With multimodal large language models we can build large-scale historical text data sets based on archival sources at faster speeds than ever before.

The past has happened. The data that survived is out there to be collected. As we are able to build larger datasets than ever before, the ability to ask the right questions becomes even more important.

With the new microlevel data abundance, we will be able to characterize the wider institutional system. We can go one level further in complexity: what made an institution inclusive or extractive in a particular region and how did this affect local economic growth?

Tools from machine learning, such as natural language processing and computer vision, will help us to measure the inclusivity and interactions of institutions on the new microlevel data. With these tools, we can transform qualitative textual sources into quantitative data (e.g., Carus and Ogilvie, 2009). Integrating these newly constructed variables into the standard toolkit in econometrics may enable us to better understand the latent mechanisms that intertwined institutions, culture and economic growth.