Reproducibility and Artifical Intelligence
Large language models and other artificial intelligence tools can be used to improve reproducibility and replicability in science. In this project, RWI supports an interdisciplinary team of computer scientists, economists and communication scientists in developing an e-toolchain for automatizing the extraction of results from published papers and their replication packages. The aim is to facilitate reproducibility checks of authors, replicators, data editors and reviewers. Other purposes are meta-reproductions to scrutinize literatures for systematic errors such as coding errors and wrongly implemented methods. Drawing on our vast experience with robustness reproductions, forensic reproductions, and meta-reproductions our contribution to this project is the verification (the ‘ground-truthing’) of automatized reproductions.
Publications
Project start:
01. May 2025
Project end:
30. April 2028
Project management:
Dr. Gunther Bensch,
Prof. Dr. Jörg Ankel-Peters
Project staff:
Martin Buchner,
Dr. Julian Rose
Project partners:
Universität Ulm,
GESIS - Leibniz-Institut für Sozialwissenschaften
Funding:
Deutsche Forschungsgemeinschaft