Action Identification Characteristics and Priming Effects in ChatGPT
Eli Fennell · 2023
Action Identification Theory (Act ID) refers to the traits and mental processes behind how people think about and describe actions in more concrete terms, called Lower-Level Action Identities, or more abstract terms, called Higher-Level Action Identities. Optimal Act ID use is associated with a variety of communicational and conversational goals and outcomes, and is deeply embedded in the structure of human language. ChatGPT is a Large Large Model (LLM)-driven Conversational AI Chatbot, trained on patterns of language and conversational dynamics, including actions and descriptions of actions at different levels of concreteness and abstraction. ChatGPT is therefore a suitable subject for novel forms of Act ID research. Across 3 pre-registered studies, the Behavior Identification Form was used to assess: 1) ChatGPT’s ability to Apply Act ID (Study 1); To measure its default Act ID level (Study 2); and to evaluate its susceptibility to priming-like effects (Study 3). Results revealed that ChatGPT: 1) Can learn and apply Act ID (r = 1.0, p < .000); 2) Displays a high-level default Act ID (M = 20.72, SD = .79); and 3) Can be experimentally induced to select more or fewer lower- or higher-level identities via simple tasks inspired by human research. These findings hold significant implications for optimal human-AI interactions, raise critical questions about the intelligence and safety of these virtual agents, help further our understanding of how Psychological theories and models can improve the study and development of these applications, and may contribute to the broader study of the human mind and consciousness.