“The kinds of work that field archaeologists do especially on excavation is notoriously extractive and destructive” Dr. Paul Zimmerman declared. This statement set the tone for a session that illuminated the dual nature of archaeology as both enlightening and destructive. Dr. Zimmerman, a field archaeologist at the University of Pennsylvania Museum highlighted a fundamental contradiction: the act of excavating artifacts, which aims to unveil ancient stories, simultaneously threatens their preservation.
The webinar explored the intersection of archaeology and emerging AI technologies. With a distinguished panel of experts in fields as diverse as machine learning, robotics, and ancient languages, the session examined how cutting-edge technologies might reconcile this destructive potential with the need to uncover and protect human history. The panel presented a range of research applications in archaeology where AI is proving invaluable. These technologies are pushing the boundaries of how data is collected, analyzed, and preserved, offering a glimpse into the future of archaeological exploration.
Dr. Zimmerman’s candid reflection on archaeology’s destructive tendencies resonated as he introduced two pioneering researchers, Thea Sommerschield and Yannis Assael. Their work, a collaboration between academia and Google DeepMind, showcased AI’s transformative role in deciphering ancient languages.
Assael explained, “Our research is about predicting the past, and because without knowledge of the past, there can be no understanding of the future.”
Their project focuses on restoring ancient Greek inscriptions, many of which are damaged or incomplete. Using a Transformer-based neural network model, named Ithaca, their AI system tackles the epigrapher’s workflow of restoration, dating, and geographic attribution. In the presentation, they demonstrated how Ithaca reconstructed missing portions of ancient texts, visualizing hypotheses that provided historians with multiple possible interpretations. This ability to produce explainable results, they explained, allows historians to make informed decisions when reconstructing the ancient world.
It was revealed that the collaboration between human historians and the AI system, Ithaca, significantly enhances the accuracy of textual restoration tasks. Initially, human experts were achieving approximately 25% accuracy on their own. However, when these experts utilized Ithaca’s capabilities, their combined efforts surpassed the performance of Ithaca alone (60%), which achieved around 70% accuracy, showcasing the substantial benefits of integrating AI with human expertise in historical studies.
The implications for historical research are profound. The AI model has also been used to date critical Athenian inscriptions, providing insights that align with recent historical breakthroughs.
As Sommerschield explained, “Ithaca’s dating predictions [have helped redate historical events], providing an eloquent example of how models like Ithaca can be used to exploit the full potential of large datasets in unprecedented ways.”
However, it was clear that AI, as promising as it might be, was not a panacea. Assael highlighted the intrinsic limitations of AI models despite their advanced predictive and restorative capabilities. He emphasized that while these tools are highly effective, human oversight is indispensable. The intention behind these technologies is not to supplant human expertise but to augment it, ensuring that the critical judgment of historians remains at the forefront of historical analysis and interpretation.
Alex Brandsen discussed enhancing the accessibility of archaeological data using the BERT model (Bidirectional Encoder Representations from Transformers). BERT processes language in context, making it highly effective for comprehending complex text. BERT enables to improve search capabilities within extensive archives of archaeological reports. To optimize BERT’s performance, human input is crucial. Researchers manually label text, teaching BERT to recognize and categorize archaeological terms accurately. This collaboration ensures that BERT can effectively navigate synonyms and homonyms, enabling deeper searches that extend beyond basic metadata.
Although the application of robotics in archaeology is relatively nascent, Tuna Kalayci presented a persuasive argument for its role. He elucidated the potential of robotic systems within archaeological frameworks. Enhanced with sophisticated computer vision and machine learning technologies, these systems are capable of detecting and tagging artifacts in real-time. Furthermore, a prototype for a robotic field assistant, capable of navigating challenging landscapes while identifying and documenting artifacts, was already in development, promising to revolutionize field operations and data collection in archaeology.
The technical challenges are many, as Kalayci explained that the diversity of archaeological sites means that there is no one-size-fits-all solution when it comes to robotic mobility.
“We are trying to understand if legged systems are superior to wheel systems or better, if we can find systems where things are interchangeable depending on your needs” said Brandsen.
Kalayci’s presentation underscored the enduring importance of human expertise in the field of archaeology, despite the increasing automation brought by robotics. He stressed that the purpose of robotic systems is not to supplant archaeologists but to augment their work by assuming responsibility for labor-intensive tasks such as sorting soil samples for micro-artifacts. By highlighting the profoundly embodied nature of archaeology, Kalayci articulated how robotics could relieve human experts from the physical demands of their work, thereby enabling them to devote more attention to the interpretation, analysis, and deeper understanding of artifacts. This thoughtful integration of robotics allows for a more focused examination of subtle historical nuances, substantially enhancing the quality and depth of archaeological research.
A recurring theme throughout the session was how AI and human expertise complement one another. Whether it was the neural networks restoring ancient Greek texts or robots enhancing excavation processes, the technology was always presented as an augmentation of human skills, not a substitute.
The discussion then shifted to the macro scale, as Iris Kramer took the audience to the skies. Her work, using AI to analyze satellite imagery and LiDAR data, offers a different lens through which to view archaeology, not as isolated digs but as part of vast, interconnected landscapes. They detect archaeological features across entire countries, identifying lost woodlands, ancient agricultural practices, and historical land use patterns.
Kramer’s research has broad applications, from protecting heritage sites threatened by urban development to informing landscape restoration efforts.
“AI in archaeology [is helping us] map historic landscapes throughout history to support planning and making informed, sustainable decisions for the future,” Kramer said. One striking example she shared was how AI had revealed 3% of England still bore traces of medieval ridge-and-furrow farming, a discovery that had profound implications for conservation.
The audience was reminded that despite the advances AI offers, it cannot shield archaeology from all risks. The deployment of AI in archaeology might mitigate some losses by preserving data and insights digitally, but as excavation continues, history is inevitably disrupted.
The panelists stressed the importance of collaboration between archaeologists, technologists, and policymakers. AI can help illuminate history’s enigmas, but it requires careful, responsible implementation. As the panelists observed, the future of archaeology, shaped by AI and robotics, will be both exciting and complex. The challenge, then, lies in harnessing these tools wisely to ensure that as we uncover history, we also preserve its integrity.
