AI Assistant for Geoanthropology

The project aims to create an AI assistant to support research in the emerging, highly interdisciplinary field of geoanthropology. The new generation of generative AI is starting to transform scientific practice across all disciplines. In particular, large language models (LLMs) are rapidly becoming better at understanding text and quickly generating accurate responses, making this technology broadly applicable across many domains, including science. While the full impact on science is not yet well understood and limitations are still being discussed, it is clear that researchers are already massively using the new technology in their scientific workflows, from low-level tasks such as proofreading text to more integral parts of scientific routines such as summarizing or querying scientific literature. LLMs undoubtedly have the potential to make scientific research more efficient, effective, and accessible, and will certainly change how we consume and create science over the next few years.

Our AI assistant is based on a domain expert Large Language Model, Geacop. A prototype of Geacop exists, which has been fine-tuned on about 5000 books, book chapters, and articles by the active participants of the institute's inaugural conference, "The Anthropocene - Crossing Boundaries." Future versions of the model will be trained on a much larger corpus of literature pertinent to the field of geoanthropology. The LLM is embedded in a digital environment called Kantropos, which gives Geacop, and thereby the researchers interacting with it, direct access to relevant scientific knowledge. Together with Krishna Gummadi and his group at the Max Planck Institute for Software Systems, we investigate how existing AI technologies can be adapted to better serve the needs arising from their use in research contexts.