The Hermeneutic Loop: Explainable AI (XAI) as a Partner in Interpretive HASS Research

Abstract

This paper presented a new method called the “Hermeneutic Loop” for using Explainable Artificial Intelligence (XAI) in humanities and social science research. The goal was to solve a major problem: complex AI often acts as a “black box,” making it challenging for researchers to understand how it reaches its results. This lack of transparency clashes with the core principles of fields that rely on deep interpretation. Our framework transformed XAI from a simple technical tool into a collaborative partner. It created a structured conversation between the researcher and the AI model, making the process of interpretation clearer and more rigorous. We demonstrated this method with three real-world examples. First, in literary studies, we used a language model (BERT) and a technique called Integrated Gradients on 500 British novels. This helped us visually trace how the meaning of the word “virtue” changed from a social concept to a psychological one over a century. Second, in historical research, we analyzed letters from the American Civil War. When a standard sentiment-analysis model failed, we used a method called LIME to understand its mistakes. This process revealed unique 19th-century ways of expressing emotion that the AI initially had missed. Finally, in a study of climate change debates on Twitter, we used “attention visualization” to see how different groups used the same words to build distinct political arguments. The Hermeneutic Loop improved our research, forcing us to question the AI’s output, which led to more reliable and insightful conclusions. It ensured that the computer’s power served human understanding. This approach provides a necessary bridge, allowing scholars to use advanced AI while staying true to the fundamental values of evidence-based interpretation.