AI Tool Analyzes Image Similarities for Art History at LMU Munich

Ludwig-Maximilians-Universität München (LMU) is developing an artificial intelligence tool at its Institute of Art History to assist researchers in analyzing visual similarities among artworks. The project, titled “Reflection-driven Artificial Intelligence in Art History,” aims to make the AI's decision-making process transparent within the context of art history.

The initiative is led by Professor Hubertus Kohle from LMU and Professor Ralph Ewerth from the German National Library of Science and Technology (TIB) in Hanover. The AI is being trained with art historical knowledge to support the analysis of similarities between images. The project is funded by the German Research Foundation (DFG) as part of the “The Digital Image” priority program.

The interdisciplinary team combines expertise from art history and computer science. Their objective is to create an algorithm with a clear and intuitive decision-making process. Researchers intend to visualize how the AI identifies similarities, considering factors such as materials used, subject matter, or historical context.

Initial steps involve examining historical automated image analysis methods and building a database of images and texts for training purposes. The project utilizes publicly available metadata from museums and images with expired copyrights. To mitigate potential biases, such as underrepresentation of female artists in training data, the system will flag any biases and explain their origins.

By transparently demonstrating its methodologies, the project seeks to foster acceptance of AI in art studies. Researchers emphasize that the AI will provide suggestions that must be validated by academics within the art historical framework, stating it will not replace human experts.