Proceedings of the Annual Meeting of the Cognitive Science Society

As Large Language Models (LLMs) transition from passive text generators to autonomous social agents, decoding their underlying "cognitive DNA" is essential for ensuring alignment with diverse human values. This study investigates the intersection of socio-emotional maturity and cultural value orientations in frontier LLMs through the lens of cognitive science and cross-cultural psychology. We introduce the Dual-Axis Social Intelligence Scale (DASIS), a novel evaluative framework designed to decouple general social sophistication from implicit cultural biases—specifically along the Individualism-Collectivism (I-C) continuum. Across three interrelated experimental paradigms, we map the latent social architecture of nine state-of-the-art models. Study 1 establishes a behavioral baseline in Chinese, revealing that while models exhibit near-human social maturity, they suffer from a profound "Trans-Linguistic Value Locking" toward individualistic schemas, regardless of their developmental origin. Study 2 interrogates the "Whorfian Effect" (Linguistic Relativity) in synthetic cognition; findings indicate a Directional Asymmetry, where English context amplifies individualism while Chinese fails to exert a reciprocal collectivist pull, suggesting an "Axiomatic Individualism" ingrained during the alignment process. Study 3 explores "Cultural Perspective Taking," testing the models' Synthetic Theory of Mind (SToM) capabilities to voluntarily override default biases. Our results demonstrate that while default "personalities" are heavily skewed, models possess a latent Cognitive Decoupling capacity that allows for high-fidelity cultural simulation under explicit framing. This work provides a critical audit of the "WEIRD" (Western, Educated, Industrialized, Rich, and Democratic) bias in modern AI. We conclude that current alignment methodologies prioritize value consistency over cultural pluralism, potentially facilitating "Cognitive Imperialism" in global AI deployment. We propose DASIS as a standard for auditing the cross-cultural adaptability of autonomous agents.