Theme:
6. A Decade after the ‘Third Science Revolution in Archaeology’
Session format:
Discussion session (with formal abstracts)
Title:
Machine Learning and the Creation of Archaeological Narratives
Content:
Machine learning in archaeology, especially in combination with social theories offers tremendous potential for the reconstruction of past interaction, social systems, and interaction. However, there is little user experience compared to other scientific techniques and more established statistical methods among archaeologists. The models - often necessarily based on ambiguous or sparse data - generate results which require interpretation. More often these are based on “commonly accepted” social theories, values, and ideals rather than embracing the problems and uncertainties in modelling. In her 2007 article, “Honoring Ambiguity/Problematizing Certitude”, Joan Gero argues for archaeological research which acknowledges and values the ambiguities and uncertainties of its data and the importance of “protecting and preserving ambiguities as a valuable rather than a necessarily painful aspect of archaeology” (323).
In the previous years our sessions (“Is ML in archaeology fact or fiction?” (2021) and “Tracing reality in archaeology using machine learning” (2020)) explored the potential, possibilities, ambiguities, and pitfalls of machine learning in social and bioarchaeological contexts. This year we specifically want to address the ambiguity of ML models, problems in their interpretation as well as their potential and danger to create narratives. Since all ML models are only capable of highlighting internal patterns of their respective data universes, they can easily perpetuate ideas of excessive certainty and amplify pre-existing biases of the data. Touching on ethics, representation, and oversimplification this perceived “sureness” has the power to distort the past, either involuntarily or voluntarily as a misuse of archaeology in the present.
We specifically invite contributions addressing ideas on how to retain ambiguity in modelling, spotting unwarranted presuppositions, colouring the conclusions, inclusion of ethics in archaeological machine learning applications, and more generally the dangers and pitfalls of creating narratives in archaeology through ML.
Keywords:
Deduction, Social Theory, Ethics
Session associated with MERC:
no
Session associated with CIfA:
no
Session associated with SAfA:
no
Session associated with CAA:
no
Session associated with DGUF:
no
Session associated with other:
Main organiser:
Chiara Girotto (Germany) 1
Co-organisers:
Henry Price (United Kingdom) 2
Affiliations:
1. Ludwig-Maximilians-Universität München
2. Imperial College London
THIS SESSION CAN NOT BE SELECTED AT THIS MOMENT