Talks and tutorials

Being critical with AI investigations at DMI Summer School 2024

This tutorial examines how using AI based models for social investigations brings in the unknown values and world views. For example, large language models are known to have more traditional and even conservative gender representations. If we use these tools to “measure the society” we automatically incorporate these worldviews into our analysis.

In the first part of the tutorial, we focus on image recognition tools and examine differences they have in labelling following methods proposed by Berg & Nelimarkka (2023). We discuss differences and implications these have and learn how to use the COSLAB tool presented in the work

In the second part of the tutorial, we embed ourselves into values in large language models. We walk through the code used for fine-tuning an LLM with Marx’s thoughts and devise our own LLM to focus on a theoretical stance we have.