Parham Pourdavood

Parham Pourdavood

Constitutional AI (CAI) aligns language models with explicitly stated normative principles, offering a transparent alternative to implicit alignment through human feedback alone. However, because constitutions are authored by specific groups of people, the resulting models may reflect particular cultural perspectives. We investigate this question by evaluating Anthropic's Claude Sonnet on 55 World Values Survey items, selected for high cross-cultural variance across six value domains and administered as both direct survey questions and naturalistic advice-seeking scenarios. Comparing Claude's responses to country-level data from 90 nations, we find that Claude's value profile most closely resembles those of Northern European and Anglophone countries, but on a majority of items extends beyond the range of all surveyed populations. When users provide cultural context, Claude adjusts its rhetorical framing but not its substantive value positions, with effect sizes indistinguishable from zero across all twelve tested countries. An ablation removing the system prompt increases refusals but does not alter the values expressed when responses are given, and replication on a smaller model (Claude Haiku) confirms the same cultural profile across model sizes. These findings suggest that when a constitution is authored within the same cultural tradition that dominates the training data, constitutional alignment may codify existing cultural biases rather than correct them--producing a value floor that surface-level interventions cannot meaningfully shift. We discuss the compounding nature of this risk and the need for globally representative constitution-authoring processes.

Parham Pourdavood, Michael Jacob, Terrence Deacon

This paper proposes a novel conceptualization of Large Language Models (LLMs) as externalized informational substrates that function analogously to DNA for human cultural dynamics. Rather than viewing LLMs as either autonomous intelligence or mere programmed mimicry, we argue they serve a broader role as repositories that preserve compressed patterns of human symbolic expression--"fossils" of meaningful dynamics that retain relational residues without their original living contexts. Crucially, these compressed patterns only become meaningful through human reinterpretation, creating a recursive feedback loop where they can be recombined and cycle back to ultimately catalyze human creative processes. Through analysis of four universal features--compression, decompression, externalization, and recursion--we demonstrate that just as DNA emerged as a compressed and externalized medium for preserving useful cellular dynamics without containing explicit reference to goal-directed physical processes, LLMs preserve useful regularities of human culture without containing understanding of embodied human experience. Therefore, we argue that LLMs' significance lies not in rivaling human intelligence, but in providing humanity a tool for self-reflection and playful hypothesis-generation in a low-stakes, simulated environment. This framework positions LLMs as tools for cultural evolvability, enabling humanity to generate novel hypotheses about itself while maintaining the human interpretation necessary to ground these hypotheses in ongoing human aesthetics and norms.