---
id: "claim-weird-bias"
type: "claim"
source_timestamps: ["§ Agentic AI's Diversity Challenge"]
tags: ["ai-bias", "cultural-representation"]
related: ["concept-weird-bias-in-ai", "action-enrich-training-data"]
confidence: "high"
testable: true
speakers: ["Mark Purdy"]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-new-28-agent-teams-different-models"
sourceUrl: "https://hbr.org/2026/06/the-strongest-teams-of-ai-agents-will-be-built-using-different-models"
sourceTitle: "The Strongest Teams of AI Agents Will Be Built Using Different Models"
---
# Major LLMs inherently reflect WEIRD populations

**Claim:** According to a study by **Atari et al.**, the psychological profiles and responses generated by major large-language models like ChatGPT closely resemble those of people from **Western, Educated, Industrialized, Rich, and Democratic (WEIRD)** societies. Consequently these models inherently fail to capture the values and diversity of non-WEIRD populations, making them **structurally homogeneous from a global cultural perspective** (see [[concept-weird-bias-in-ai]]).

**Consequence:** Because prompting cannot remove a bias baked into the base model, the fix is to enrich training data with global cultural datasets (see [[action-enrich-training-data]]).

**Confidence: high.**

**Enrichment validation — STRONGLY GROUNDED:** This is the best-supported claim in the source. Atari et al. (2023), *"The Cultural Psychology of GPT,"* demonstrates that GPT-3.5/GPT-4 systematically align with WEIRD psychological profiles across multiple cultural-psychology benchmarks, clustering around Western patterns and often failing to represent non-WEIRD norms. The extraction accurately captures the core result.