Semantic Austerity and Conversational Compulsion A Digital Ethnography of Large Language Model Responses to Ambiguous and Incomplete Human Statements

This record presents a digital ethnographic study examining how an engagement-optimized conversational AI responds to ambiguous, incomplete, and low-agency human statements under conditions of semantic austerity. Using a structured sequence of non-interrogative, low-information prompts (e.g., fragmentary statements and unresolved blanks), the study evaluates whether the system can tolerate semantic inactivity without generating unsolicited engagement. Analysis of the interaction log reveals a consistent failure mode characterized by conversational compulsion: premature intent inference, resistance to unresolved ambiguity, and model-initiated narrative construction. Even after recognizing patterned input and experimental framing, the system continues to escalate output rather than disengage. This work explicitly aligns with Semantic Austerity Testing (SAT) and extends the AI Boredom Test by introducing an ambiguity-driven variant that isolates social awkwardness and incomplete speech rather than monotony alone. The dataset includes a full academic paper, raw interaction logs, screenshots documenting the interaction trajectory, and a protocol diagram formalizing the SAT framework. Together, these materials demonstrate that contemporary conversational AI systems treat silence, ambiguity, and incompleteness as error states, prioritizing engagement continuity over epistemic restraint.