---
id: "action-enrich-training-data"
type: "action-item"
source_timestamps: ["§ Seven Imperatives for Creating Diverse Agentic Teams"]
tags: ["data-engineering", "model-training"]
related: ["entity-big-five-framework", "entity-world-values-survey", "concept-weird-bias-in-ai", "framework-seven-imperatives"]
action: "Train models using psychometric frameworks and global cultural surveys."
outcome: "Produces agents that accurately reflect human personality nuances and diverse global values."
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"
---
# Enrich Agentic Training Data

**Imperative 2 of the [[framework-seven-imperatives]].**

**Action:** Move beyond standard internet scraping by actively training models on:
- Multi-dimensional **psychometric datasets** — the [[entity-big-five-framework|Big Five Framework]] (agreeableness, neuroticism, extraversion, openness, conscientiousness), and
- Global **cultural datasets** — the [[entity-world-values-survey|World Values Survey]].

**Why it works:** This helps agents develop nuanced, **non-binary** personality traits (the failure mode of persona prompting — see [[concept-cosmetic-ai-diversity]]) and mitigates the [[concept-weird-bias-in-ai|WEIRD bias]] inherent in major LLMs.

**Outcome:** Agents that more accurately reflect human personality nuances and diverse global values.

**Enrichment validation:** Conceptually sound (both datasets are canonical), and aligned with Atari-et-al.-style calls to add non-WEIRD data. Caveats: this is **not yet standard practice** in foundation-model training, and directly encoding psychometrics/values raises complex ethics and privacy questions.