AUTOMATION IN THE PRACTICE OF ARCHAEOLOGICAL SURVEY

Thanks to AHRC Collaborative Doctoral Partnership funding held jointly by Historic Environment Scotland (HES) and the University of Glasgow, we are offering a 45 month (3.75 years) PhD scholarship on developing approaches to integrate automation-led detection routines into workflows used in the professional practice of archaeological prospection and landscape archaeology, notably for large scale heritage management. The supervisors will be Dr Rachel Opitz (Archaeology) and Dr Jan Paul Siebert (Computer Science) at University of Glasgow, and Dr Lukasz Banaszek and Mr David Cowley (HES).

Aims

To critically assess the challenges of emerging automation-led approaches to current practice, propose potential new workflows that incorporate automation, and evaluate their design.
To explore synergies between the existing manually created knowledge base, the role of such information in training neural networks and designing object detection algorithms, and the validation and expectations of outputs, investigating how to bridge current and emerging practice.
To envisage innovative, effective workflows that reconfigure the knowledge-creation cycle for large scale archaeological prospection, inspired by and enabling the integration of AI and other automation-led approaches
Objectives
To build on work recently completed for the island of Arran and ongoing HES survey projects with their Rapid Archaeological Mapping Programme, together with case studies elsewhere in Europe (Germany, Norway, Netherlands) to assess current practice in automation-led prospection.
To iteratively research, design, prototype and evaluate workflows that reflect different approaches to integrating AI-led methods within the framework of large area and national archaeological mapping by heritage agencies.
To explore the specific stages and aspects of the socio-technical process, e.g. training and test set definition, choice of network architecture, role of field checks of identifications, and the role of collaboration between archaeologists and computer scientists.
Automated detection routines have been viewed as potentially useful or even transformative for several decades, and recent progress in artificial intelligence (AI) based in machine learning and computer vision has moved these approaches from potentially interesting to practically implementable across a variety of applications. Within archaeology, the potential of AI-led approaches and heavily automated image processing for partially automating the identification of archaeological features and landscape changes has been demonstrated in several studies. Their implementation has brought measurable benefits, leading to increased investment in their development. While the technologies themselves are being pursued, less attention has been paid to the analytical and interpretive frameworks within which semi-automated computational approaches to feature identification, notably AIs, are integrated into practices of archaeological landscape interpretation, particularly within heritage management bodies. Current practice in this field is largely based on manual processes dominated by individual visual observation, and the integration of AI-led and other automated approaches requires a critical examination of interpretive routines and the cycle of knowledge creation.

In order to examine the impact of automated detection on interpretive practice, the student will iteratively research, design, prototype and evaluate workflows that reflect different approaches to integrating AI-led methods within the framework of large area and national archaeological mapping by heritage agencies. The project will draw on work recently completed for the island of Arran and ongoing HES survey projects with their Rapid Archaeological Mapping Programme, together with case studies elsewhere in Europe (Germany, Norway, Netherlands). Within these case studies, the project will explore the specific stages and aspects of the socio-technical process, e.g. training and test set definition, choice of network architecture, role of field checks of identifications, and the role of collaboration between archaeologists and computer scientists.

This research will highlight key points of impact, taking an essential first step toward the definition of exemplar workflows for a new knowledge-creation cycle in large-scale archaeological prospection. In doing so, this project will explore synergies between the existing manually created knowledge base, the role of such information in training neural networks and designing object detection algorithms, and the validation and expectations of outputs, investigating how to bridge current and emerging practice.

AI-led and other automated approaches to archaeological feature identification promise to profoundly change the practices of archaeological prospection and landscape archaeology, effecting fundamental shifts in workflow and routine practice, bringing significant benefits. However, the benefits of these approaches can only be realised if the necessary adaptation of routines based on the traditional primary role of the human observer/interpreter is deliberately considered and thoughtfully managed. Envisaging innovative, effective workflows that reconfigure the knowledge-creation cycle for large scale archaeological prospection, inspired by and enabling the integration of AI and other automation-led approaches, is the core focus of this project.