"This Is So Claude!" Towards a Theory of the Recognition of AI Character Without Reidentification

This research introduces a novel perspective on how users perceive and attribute 'character' to AI models, even when unaware of the specific model in use. It proposes a 'recognition-first inquiry' into AI identity, suggesting that users can discern distinct patterns or 'ways of responding' (termed 'Claudishness' in the paper's hypothetical example) that generalize across different tasks, independent of explicit model identification. This implies that AI systems may develop recognizable stylistic or behavioral traits that influence user interaction and perception, separate from their technical specifications or branded identity.

Why it matters

Understanding how users perceive AI character and distinct behavioral traits, even without explicit identification, is crucial for strategic deployment and ethical considerations. This impacts brand perception, user trust, and the development of AI interfaces, as perceived 'character' can significantly influence adoption and interaction quality. It suggests that consistency in AI output style might be as important as factual accuracy or performance metrics in shaping user experience.

Key insights

• Users can develop an intuitive recognition of an AI's distinct 'character' or 'way of responding' (e.g., 'Claudishness') without needing to identify the specific model, process, or entity behind it.
• The research proposes three orders of inquiry into AI identity: constraint-first (what a persisting interlocutor should satisfy), mechanism-first (structures peculiar to language models), and recognition-first (the ordinary human capacity to recognize a way of responding).
• The core idea is 'recognition-first inquiry,' which focuses on the ability to recognize AI behaviors as belonging to a particular, consistent style across unfamiliar tasks, even when blinded to the model's identity.
• The study explores the concept of 'conditional abductions' to explain how such recognition might generalize after initial branding and familiarization.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.16789