Beyond Cheating: Using ChatGPT to Structure Data

Alexandra Miller, George Mason University

In the world today, Artificial Intelligence seems to be everywhere. We have self-driving cars, digital assistants, and the notorious ChatGPT. In academia, much hay has been made about the potential for students to abuse ChatGPT in the classroom. However, the web-based application has many more uses. Users can generate art, translation, code, and more. Usefully, users can also prompt ChatGPT to structure data from OCR text. In Historical GIS, structuring data can be one of the most time-consuming steps in the research process. However, using ChatGPT speeds up an otherwise slow and tedious manual process. Historians can copy and paste OCR text into the ChatGPT prompt box, then prompt the application to structure the data into an information table. Historians can create unique prompts to structure their data in a way that most suits both their sources and their research mode. Because the user writes their own prompt to structure data, they exercise a great deal of control over the data structuring process. And, because ChatGPT structures data faster than a historian can manually do so, historians can also revise their data tables as they go. There are a few drawbacks to using ChatGPT which historians should be aware of. Since ChatGPT is an application based on Artificial Intelligence, it works to predict and reproduce patterns. As a result, the application occasionally produces hallucinations or misaligns the information table. However, ChatGPT produces largely accurate results. In conclusion, I argue that ChatGPT is a useful tool for historical mapping and data-based research. Furthermore, I suggest that using ChatGPT is a legitimate research practice which does not undermine the validity of data or mapping results.

No extended abstract or paper available

 Presented in Session 167. New Methods in Historical Mapping: Deep Maps, ChatGPT, and New Media