Abstract for: Calling All U.S. Presidents for a Friendly Chat: Applying Generative AI to Study Political History

What would George Washington say if he were asked, “what factors influence the economy?” How would Theodore Roosevelt’s or Barack Obama’s thoughts differ? Advancements in generative artificial intelligence provide an opportunity to analyze historical documents and extract responses to desired questions which might not be directly available. We create 60 generative agents using an LLM trained on US presidents’ inaugural addresses. We ask each agent “what factors influence the economy?” We validate the responses with other LLMs tasked with predicting the president most likely to have given each response. We use the SD Bot to extract variables and relationships from the responses and depict mental models. Finally, we quantify and visualize presidents’ relative similarities to each other. Other LLMs identified presidents with 80% accuracy. A network graph and heatmap show presidents’ relative similarities. Over time, presidents’ focus has shifted from concepts such as virtue to other variables such as infrastructure and education; variables such as unity and public management performance are more perennial. Collectively, presidents believe that unity and good governance are important and that a reinforcing feedback loop exists between the economy, jobs, and markets. Studying mental models of historical characters can become more feasible using a combination of systems thinking approaches and generative AI. This study provides a showcase of this opportunity, including the increased efficiency gained through automation of the process. Assistance with R code syntax.