Angelie Kraft

Person

Angelie Kraft is an interdisciplinary researcher in the area of natural language processing and AI ethics with a training in computer science and psychology. As a fellow at the Weizenbaum Institute, she is working in the group "Digitalization and Opening up Science". In her project, she examines popular benchmarks that aim to measure the "knowledgeability" of large language models (LLMs). She analyzes these benchmarks with regards to their representativeness, assuming that they are biased and foster the optimization of language models towards the knowledge mainly relevant to a male, white and Western standard.

Angelie Kraft has been pursuing a PhD in Computer Science since October 2021 at University of Hamburg and has been working as a researcher at Leuphana University Lüneburg in the work group "Artifical Intelligence and Explainability". Her research focuses on the ethically relevant limitations of LLMs, particularly in relation to questions of fairness and factual fidelity. Her work shows that knowledge and knowledge bases are not objective and instead underrepresent the perspectives of marginalized groups. This can contribute to reinforcing existing power imbalances, as the knowledge of a dominant group in particular is perpetuated and disseminated by influential LLM technologies.