AI and Archaeological Pottery Identification: Reflections on the importance of open and persistent data

Abstract

This article is a discussion of the work of the Archaeological Automatic Interpretation and Documentation of cEramics (ArchAIDE) project, in the context of the new collaborative work now being undertaken in the AUTOMated enriched digitisation of Archaeological liThics and cerAmics (AUTOMATA) project, including partners from both ArchAIDE and the Arch-I-Scan Project (and described in the other articles within this volume). Completed in 2019, the purpose of the ArchAIDE project was to develop new technologies and methodologies to optimise and economise the identification process for archaeological pottery. It resulted in a range of outputs, primarily a new proof-of-concept mobile app that streamlined the practice of pottery recognition by type using Artificial Intelligence (AI). ArchAIDE represented one of the earliest contributions to research in the area of automated image recognition of archaeological material, but more importantly, its development was driven entirely by the needs of archaeologists, and archaeologists were always at the heart of the implementation of the project. As such, ArchAIDE focused on low-tech approaches accessible to the widest range of archaeologists possible, with an emphasis on automating the mundane aspects of existing workflows, while ensuring key points of decision-making within the identification process were left intact. The overarching lesson learned from ArchAIDE was that there is much work to do to create a sufficient amount of well-identified training data to ensure good results from AI-driven workflows in archaeology. ArchAIDE and the Arch-I-Scan Project have shown that AI applications and workflows incorporating automated image recognition using trained neural networks can speed the identification of archaeological pottery. They can better automate the documentation and creation of reliable training data, but to accomplish this, at the kind of scale necessary to make this work more than a proof-of-concept, further work is necessary. The AUTOMated enriched digitisation of Archaeological liThics and cerAmics (AUTOMATA) project is taking many of the lessons learned from ArchAIDE and the Arch-I-Scan Project, while also incorporating robotics to work towards implementing these workflows at scale. This article focuses on the contributions of these types of projects, from a digital stewardship and sustainability perspective, and the important lessons learned regarding how archaeology navigates the use of AI applications in the future.