are inclusive, promote diversity, and can be used to combat misinformation? The HCDA group addresses these important questions. Our work includes a wide range of techniques, such as statistical AI (machine learning), symbolic AI (knowledge graphs, reasoning), and human computation (crowdsourcing). By analyzing empirical evidence of human interactions with data and systems, we derive insights into the impact of design and implementation choices on users. We maintain close collaborations with professionals from the culture and media sectors, as well as social scientists and humanities scholars, through the Cultural AI Lab and the AI, Media and Democracy Lab. These interdisciplinary labs provide us with opportunities to work with real data and real-world use cases.