Session: #235

Content:

Based on last year’s session “Traum -Tracing Reality in Archaeology Using Machine learning” we aim to expand on this idea. The tremendous potential of machine learning is starting to be explored in archaeology alongside network models and other more traditional statistical applications. However, if these are used to explore new societal structure quite often social theory as well as methodological assumptions are neither discussed nor explored. Predominantly, this applies to the fact that human society is a complex system and therefore inherently more than the sum of its part.
We fully endorse the idea that including complexity is not requisite in all models and one should always choose the most basic approach that still explains the phenomenon. However, current social models are often harnessing the power of machine learning and modelling as they are able to spot new patterns, invisible to humans and standard approaches. This power comes with new problems, as the usual approaches are black box based. Whilst these can be judged by their accuracy, they cannot be easily understood. Without prior white box modelling and feature engineering these results are almost non-interpretable. Without strict, clear feature classification and knowledge of the data itself the model’s outcome more often resembles magic than usable results.
In our session we aim to discuss new approaches to Machine Learning in archaeology, to work on a framework for these models in social and theoretical archaeology as well as explore their ties to complexity theory and how it can improve results.

Session: #235

Theme & Session Format

Title & Content

Organisers