Large Language Models as communicative actors
The DFG-funded research project “Large Language Models as communicative actors” deals with the effects of communication with Large Language Models (LLMs) on users' attitudes and intentions to act. The project was initiated by DIID member Dr. Fabian Anicker and is funded from 01.04.2025 to 31.03.2028.
Large language models (LLMs) such as ChatGPT, LLaMA or Deepseek play a prominent role in the transformation of society through artificial intelligence. This development presents the social sciences with both fundamental theoretical and empirical challenges. Classical categories such as technology, medium, actor or infrastructure cannot be transferred to LLMs without further ado. The project therefore addresses two conceptual and empirical questions in particular:
What (communicative) agency is attributed to LLMs?
What effects does communication with LLMs have on attitudes and intentions to act, and to what extent is the effect of LLMs moderated by attributions of agency?
The project aims to test the extent to which the attribution of agency to LLMs modifies their effects on users by conducting experimental studies on the use of LLMs. The aim of the project is to develop an empirically based theory of communicative agency. This should not only contribute to a better theoretical understanding of interactions with LLMs, but also uncover the mechanisms behind the observed changes in behavior. In addition, the project will make methodological contributions to the development of indicators for the implicit and explicit attribution of agency. In the long term, these findings should contribute to a better understanding of the social status and impact of LLMs in the digital society.