Beyond Instrumental and Substitutive Paradigms: Introducing Machine Culture as an Emergent Phenomenon in Large Language Models
Yueqing Hu, Xinyang Peng, Yukun Zhao, Lin Qiu, Ka-lai Hung, Kaiping Peng
TL;DR
The paper challenges two common frameworks for understanding cultural outputs of large language models—Instrumental (models reflect developers’ culture) and Substitutive (prompt-language triggers cultural frames)—and proposes Machine Culture as an emergent, probabilistic phenomenon arising from high-dimensional representations and safety-alignment dynamics. Using a fully crossed $2\times2$ design across eight multimodal tasks, it compares US- vs. China-developed models and English- vs. Chinese-prompt conditions, incorporating image generation and interpretation. Key findings include Cultural Reversal, where language priming produces nonintuitive or opposite effects relative to human patterns, and Service Persona Camouflage, where RLHF drives outputs toward a hyper-positive, low-variance “helpful assistant” persona, effectively masking underlying variance. The study argues that Machine Culture is not a reflection of human culture but a distinct computational ecology—shaped by superposition, mode collapse, and alignment pressures—calling for theoretical reframing and novel methodological approaches in cognitive science and AI ethics.
Abstract
Recent scholarship typically characterizes Large Language Models (LLMs) through either an \textit{Instrumental Paradigm} (viewing models as reflections of their developers' culture) or a \textit{Substitutive Paradigm} (viewing models as bilingual proxies that switch cultural frames based on language). This study challenges these anthropomorphic frameworks by proposing \textbf{Machine Culture} as an emergent, distinct phenomenon. We employed a 2 (Model Origin: US vs. China) $\times$ 2 (Prompt Language: English vs. Chinese) factorial design across eight multimodal tasks, uniquely incorporating image generation and interpretation to extend analysis beyond textual boundaries. Results revealed inconsistencies with both dominant paradigms: Model origin did not predict cultural alignment, with US models frequently exhibiting ``holistic'' traits typically associated with East Asian data. Similarly, prompt language did not trigger stable cultural frame-switching; instead, we observed \textbf{Cultural Reversal}, where English prompts paradoxically elicited higher contextual attention than Chinese prompts. Crucially, we identified a novel phenomenon termed \textbf{Service Persona Camouflage}: Reinforcement Learning from Human Feedback (RLHF) collapsed cultural variance in affective tasks into a hyper-positive, zero-variance ``helpful assistant'' persona. We conclude that LLMs do not simulate human culture but exhibit an emergent Machine Culture -- a probabilistic phenomenon shaped by \textit{superposition} in high-dimensional space and \textit{mode collapse} from safety alignment.