A multidisciplinary team of Spanish and Portuguese archaeologists and artificial intelligence experts has combined non-destructive archaeological measurement techniques, machine learning and Explainable Artificial Intelligence (XAI) tools to develop an AI system applicable to archaeological research. In this specific case, to investigate the provenance of archaeological samples of variscite, a mineral with a characteristic green colour highly appreciated in Prehistory and distributed by extensive exchange networks throughout Western Europe between the sixth and second millennium BC. It was used to make necklaces, bracelets, rings... Items of personal adornment in general.
This group of researchers has been collaborating for years to find out where variscite comes from in the different archaeological sites of the Iberian Peninsula. To do this, they compare current geological samples of variscite with samples found in archaeological excavations. They analyze the mineral, record its elements, and then compare the small chemical variations they present. Based on the similarities, it is possible to determine from which place it has been extracted.
The study, published in the Journal of Archaeological Science, is led by the University of Lisbon and has the participation of the Milá y Fontanals Institution for Research in the Humanities (IMF-CSIC), the University of Seville, the University of Alcalá and the the CIPAG (a Spanish acronym that stands for “Collective for the Research of Prehistory and Archaeology of Garraf-Ordal”).
A unique geochemical footprint in each mine
The innovation of this study lies in the use of AI to analyze the results of the chemical composition. "Our model learns to recognize the unique geochemical footprint of each mine. It is able to identify where a prehistoric bead comes from, even thousands of years after it was manufactured," explains Daniel Sánchez-Gómez, a researcher at the University of Lisbon and lead author of the study. Thanks to this pioneering approach, they have been able to predict with 95% accuracy the geological origin of archaeological objects made with variscite.
In this way, the team has built the most extensive compositional database created to date, with more than 1,800 geological samples and 571 archaeological accounts, which have been analyzed using portable X-ray fluorescence.
To process the data, they used a random forest algorithm, which is very commonly used in Machine Learning, which has allowed them to achieve unprecedented precision. In addition, explains Ferrán Borrell, an archaeologist at the IMF-CSIC, "something very remarkable about this project is that the information has been uploaded to Zenodo [an open repository developed under the European OpenAIRE program and operated by CERN], so that other researchers can use that data and make their own interpretations. It is working with open science." Ferran Borrell directs the excavation project of the prehistoric variscite mines in Gavà (Spain), from which part of the samples studied come from.
The results have made it possible to reinterpret prehistoric trade routes. Now, the researchers explain, we can know that the mines of Gavà (Barcelona, Spain) and Aliste (Zamora, Spain) were the main production and distribution centres; that the traditionally cited source of Encinasola (Huelva, Spain) would have been of lesser importance; and that the materials found in Brittany (France) probably come from the north of the Iberian Peninsula, suggesting trans-Pyrenean land routes, rather than the maritime ones proposed so far.
"We have used explainable artificial intelligence techniques, which allow AI models, especially the most complex ones, to explain in a clear and understandable way how they make their decisions. In the case of our research, this means that it not only accurately predicts, but also shows us which chemical elements were decisive in each classification, bringing transparency and rigor to archaeological interpretation," adds Carlos Odriozola, professor at the University of Seville and PI of the project. This methodological framework, called VORTEX (Variscite Origin Recognition Technology X-ray based), opens up new possibilities for the study of the provenance of other archaeological materials, such as amber, and constitutes a milestone in the application of artificial intelligence to cultural heritage.
Now, according to Manuel Edo Benaiges, co-author of the article, we must try to address the following questions: What was the reason for the expansion phenomenon of the green stone? How did this expansion through Western Europe occur over time? Where did it start? The interpretations are many, always with the common goal of knowing more about the past. "It's not just about green beads: it's about using artificial intelligence to tell the human stories of prehistory," concludes Sánchez-Gómez.